JP2018088101A - Synonymous expression extraction device, synonymous expression extraction method and synonymous expression extraction program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To accurately extract synonymous expressions of compound nouns from document data. A synonym expression extraction device includes a word pair setting unit, a semantic similarity learning unit, a word synonym determination unit, and a compound noun synonym determination unit. The word pair setting unit identifies a plurality of word pairs in the pair of compound nouns extracted from the document as synonymous word pairs and undetermined word pairs. The semantic similarity learning unit performs learning processing of the semantic similarity between words a plurality of times for each of the undetermined word pair and the plurality of synonymous word pairs in the document while adding cases. The word synonym determination unit, the relationship between the number of learning processing and meaning similarity in the undetermined word pair, based on the correlation coefficient for the relationship between the number of learning processing and synonym similarity in the synonymous word pair, It is determined whether the words of the undetermined word pair have the same meaning. The compound noun synonym determination unit determines that the pair of compound nouns is a synonymous expression when all the word pairs in the pair of compound nouns are synonymous word pairs. [Selection diagram] Figure 1

Description

  The present invention relates to a synonym expression extraction device, a synonym expression extraction method, and a synonym expression extraction program.

  As one technique for extracting synonyms from document data, the notation similarity between candidates is calculated based on the context co-occurrence and the notation editing distance for synonym candidates extracted from document data. There is known a method for determining whether or not they are synonymous based on the degree (see, for example, Patent Document 1). In this type of synonym extraction method, it is possible to extract synonyms of words and compound nouns used in a specific field.

International Publication No. 2014/002776

  In the above synonym extraction method, when determining whether or not a pair of compound nouns is synonymous, the determination is made based on the similarity between the notation of one compound noun and the notation of the other compound noun. For this reason, if there are few overlapping parts in the compound noun notations in the compound noun pairs that are actually synonymous, the compound noun pairs may be erroneously determined not to be synonymous expressions.

  In one aspect, an object of the present invention is to accurately extract synonymous expressions of compound nouns from document data.

  A synonym expression extracting device as one aspect includes a word pair setting unit, a semantic similarity learning unit, a word synonym determining unit, and a compound noun synonym determining unit. The word pair setting unit divides the compound noun pair extracted from the document data into a plurality of word pairs, refers to a synonym word dictionary in which synonymous word pairs are registered, and sets the plurality of word pairs as synonymous word pairs. , It is identified as an unconfirmed word pair whose synonym is uncertain. The semantic similarity learning unit learns the semantic similarity between words for each of a plurality of synonym word pairs in the document data including unconfirmed word pairs and synonym word pairs in compound noun pairs. Perform multiple times. The word synonym determination unit determines whether the words of the undefined word pair are synonymous based on the learning result of the semantic similarity of the undefined word pair and the learning result of the semantic similarity in each of the plurality of synonymous word pairs. Determine whether or not. The compound noun synonym determination unit determines that a pair of compound nouns is a synonym expression when a plurality of word pairs in the compound noun pair are all synonymous word pairs. The semantic similarity learning unit in the synonym expression extracting device adds a case to be used for the learning process for each of the semantic similarity learning processes for each of a plurality of word pairs to be processed. Further, the word synonym determination unit includes a relationship between the number of learning processes and the semantic similarity in the learning result of the semantic similarity of the undefined word pair, and the number of learning processes in the learning result of the semantic similarity of the synonym word pair. A correlation coefficient for the relationship with the semantic similarity is calculated. The word synonym determination unit determines whether or not the words of the unconfirmed word pair are synonymous based on the calculated correlation coefficient.

  According to the above-described aspect, it is possible to accurately extract synonymous expressions of compound nouns from document data.

It is a figure which shows the functional structure of the synonymous expression extraction apparatus which concerns on 1st Embodiment. It is a figure which shows the example of a synonym word dictionary. It is a figure explaining the determination method of whether it is synonymous with the example of a pair of compound nouns. It is a flowchart explaining the process which the synonymous expression extraction apparatus which concerns on 1st Embodiment performs. It is a flowchart explaining the content of a synonymous compound noun specific process. It is a flowchart explaining the content of a semantic similarity learning process. It is a flowchart (the 1) explaining the content of the synonym determination process about a word pair. It is a flowchart (the 2) explaining the content of the synonym determination process about a word pair. It is a flowchart explaining the content of the synonym determination process about the pair of compound noun. It is a figure which shows the example of a character string and the result of a morphological analysis. It is a figure which shows the example of the extraction result of a pair of compound nouns, and the list of word pairs. It is a figure which shows the example of a similarity transition table. It is a figure explaining the determination method whether an unconfirmed word pair is synonymous expression. It is a graph explaining the determination method of the frequency | count of a learning process. It is a figure which shows the functional structure of the synonymous expression extraction apparatus which concerns on 2nd Embodiment. It is a flowchart explaining the content of the synonymous compound noun specific process which concerns on 2nd Embodiment. It is a flowchart explaining the content of the setting process of a determination threshold value. It is a figure explaining the specific example of the setting method of a determination threshold value. It is a figure which shows the system configuration | structure of the synonym dictionary creation system which concerns on 3rd Embodiment. It is a figure which shows the system configuration | structure of the document rewriting system which concerns on 4th Embodiment. It is a figure which shows the functional structure of a document data rewriting apparatus. It is a flowchart explaining the process which a document data rewriting apparatus performs. It is a figure which shows the hardware constitutions of a computer.

[First Embodiment]
FIG. 1 is a diagram illustrating a functional configuration of the synonym expression extracting device according to the first embodiment.

  As shown in FIG. 1, the synonym expression extraction device 1 according to the present embodiment includes a character string extraction unit 110, a morpheme analysis unit 120, a compound noun extraction unit 130, and a synonym compound noun identification unit 140. The synonym expression extracting device 1 also stores a document set 191, an analysis dictionary 192, an analysis result corpus 193, a synonym word dictionary 194, a similarity transition table 195, and a synonym expression list 196 ( (Not shown).

  The character string extraction unit 110 extracts a character string from the document set 191. The document set 191 includes a plurality of document data used for extracting synonymous expressions of compound nouns. Each of the plurality of document data in the document set 191 includes, for example, terms used in a specific technical field, a specific industry, and the like. For example, the document set 191 is stored in the storage unit of the synonym expression extraction device 1 via an input receiving unit (not shown).

  The morpheme analysis unit 120 refers to the analysis dictionary 192 and performs morpheme analysis on the character string extracted by the character string extraction unit 110. The analysis dictionary 192 includes word information used for morphological analysis. The morpheme analysis unit 120 registers the result of the morpheme analysis in the analysis result corpus 193 and passes it to the compound noun extraction unit 130.

  The compound noun extraction unit 130 extracts compound nouns in the character string based on the result of morphological analysis. For example, the compound noun extraction unit 130 extracts a word string that satisfies the conditions of the compound noun based on the order of parts of speech in the character string. In addition, the compound noun extraction unit 130 extracts a pair of compound nouns that may be synonymous expressions from the extracted compound nouns.

  Based on the analysis result corpus 193 and the synonym word dictionary 194, the synonymous compound noun specifying unit 140 specifies a compound noun pair that is a synonym expression from the compound noun pairs extracted by the compound noun extracting unit 130. The synonym word dictionary 194 stores a plurality of synonymous word pairs. The synonym compound noun specifying unit 140 generates a word pair by dividing each compound noun into words for compound noun pairs that may be synonymous expressions, and refers to the synonym word dictionary 194 to generate the word pairs. Whether or not is synonymous. When there is a word pair in which it is uncertain whether or not it is synonymous among the word pairs in the compound noun pair, the synonym compound noun specifying unit 140 makes semantic similarity to all the word pairs specified from the result of the morphological analysis. Do the degree learning. In the following description, a word pair in which words are synonymous is referred to as a synonymous word pair, and a word pair in which whether or not the words are synonymous is referred to as an indeterminate word pair.

  When learning semantic similarity, the synonymous compound noun specifying unit 140 performs learning processing (calculation processing) of semantic similarity by a plurality of cases a plurality of times. At this time, the synonymous compound noun identification unit 140 adds a plurality of cases and performs the next learning process every time one learning process ends. Here, one case is a result of morphological analysis for one character string, for example. The synonymous compound noun identification unit 140 stores the learning result in the similarity estimation table 195 every time the semantic similarity learning process ends. Thereafter, the synonymous compound noun specifying unit 140 determines whether the number of cases (number of processing times) in the undefined word pair and the semantic similarity is based on the relationship between the number of cases in the synonymous word pair and the semantic similarity. It is determined whether the words of the confirmed word pair are synonymous. When the words of the undefined word pair are synonymous, the synonymous compound noun specifying unit 140 changes the undefined word pair to a synonymous word pair. After determining whether or not the uncertain word pair is a synonym word pair, the synonym compound noun specifying unit 140 determines that all the plural word pairs in the compound noun pair that may be synonymous expressions are synonymous word pairs. It is determined whether or not. When all the plural word pairs in the compound noun pair are synonymous word pairs, the synonymous compound noun specifying unit 140 determines that the compound nouns in the compound noun pair including the plurality of word pairs are synonymous expressions. The synonym compound noun specifying unit 140 registers the pair of compound nouns determined to be synonymous expressions in the synonym expression list 196.

  The synonymous compound noun identifying unit 140 having the above functions includes a word pair setting unit 141, a similarity transition table creating unit 142, a semantic similarity learning unit 143, a word synonym determining unit 144, and a compound noun synonym determining unit 145. And including.

  The word pair setting unit 141 generates a plurality of word pairs by dividing each compound noun in a compound noun pair having a possibility of synonymous expression into words, and the plurality of word pairs is synonymous with an uncertain word pair. And classify. The word pair setting unit 141 refers to the synonym word dictionary 194 and classifies a plurality of word pairs into unconfirmed word pairs and synonym word pairs. Further, the word pair setting unit 141 refers to the analysis result corpus 193 and the synonym word dictionary 194, and from the words that are different from the words included in the compound noun pairs that may be synonymously expressed in the document set 191. Extract synonym word pairs. The word pair setting unit 141 sets the unconfirmed word pair and the synonym word pair as word pairs used for determining whether or not the compound nouns in the compound noun pair are synonymous expressions.

  The similarity transition table creation unit 142 generates a similarity transition table 195 that stores the learning results of the semantic similarity for the unconfirmed word pairs and all the synonym word pairs. The similarity transition table calculation unit 142 causes the semantic similarity learning unit 143 to learn (calculate) the semantic similarity of the word pair, and stores the learning result of the semantic similarity of the word pair in the similarity transition table 195. When the total number of word pairs is N and the number of executions of the semantic similarity learning process is M, the similarity degree transition table creating unit 142 generates, for example, an N × M cell similarity degree transition table 195. .

  The semantic similarity learning unit 143 generates a semantic vector for the word of the case based on the result (case) of the morphological analysis stored in the analysis result corpus 193, and the semantic similarity of the word pair based on the generated semantic vector The process of calculating (learning) the degree is repeated M times. The semantic similarity learning unit 143 adds H cases used for calculating the semantic similarity every time the process of calculating the semantic similarity of the word pair is completed once. For example, the semantic similarity learning unit 143 sets the number of cases Ht in the m-th process (m = 1, 2,..., M) as Ht = m × H.

  The word synonym determination unit 144 is based on the semantic similarity score for the synonymous word pair and the semantic similarity score for the unconfirmed word pair in the semantic similarity learning result (similarity transition table 195). It is determined whether the words of the unconfirmed word pair are synonymous expressions. The word synonym determination unit 144 determines the undefined word pair when the correlation coefficient between the transition of the semantic similarity for the synonymous word pair and the transition of the semantic similarity for the undefined word pair is equal to or greater than a threshold. Are determined to be synonymous expressions.

  Based on the result of the word synonym determination unit 144, the compound noun synonym determination unit 145 determines whether or not the compound nouns of a pair of compound nouns that may be synonymous expressions are synonymous expressions. The compound noun synonym determination unit 145 determines that compound nouns of a pair of compound nouns are synonymous expressions when a plurality of word pairs in the compound noun pair are all synonymous pairs. The compound noun synonym determination unit 145 registers the compound noun pair determined to be synonymous expressions in the synonym expression list 196.

FIG. 2 is a diagram illustrating an example of a synonym word dictionary.
As shown in FIG. 2, in the synonym word dictionary 194, a plurality of pairs of words (synonym word pairs) that are synonymous expressions are registered. The types of words registered in the synonym word dictionary 194 and the number K of synonym word pairs can be set as appropriate. For example, the synonym word pair registered in the synonym word dictionary 194 may be a representative synonym pair having a high appearance frequency in the field of documents included in the document set 191, or the field of documents included in the document set 191. It may be a synonym word pair extracted at random regardless of.

  As described above, the synonym word dictionary 194 is used to determine whether or not compound nouns in a pair of compound nouns that may have synonymous expressions are synonymous.

FIG. 3 is a diagram for explaining a method for determining whether or not it is synonymous with an example of a pair of compound nouns.
In FIG. 3, as an example of a pair of compound nouns, a first compound noun (word string) 2 called “fare calculation module” and a second compound noun (word string) 3 called “transportation payment function” are shown. Show. The first compound noun 2 is a compound noun combining a first word 201 “Fare”, a second word 202 “Calculation”, and a third word 203 “Module”. On the other hand, the second compound noun 3 is a compound noun combining a first word 301 “transportation expense”, a second word 302 “adjustment”, and a third word 303 “function”. .

  For example, the first compound noun 2 and the second compound noun 3 described above are compound nouns used in the field of application software for calculating a fare (transportation cost). However, the appearance frequency of the first compound noun 2 in the document set 191 for the field is often lower than the appearance frequency of the words 201 to 203 in the first compound noun 2. Similarly, the appearance frequency of the second compound noun 3 in the document set 191 for the field is often lower than the appearance frequency of the words 301 to 303 in the second compound noun 3. Therefore, whether or not the two compound nouns are synonymous is accurately determined based on the appearance frequency, context similarity, and the like of the first compound noun 2 and the second compound noun 3 in the document set 191. Difficult to do.

  On the other hand, in this embodiment, as shown in FIG. 3, two compound nouns 2 and 3 are each divided into words to generate a plurality of word pairs WP1, WP2, and WP3, and for each word pair, It is determined with reference to the synonym word dictionary 194 whether it is synonymous. In the example shown in FIG. 3, first, a word of the first word 201 (“fare”) in the first compound noun 2 and the first word 301 (“transportation cost”) in the second compound noun 3. It is determined whether or not the pair WP1 is synonymous. In the synonym word dictionary 194 shown in FIG. 2, word pairs of “fare” and “transportation cost” are registered. For this reason, the synonym expression extraction device 1 determines that the word pair WP1 between the first word 201 in the first compound noun 2 and the first word 301 in the second compound noun 3 is synonymous. Similarly, the synonym expression extraction device 1 uses the word pair of the second word 202 (“calculation”) in the first compound noun 2 and the second word 302 (“checkout”) in the second compound noun 3. It is determined that WP2 is a synonym word pair.

  Here, further, the word pair WP3 of the third word 203 (“module”) in the first compound noun 2 and the third word 303 (“function”) in the second compound noun 3 is shown in FIG. Assume that it is registered in the synonym word dictionary 194. In this case, the synonym expression extraction device 1 determines that the word pair WP3 between the third word 203 in the first compound noun 2 and the third word 303 in the second compound noun 3 is synonymous. As described above, when it is determined that the plurality of word pairs WP1 to WP3 in the compound noun pair are all synonymous word pairs, the synonym expression extracting device 1 determines that the compound nouns in the compound noun pair are synonymous. .

  On the other hand, when the word pair WP3 of “module” and “function” is not registered in the synonym word dictionary 194 in FIG. 2, the synonym expression extraction device 1 determines the word pair WP3 in the compound noun pair shown in FIG. It is determined as an indeterminate word pair. In this case, the synonym expression extraction device 1 determines whether or not the word pair WP3 is a synonym word pair in order to determine whether or not the first compound noun 2 and the second compound noun 3 are synonymous expressions. Processing to determine is performed. The synonym expression extraction apparatus 1 is based on the transition of the learning result of the semantic similarity for each of the plurality of synonymous word pairs collected from the document set 191 and the transition of the learning result of the semantic similarity for the unconfirmed word pair. Then, it is determined whether or not the unconfirmed word pair is a synonymous word pair. When it is determined that the word pair WP3 of “module” and “function” is a synonym word pair by this process, the synonym expression extraction device 1 uses the first compound noun 2 (“fare calculation module”), the second And compound noun 3 (“transportation payment function”) is determined to be synonymous.

  Hereinafter, with reference to FIGS. 4 to 8, processing performed by the synonymous expression extraction apparatus 1 according to the present embodiment will be described.

  FIG. 4 is a flowchart for explaining processing performed by the synonymous expression extraction apparatus according to the first embodiment.

  As shown in FIG. 4, the synonym expression extracting apparatus 1 of the present embodiment first extracts a character string from the document set 191 (step S1). The character string extraction unit 110 performs the process in step S1. The character string extraction unit 110 extracts a character string from each of a plurality of document data included in the document set 191 according to a known extraction method.

  Next, the synonym expression extraction device 1 performs morphological analysis of the extracted character string (step S2). The process of step S2 is performed by the morphological analysis unit 120. The morpheme analysis unit 120 performs morpheme analysis on each of the extracted character strings with reference to the analysis dictionary 192 according to a known analysis method. The morpheme analysis unit 120 stores the result of the morpheme analysis in the analysis result corpus 193 and passes it to the compound noun extraction unit 130. The morpheme analysis unit 120 stores the result of the morpheme analysis in the analysis result corpus 193 for each sentence in the document set 191.

  Next, the synonym expression extracting device 1 extracts a compound noun from the character string based on the result of the morphological analysis (step S3). The compound noun extraction unit 130 performs the process in step S3. The compound noun extraction unit 130 extracts a word string satisfying the condition of the compound noun from the character string based on the result of the morphological analysis for each character string.

  Next, the synonym expression extraction device 1 determines whether or not a plurality of types of compound nouns have been extracted (step S4). For example, the compound noun extraction unit 130 performs the determination in step S4. When no compound noun is extracted, or when there is only one extracted compound noun (step S4; NO), there is no compound noun pair for determining whether or not the compound noun is synonymous. The (synonymous expression extraction device 1) ends the process.

  On the other hand, when a plurality of types of compound nouns are extracted (step S4; YES), the synonym expression extracting device 1 next identifies a pair of compound nouns that may be synonymous expressions (step S5). The process of step S5 is performed by the compound noun extraction unit 130, for example. In step S5, the compound noun extraction unit 130 identifies a pair of compound nouns that may have synonymous expressions according to a known identification method. For example, the compound noun extraction unit 130 calculates a context similarity for a compound noun for each of a plurality of combinations of compound noun pairs, and based on the context similarity, compound nouns that may have synonymous expressions are calculated. Identify pairs.

  Next, the synonym expression extraction device 1 determines whether or not there is a pair of compound nouns that may be synonymous expressions (step S6). For example, the compound noun extraction unit 130 performs the determination in step S6. If there is no compound noun pair that has the possibility of synonymous expression (step S6; NO), there is no compound noun pair for determining whether or not it is synonymous, so the compound noun extracting unit 130 (synonymous expression extracting device) 1) terminates a series of processes for one set of documents 191.

  On the other hand, when there is a pair of compound nouns that may be synonymous expressions (step S6; YES), the synonym expression extracting apparatus 1 performs a synonymous compound noun specifying process (step S7). The process of step S7 is performed by the synonymous compound noun specifying unit 140. The synonymous compound noun identifying unit 140 performs the above-described processing, and identifies a pair of compound nouns that are synonymous expressions. When a compound noun pair that is a synonym expression is specified, the synonym compound noun specifying unit 140 registers the specified compound noun pair in the synonym expression list 196. When the synonymous compound noun identification process (step S7) for the compound noun pair that has the possibility of synonymous expression is completed, the synchronous expression extracting device 1 ends the series of processes for the set of documents 191.

  As described above, the process (synonymous compound noun identifying process) of step S7 is performed by the synonymous compound noun identifying unit 140. The synonymous compound noun specifying unit 140 performs, for example, the process shown in FIG. 5 as the synonymous compound noun specifying process.

FIG. 5 is a flowchart for explaining the contents of the synonymous compound noun identification process.
In the synonym compound noun identification process, the synonym compound noun identification unit 140 first identifies a synonym word pair and an unconfirmed word pair in a compound noun pair that may be synonymously expressed (step S71). The processing in step S71 is performed by the word pair setting unit 141 of the synonymous compound noun specifying unit 140. As shown in FIG. 3, the word pair setting unit 141 divides a compound noun pair that has the possibility of synonymous expression into a plurality of word pairs and determines whether each word pair is a synonym word pair or an undefined word pair. Is identified. The word pair setting unit 141 refers to the synonym word dictionary 194 to identify whether the word pair is a synonym word pair or an undefined word pair. Moreover, in step S71, the word pair setting part 141 produces | generates the word pair list | wrist which shows whether a some word pair is a synonymous word pair, respectively, respectively, for example.

  Next, the synonymous compound noun specifying unit 140 collects synonymous word pairs (synonymous word pairs) registered in the synonym word dictionary 194 from the character strings of the document set 191 (step S72). The word pair setting unit 141 performs the process of step S72. The word pair setting unit 141 adds, to the word pair list, synonym word pairs that do not overlap with the synonym word pairs in the compound noun pair among the synonym word pairs collected in step S72.

  Next, the synonymous compound noun specifying unit 140 performs a semantic similarity learning process (step S73) with the unconfirmed word pair and the synonymous word pair identified and collected in steps S71 and S72 as processing target word pairs. The semantic similarity learning process in step S73 is performed by the similarity transition table creating unit 142 and the semantic similarity learning unit 143 of the synonymous compound noun specifying unit 140.

  In step S <b> 73, the similarity transition table creation unit 142 generates a similarity transition table 195 that stores a learning result of semantic similarity for each word pair to be processed. Further, the similarity transition table calculation unit 142 stores the learning result of the semantic similarity for each word pair to be processed in the semantic similarity learning unit 143 in a predetermined column (cell) of the similarity transition table 195. When the total number of word pairs to be processed is N and the semantic similarity is learned M times, the similarity transition table creating unit 142 has, for example, a similarity transition table having N × M columns (cells). 195 is generated. In step S73, the semantic similarity learning unit 143 learns a semantic vector for each word of the word pair based on the result (example) of the morphological analysis, and the meaning of the words of the word pair based on the semantic vector. Processing for calculating the similarity is performed. At this time, the semantic similarity learning unit 143 repeats the process of calculating the above semantic similarity for a set of word pairs M times. Furthermore, the semantic similarity learning unit 143 adds the number of cases used for calculating the semantic similarity every time the process of calculating the semantic similarity for one set of word pairs is completed once. For example, the semantic similarity learning unit 143 sets the number of cases Ht in the m-th (m = 1, 2,..., M) processing for one set of word pairs as Ht = m × H (for example, H = 1000). And

  When the process of step S73 is completed, the synonymous compound noun specifying unit 140 next performs a synonym determination process (step S74) for the unconfirmed word pair. The process of step S74 is performed by the word synonym determination unit 144 of the synonym compound noun specification unit 140. The word synonym determination unit 144 refers to the similarity transition table 195 to calculate a correlation coefficient between the transition of the semantic similarity of the synonym word pair and the transition of the semantic similarity of the undefined word pair, and the correlation coefficient Is equal to or greater than the threshold, it is determined that the words of the undefined word pair are synonymous. The word synonym determination unit 144 changes an unconfirmed word pair that has been determined to be synonymous to a synonym word pair.

  Next, the synonymous compound noun specifying unit 140 performs synonym determination processing for a pair of compound nouns (step S75). The process of step S75 is performed by the compound noun synonym determination unit 145 of the synonym compound noun identification unit 140. The compound noun synonym determination unit 145 determines whether or not each of a plurality of word pairs in a compound noun pair having the possibility of synonymous expression is a synonym word pair. When all the word pairs in the compound noun pair are synonymous word pairs, the compound noun synonym determination unit 145 determines that the compound noun pair is a compound noun pair of synonymous expressions, and the compound noun pair is synonymously expressed. Register in the list 196.

  When the process of step S75 is completed, the synonymous compound noun specifying unit 140 ends the synonymous compound noun specifying process (step S7).

  The semantic similarity learning process (step S73) of the synonymous compound noun specifying process performed by the synonymous compound noun specifying unit 140 is performed by the similarity transition table creating unit 142 and the semantic similarity learning unit 143 as described above. . The similarity transition table creation unit 142 and the semantic similarity learning unit 143 perform, for example, the process illustrated in FIG. 6 as the semantic similarity learning process.

FIG. 6 is a flowchart for explaining the content of the semantic similarity learning process.
In the semantic similarity learning process, as shown in FIG. 6, first, a similarity transition table 195 is prepared that stores M learning results for each of N word pairs (step S7301). The process of step S7301 is performed by the similarity transition table creation unit 142. For example, the similarity transition table creation unit 142 generates a table having N × M data storage columns (cells).

  After the process of step S7301, the similarity transition table creating unit 142 and the semantic similarity learning unit 143 repeat the first loop process (steps S7302 to S7309) M times. One process in the first loop process is a second loop process (steps S7303 to S7308). At the loop end (steps S7302 and S7309) of the first loop process, a process of counting the number m of processes of the second loop process and a determination of whether or not the number m of processes is m ≧ M are performed. The similarity transition table creation unit 142 performs counting and determination of the number m of processes. When the second loop process is repeated M times, the similarity transition table creation unit 142 ends the first loop process and ends the semantic similarity learning process.

  In the second loop process (steps S7304 to S7307), the semantic similarity is calculated by learning the semantic vector of each word in a set of word pairs, and the calculated semantic similarity is stored in the similarity transition table. The process is stored in 195. At the loop end (steps S7303 and S7308) of the second loop process, whether or not the process of selecting one word pair from the plurality of word pairs and the processes of steps S7304 to S7307 have been performed on all the word pairs. Is determined. The selection and determination at the loop end of the second loop processing is performed by the similarity transition table creation unit 142. When the processing of steps S7304 to S7307 is performed for all word pairs, the similarity transition table creation unit 142 ends one second loop processing.

  In one process in the second loop process, first, H morphological analysis results are acquired from the analysis result corpus 193 as learning examples (step S7304). The semantic similarity learning unit 143 performs the processing in step S7304. The semantic similarity learning unit 143 acquires H morphological analysis results that have not yet been acquired in the current semantic similarity learning processing from the morphological analysis results registered in the analysis result corpus 193. When the second loop process is the second to Mth times, the semantic similarity learning unit 143 adds the acquired H learning examples to the acquired learning examples.

  The semantic similarity learning unit 143 that has acquired the learning example continuously learns the semantic vector of the word pair based on the acquired learning example (step S7305), and calculates the semantic similarity of the word pair based on the learned semantic vector. Calculate (step S7306). In step S7305, the semantic similarity learning unit 143 learns a semantic vector for each word in a set of word pairs currently being processed according to a known learning method. In step S7305, the semantic similarity learning unit 143 calculates the semantic similarity of the word pair based on the learned semantic vector of each word. The semantic similarity learning unit 143 sends the semantic similarity calculated for the word pair currently being processed to the similarity transition table creating unit 142. Upon receiving the semantic similarity calculation result, the similarity transition table creating unit 142 stores the received semantic similarity value in the similarity transition table 195 (step S7307). In step S7307, the similarity transition table creation unit 142 identifies the word pair currently being processed and the processing count m in the N × M data storage columns (cells) of the similarity transition table 195. The calculation result of the semantic similarity is stored in the cell.

  When the processing of steps S7304 to S7307 is performed on all word pairs, the similarity transition table creation unit 142 ends the second loop processing. Thereafter, the similarity transition table 142 updates the processing count m to m = m + 1 at the loop end (step S7302 or S7309) of the first loop processing, and repeats the second loop processing until m ≧ M. When the number m of processing reaches m ≧ M, the similarity transition table creation unit 142 ends the first loop process and ends the semantic similarity learning process.

  When the semantic similarity learning process (step S73) by the similarity transition table creating unit 142 and the semantic similarity learning unit 143 is completed, the synonymous compound noun specifying unit 140 then performs a synonym determination process for the word pair (step S74). I do. The process of step S74 is performed by the word synonym determination unit 144 of the synonym compound noun specification unit 140. The word synonym determination unit 144 performs the processes shown in FIGS. 7A and 7B as the synonym determination process in step S74.

  FIG. 7A is a flowchart (part 1) illustrating the contents of synonym determination processing for a word pair. FIG. 7B is a flowchart (part 2) for explaining the contents of synonym determination processing for word pairs.

  In the synonym determination process for word pairs, the word synonym determination unit 144 first obtains a list of word pairs collected from the document set 191 and a similarity transition table 195 as shown in FIG. 7A (step S7401). .

  Next, the word synonym determination unit 144 performs first loop processing (S7402 to S7411). One process (steps S7403 to S7410) in the first loop process includes a second loop process (steps S7403 to S7406) and a process of steps S7407 to S7410 performed after the second loop process. At the loop end (steps S7402 and S7411) of the first loop process, the word synonym determination unit 144 performs a process of selecting unconfirmed word pairs and a process of steps S7403 to S7410 for all unconfirmed word pairs. It is determined whether or not. When the processing of steps S7403 to S7410 is performed for all unconfirmed word pairs, the word synonym determination unit 144 ends the first loop processing and ends the synonym determination processing for the word pairs.

  One processing (steps S7404 and S7405) in the second loop processing is performed by calculating the correlation coefficient of the transition of the semantic similarity between the unconfirmed word pair currently selected as a processing target and one set of synonymous word pairs. It is a process to calculate. First, the word synonym determination unit 144 obtains the semantic similarity of the unconfirmed word pair that is the current processing target and the semantic similarity of the synonym word pair from the similarity transition table 195 (step S7404). Subsequently, the word synonym determination unit 144 calculates a correlation coefficient between the acquired semantic similarities (step S7405). In step S7405, the word synonym determination unit 144 calculates a correlation coefficient by the following equation (1), for example.

In Expression (1), x _m is the result of the m-th semantic similarity learning process (semantic similarity) for the undefined word pair, and xa is the addition of the results of M learning processes for the undefined word pair. Average. Further, in the equation (1), y _m is the result of a learning process of the m-th mean similarity to synonymous word pairs (meaning similarity), ya is additive results M times of learning process for the synonymous word pairs Average.

  At the loop end of the second loop processing (steps S7403 and S7406), processing for selecting a synonym word pair to be processed from all synonymous word pairs and processing for calculating a correlation coefficient for all synonymous word pairs. It is determined whether or not it has been performed. If the process which calculates a correlation coefficient with all the synonym word pairs is performed, the word synonym determination part 144 will complete | finish a 2nd loop process.

  When the second loop process ends, the word synonym determination unit 144 next calculates the average value of the correlation coefficients of the semantic similarity for the unconfirmed word pair that is the current processing target (step S7407). In step S7407, the word synonym determination unit 144 calculates an average value of correlation coefficients of semantic similarity with each synonym word pair calculated in the second loop process.

  Next, as shown in FIG. 7B, the word synonym determination unit 144 determines whether or not the calculated average value of the correlation coefficients is equal to or greater than a threshold value (step S7408). When the average value of the correlation coefficients is equal to or greater than the threshold (step S7408; YES), the word synonym determination unit 144 changes the unconfirmed word pair that is the current processing target of the word pair list to a synonym word pair (step S7409). . On the other hand, when the average value of the correlation coefficients is smaller than the threshold (step S7408; NO), the word synonym determination unit 144 deletes the unconfirmed word pair that is the current processing target from the word pair list (step S7410).

  When the second loop processing for the unconfirmed word pair that is the current processing target and the processing in steps S7407 to S7410 are finished, the word synonym determination unit 144 determines whether there is an unconfirmed word pair that has not been subjected to these processing. Is determined (step S7402 or S7411). When there is an unconfirmed word pair that has not been subjected to the second loop process and the processes of steps S7407 to S7410, the word synonym determination unit 144 performs the second loop process for the unprocessed unconfirmed word pair and the steps S7407 to S7407. The process of S7410 is performed. Then, when the second loop process and the processes of steps S7407 to S7410 are performed on all unconfirmed word pairs, the word synonym determination unit 144 ends the first loop process and determines the synonyms for the word pairs. The process ends.

  When the word synonym determination process (step S74) by the word synonym determination unit 144 is completed, the synonym compound noun determination unit 140 then performs a synonym determination process (step S75) for a pair of compound nouns. The process of step S75 is performed by the compound noun synonym determination unit 145 of the synonym compound noun determination unit 140. The compound noun synonym determination unit 145 performs the process shown in FIG. 8 as the synonym determination process in step S75.

  FIG. 8 is a flowchart for explaining the contents of synonym determination processing for a pair of compound nouns.

  In the synonym determination process for a pair of compound nouns, the compound noun synonym determination unit 145 performs a loop process (S7501 to S7505). One processing in the loop processing (steps S7502 to S7504) is performed when a plurality of word pairs in a pair of compound nouns that may be synonymous expressions are all synonymous word pairs. This is a process of registering in the synonym expression list 196. The compound noun synonym determination unit 145 performs the processing of steps S7502 to S7504 for each of all compound noun pairs that may have synonymous expressions. At the loop end (steps S7501 and S7505) of the loop processing, the compound noun synonym determination unit 145 performs processing for selecting a compound noun pair and steps S7502 for all compound noun pairs that may have synonymous expressions. It is determined whether or not the processing of S7504 has been performed.

  In one process in the loop process, the compound noun synonym determination unit 145 first refers to the result of the synonym determination process for each of a plurality of word pairs in the compound noun pair currently selected for processing (step). S7502).

  Next, the compound noun synonym determination unit 145 determines whether or not all the word pairs in the compound noun pair are synonymous word pairs (step S7503). If all are synonymous word pairs (step S7503; YES), the compound noun synonym determination unit 145 registers the compound noun pair currently being processed in the synonym expression list 196 (step S7504), and ends the loop processing? It is determined whether or not (step S7505). On the other hand, when a word pair that is not synonymous is included in the plurality of word pairs (step S7503; NO), the compound noun synonym determination unit 145 determines whether to skip step S7504 and end the loop processing. (Step S7505). When the processes of steps S7502 to S7504 are performed on all compound noun pairs, the compound noun synonym determination unit 145 ends the loop process and ends the synonym determination process for the compound noun.

  As described above, in the synonymous expression determination process for the compound noun according to the present embodiment, each compound noun in the compound noun pair is divided into words to generate a plurality of word pairs, and all of the plurality of word pairs are all synonymous. A pair of compound nouns that are word pairs is determined as a synonymous expression. In addition, for compound noun pairs that have the possibility of synonymous expressions, based on the results of semantic similarity learning performed for each of the word pairs whose synonyms are uncertain and those that are synonymous. Thus, it is determined whether or not a word pair for which whether or not it is synonymous is synonymous. At this time, the learning of the semantic similarity of each word pair is performed in such a manner that learning examples are added for each time in a plurality of times, and the relationship between the number of cases and the semantic similarity is calculated. When the semantic similarity of word pairs is learned by such a method, the number of cases increases and the value of semantic similarity increases for synonymous word pairs. Moreover, the same tendency is seen in the relationship between the number of cases and the semantic similarity in a plurality of synonymous word pairs. For this reason, if the relationship between the number of cases and semantic similarity in a word pair for which it is uncertain whether or not they are synonymous has the same tendency as the relationship between the number of cases and semantic similarity in a synonymous word pair, It becomes possible to determine that the words of the word pair are synonymous. Therefore, according to this embodiment, even when there is a small number of document data in the document set 191 or when there is a document including a compound noun used only in a specific field, a synonym expression of the compound noun can be extracted with high accuracy. It becomes possible.

  FIG. 9 is a diagram illustrating an example of a character string and a result of morphological analysis. FIG. 10 is a diagram illustrating an example of a compound noun pair extraction result and a list of word pairs.

  An example of character strings extracted from the document set 191 is shown in the table 401 in FIG. In the synonym expression extraction apparatus 1 according to the present embodiment, the morpheme analysis unit 120 performs morpheme analysis (step S2) on each of the character strings extracted from the document set 191. The table 402 in FIG. 9B includes the result of the morphological analysis for the character string “execute the fare calculation module to obtain the transportation cost” shown in the table 401 in FIG. The result of the morphological analysis for the character string “I will carry out the transportation expense settlement function to find the The symbol “/” in the table 402 indicates a morpheme break. In the analysis result corpus 193, for example, the result of morphological analysis for each sentence (each character string) is stored in a format such as the table 402.

  After performing the morphological analysis, the synonym expression extraction device 1 extracts a compound noun from the character string in the compound noun extraction unit 130 (step S3), and further identifies a pair of compound nouns that may be synonymous expressions ( Step S5). When the condition of a word string in which a plurality of nouns are continuous is set as one of the conditions of the compound noun, the character string “Execute the fare calculation module to obtain transportation costs” Is extracted. Also, a compound noun “transportation cost settlement function” is extracted from the character string “perform transportation cost settlement function to obtain transportation cost”. Although not shown, compound nouns are extracted from other character strings according to similar conditions.

  After extracting the compound noun, the compound noun extraction unit 130 identifies a pair of compound nouns that may be synonymous expressions based on the context similarity of the compound noun. Both of the above-described two character strings are sentences such as “to obtain transportation expenses (compound nouns)...”, And the context similarity is high. Therefore, as shown in the table 403 of FIG. 10A, the compound noun output unit 130 converts the “fare calculation module” and the “transportation cost adjustment function” into a pair of compound nouns that may be synonymously expressed. Identify.

  Thereafter, in the synonym expression extraction device 1, the word pair setting unit 141 identifies a plurality of word pairs in the compound noun pair as synonymous word pairs or determination target word pairs, and collects synonym word pairs from the document set 191 ( Steps S71 and S72).

  As described above, the “fare calculation module” and the “transportation cost adjustment function” are compound nouns each combining three words (nouns). For this reason, from the pair of the “fare calculation module” and the “transportation cost adjustment function”, as shown in the word pair list 404 of FIG. 10B, the word pair of “fare” and “transportation cost”, A word pair of “calculation” and “settlement” and a word pair of “module” and “function” are generated. The word pair setting unit 141 refers to the synonym word dictionary 194 and identifies which of these three word pairs is a synonym word pair or a determination target word pair. When referring to the synonym word dictionary 194 in FIG. 2, the word pair of “fare” and “transportation cost” and the word pair of “calculation” and “settlement” are synonymous word pairs. In contrast, the word pair of “module” and “function” is an undefined word pair. For this reason, the synonym word dictionary 194 registers three word pairs and attributes (unconfirmed or synonymous) of each word pair in the word pair list 404. Further, when registering a word pair in the word pair list 404, an array ID for identifying each word pair is assigned as shown in FIG.

  Further, the determination target identification unit 141 collects synonym word pairs registered in the synonym word dictionary 194 from the results of morphological analysis for all character strings, and registers them in the word pair list 404. In the synonym word dictionary 194 of FIG. 2, a set of “execution” and “execution” is registered as a synonym word pair. Further, the result of the morphological analysis shown in the table 402 of FIG. 9B includes a character string including the word “execution” and a character string including the word “execution”. Therefore, the determination target identification unit 141 registers the word pairs of “execution” and “execution” as synonymous word pairs as in the word pair list 404 of FIG. 10B. Further, the determination target identification unit 141 collects all the synonym word pairs registered in the synonym word dictionary 194 from the morphological analysis results for all the character strings, and registers them in the word pair list 404.

  After registering the uncertain word pair in the compound noun pair and the synonym word pair registered in the synonym word dictionary 194 in the word pair list 404, the synonym expression extracting device 1 performs the semantic similarity learning process (step S73). I do. In the semantic similarity learning process, the similarity transition table creating unit 142 and the semantic similarity learning unit 143 perform the processing of steps S7301 to S7309 in FIG. 6 and the similarity transition table 195 as shown in FIG. create.

FIG. 11 is a diagram illustrating an example of the similarity transition table.
As shown in FIG. 11, the similarity transition table 195 has a data storage column (cell) for storing the results of M learning processes for each of N word pairs. The result of the M learning processes for each word pair is stored in a data storage column designated by the array ID assigned to the word pair and the learning count m. Further, in the similarity transition table 195 in FIG. 11, the number of cases H acquired in one semantic similarity learning process is set to H = 1000. That is, the result of the m-th learning process in the similarity transition table 195 in FIG. 11 is the result of the learning process performed using m × 1000 learning cases (m = 1, 2,..., M). Become.

  After creating the similarity transition table 195, the synonym expression extraction device 1 performs synonym determination processing (step S74) for the word pair in the word synonym determination unit 144. The process of step S74 is a process of determining whether or not the unconfirmed word pair is a synonymous word pair. The word synonym determination unit 144 performs the processes of steps S7401 to S7411 in FIGS. 7A and 7B to determine whether or not the unconfirmed word pair is a synonym word pair.

FIG. 12 is a diagram illustrating a method for determining whether or not an undefined word pair is a synonymous expression.
FIG. 12A shows three straight lines indicating the relationship between the number of cases for the synonymous word pair and the semantic similarity in the similarity transition table 195, and the number of cases and the semantic similarity for the undefined word pair. And two straight lines showing the relationship. A synonym word pair is a pair of words defined as synonymous in the synonym word dictionary 194. Each word in the synonym word pair has a higher appearance frequency in the document set 191 than a compound noun combining a plurality of words. For this reason, the semantic similarity of the synonym word pair calculated based on the learning result of the semantic vector of each word in the synonym word pair is represented by the number of cases (learning data amount) as shown in FIG. There is a tendency to increase proportionately. That is, when two words in a word pair are synonymous, it can be said that the semantic similarity of the word pair tends to increase in proportion to the number of cases (learning data amount).

  Therefore, when two words in an unconfirmed word pair are synonymous, the relationship between the number of cases and semantic similarity in the word pair has the same tendency as the relationship between the number of cases and semantic similarity in the synonymous word pair It is thought that there is. That is, when two words in an undefined word pair are synonymous, the semantic similarity of the word pair is proportional to the number of cases (learning data amount) as shown by a thick solid line in FIG. It will be higher. On the other hand, when the two words in the uncertain word pair are not synonymous, the relationship between the number of cases and the semantic similarity for the word pair is synonymous as shown by the thick dotted line in FIG. It can be said that the relationship between the number of cases and the degree of semantic similarity in word pairs tends to be different.

  Therefore, in the present embodiment, processing as in steps S7403 to S7407 in FIG. 7A is performed to calculate a correlation coefficient between the transition of the semantic similarity of the undefined word pair and the transition of the semantic similarity of the synonymous word pair. . For example, in the similarity transition table 195 in FIG. 11, the word pair with the array ID “1” is an unconfirmed word pair, and the word pairs with the other array ID are synonymous word pairs (see (b) in FIG. 10). reference). For this reason, the word synonym determination unit 144 calculates the correlation coefficient of the semantic similarity for each of the array ID pairs in which one array ID is “1” as in the table 405 in FIG. To do. Thereafter, the word synonym determination unit 144 calculates the average value of the calculated correlation coefficients of the plurality of semantic similarity degrees, and if the average value is equal to or greater than the threshold, the word whose array ID is “1” It is determined that the pair is a synonym word pair. In the example shown in FIG. 12B, the average value of the correlation coefficient is 0.68. Therefore, when the determination threshold used in step S7408 in FIG. 7B is 0.68 or less, it is determined that the word pair (pair of “module” and “function”) having the array ID “1” is synonymous. The In this case, for example, the word synonym determination unit 144 changes the attribute of the word pair of “module” and “function” in the word pair list 404 of FIG. 10B from unconfirmed to synonymous (step S7409). . On the other hand, when the determination threshold value used in step S7408 of FIG. 7B is larger than 0.68, it is determined that the word pair with the array ID “1” is not synonymous. In this case, the word synonym determination unit 144 deletes the word pair of “module” and “function” from the word pair list 404 (step S7410).

  After the process of step S74, in the synonym expression extraction device 1, the compound noun synonym determination unit 145 determines whether or not a pair of compound nouns that may be synonymous expressions are synonymous expressions. As described above, a compound noun pair having the possibility of synonymous expression includes an undefined word pair among a plurality of word pairs in the compound noun pair. For the unconfirmed word pair, whether or not it is a synonym word pair has been determined by the synonym determination process for the word in step S74. When the unconfirmed word pair is a synonymous word pair, the attribute of the unconfirmed word pair in the word pair list 404 is changed from unconfirmed to synonymous. On the other hand, when the unconfirmed word pair is not a synonymous word pair, the unconfirmed word pair is deleted from the word pair list 404. The synonym word pairs in the compound noun pairs are registered in the word pair list 404. Therefore, when all of a plurality of word pairs in a compound noun pair are registered as synonymous word pairs in the word pair list 404, the compound noun synonym determination unit 145 determines that the compound noun pair is a synonym expression, Register in the synonym expression list 196.

  As described above, in the synonym expression extraction device 1 according to the present embodiment, the pair of compound nouns extracted from the document set 191 is used for learning with the word pair whose synonym is uncertain and the synonym word pair. Learning semantic similarity multiple times while adding cases (data). If the average value of the correlation coefficients for the relationship between the number of learnings in the synonymous word pair and the semantic similarity and the relationship between the number of learnings in the unconfirmed word pair and the semantic similarity is equal to or greater than a threshold value, The confirmed word pair is determined as a synonymous word pair. In the present embodiment, whether or not a pair of compound nouns is a synonymous expression is determined based on whether or not a plurality of word pairs in the compound noun pair are all synonymous word pairs. Therefore, according to the present embodiment, it is possible to improve the accuracy of determination as to whether or not a synonym expression is made for a pair of compound nouns with a low appearance frequency in the document set 191.

  Note that the number M of learning processes of the semantic similarity of each word pair in the present embodiment can be set as appropriate, but is preferably 20 times or more (M ≧ 20).

FIG. 13 is a graph illustrating a method for determining the number of learning processes.
In the graph of FIG. 13, the horizontal axis represents the number of executions of the semantic similarity learning process. In the graph of FIG. 13, the left vertical axis is the significance test value for the correlation coefficient of semantic similarity, and the right vertical axis is the processing time required for the learning process.

  The relationship between the number of executions of the learning process for each word pair and the significance test value is as shown in FIG. 13 by a thick solid line, where the significance test value for the correlation coefficient of the semantic similarity increases as the number of executions increases. Decrease. The significance test value is a result of testing whether or not the correlation coefficient is statistically significant, and is calculated by a known test method. This significance test value is related to the reliability of the correlation coefficient of semantic similarity, and the smaller the significance test value, the higher the reliability of the correlation coefficient of semantic similarity. That is, as shown in FIG. 13, when the number of executions of learning processing for each word pair is increased, the reliability of the correlation coefficient of the semantic similarity of two word pairs is increased. However, if the number of executions of the learning process for each word pair is increased, the processing time required for the learning process increases as shown by a thick dotted line in FIG. Therefore, the number of executions M of the learning process for each word pair is the reliability of the correlation coefficient of the desired semantic similarity (determination accuracy of whether the word pair is a synonym expression) or the process of the synonym expression extraction device 1. What is necessary is just to set suitably based on capability, a desired response, etc. For example, when a reliability value that is 0.5 or less in the significance test value (99% or more reliability) is desired, the number of executions M of the learning process for each word pair may be set to about 30 times. Also, for example, when a reliability level of 0.6 or less is desired for the significance test value (99% or higher reliability), the number of executions M of learning processing for each word pair can be reduced to about 20 times. It becomes.

  Note that the flowcharts of FIGS. 4 to 8 are merely examples of synonym expression extraction processing according to the present embodiment. The synonymous expression extraction processing according to the present embodiment is not limited to the flowcharts of FIGS. 4 to 8, and the processing content can be appropriately changed without departing from the gist of the present embodiment. For example, in the semantic similarity learning process (step S73), instead of generating the similarity transition table 195 as shown in FIG. 11, N pieces of learning results are sequentially stored in each of N word pairs. May be generated. Further, for example, in the synonym determination process (step S74) for the word pair, the semantic similarity of the undefined word pair that is the current processing target may be acquired before the second loop process in FIG. 7A is started. In this case, in step S7404 in FIG. 7A, only the semantic similarity of the synonym word pair that is the current processing target is acquired. For this reason, in the second loop processing, it is possible to suppress an increase in processing time due to acquiring the semantic similarity of the undefined word pair each time.

  Further, the synonym expression extraction device 1 according to the present embodiment is not limited to the configuration illustrated in FIG. 1, and may be a device including an acquisition unit that acquires the document set 191 from an external device, for example. Further, the synonym expression extraction device 1 may be a device including a display unit that displays the processing result of the compound noun synonym identification process (step S7), the synonym expression list 196, or the like, or an output unit that outputs to an external device. Good.

[Second Embodiment]
FIG. 14 is a diagram illustrating a functional configuration of the synonym expression extracting device according to the second embodiment.

  As shown in FIG. 14, the synonym expression extraction device 1 according to the present embodiment includes a character string extraction unit 110, a morpheme analysis unit 120, a compound noun extraction unit 130, and a synonym compound noun identification unit 140. The synonym expression extracting device 1 also stores a document set 191, an analysis dictionary 192, an analysis result corpus 193, a synonym word dictionary 194, a similarity transition table 195, and a synonym expression list 196 ( (Not shown).

  The character string extraction unit 110, the morpheme analysis unit 120, the compound noun extraction unit 130, and the synonym compound noun identification unit 140 in the synonym expression extraction device 1 of the present embodiment each have the functions described in the first embodiment. Note that the synonym compound noun specifying unit 140 in the synonym expression extracting device 1 of the present embodiment uses a determination threshold used for determining whether or not an undefined word pair is a synonymous pair in addition to the functions described in the first embodiment. With the ability to set

  The synonym compound noun identifying unit 140 in the synonym expression extracting device 1 of the present embodiment includes a word pair setting unit 141, a similarity transition table creation unit 142, a semantic similarity learning unit 143, a word synonym determination unit 144, and a composite A noun synonym determination unit 145. The word pair setting unit 141, the similarity transition table creating unit 142, the semantic similarity learning unit 143, the word synonym determining unit 144, and the compound noun synonym determining unit 145 in the synonymous compound noun specifying unit 140 according to the present embodiment, respectively, The function described in the first embodiment is provided.

  In addition, the synonymous compound noun specifying unit 140 according to the present embodiment further includes a determination threshold setting unit 146. The determination threshold setting unit 146 sets a determination threshold used for determining whether or not the unconfirmed word pair is a synonymous pair. The determination threshold setting unit 146 sets a determination threshold based on the learning result of the semantic similarity in each of a plurality of synonymous word pairs collected from the document set by the word pair setting unit 141. That is, the determination threshold setting unit 146 sets a determination threshold based on the learning result of the synonym word pair in the similarity transition table 195 created in the semantic similarity learning process (step S73) described in the first embodiment. .

  The synonym expression extraction apparatus 1 according to the present embodiment extracts a synonym expression of compound nouns from the document set 191 by, for example, the processes of steps S1 to S7 illustrated in FIG. Of the processes in steps S1 to S7 performed by the synonymous expression extraction apparatus 1 according to the present embodiment, the processes in steps S1 to S6 may be the same as described in the first embodiment. On the other hand, the synonym expression extraction device 1 according to the present embodiment is used to determine whether or not the undefined word pair is a synonym pair in the semantic similarity learning process included in the synonym compound noun identification process in step S7. Set the judgment threshold. For this reason, the synonym expression extraction apparatus 1 which concerns on this embodiment performs the process shown, for example in FIG. 15 as a synonymous compound noun specific process of step S7.

  FIG. 15 is a flowchart for explaining the contents of the synonymous compound noun specifying process according to the second embodiment.

  Also in the present embodiment, the synonymous compound noun identifying process is performed by the synonymous compound noun identifying unit 140. As shown in FIG. 15, first, the synonym compound noun specifying unit 140 is registered in the synonym word dictionary from the document set (step S71) for identifying the synonym word pair and the unconfirmed word pair in the compound noun pair. The process (step S72) which collects the same synonym word pair is performed. The word pair setting unit 141 performs the processes in steps S71 and S72. The word pair setting unit 141 performs the processing described in the first embodiment, and generates, for example, a word pair list 404 as illustrated in FIG.

  Next, the synonymous compound noun specifying unit 140 performs a semantic similarity learning process (step S73) with the unconfirmed word pair and the synonym word pair obtained by the processes of steps S71 and S72 as the word pairs to be processed. The semantic similarity learning process in step S73 is performed by the similarity transition table creating unit 142 and the semantic similarity learning unit 143 of the synonymous compound noun specifying unit 140. For example, the similarity transition table creation unit 142 and the semantic similarity learning unit 143 perform the processing described in the first embodiment (see FIG. 6), and learn the semantic similarity in each of a plurality of word pairs. A similarity transition table 195 (see FIG. 11) showing the transition of That is, also in this embodiment, the semantic similarity learning unit 143 executes the process of learning the semantic similarity for one set of word pairs a plurality of times, and each time it is executed, a case (morpheme analysis) used for learning Result). For example, the semantic similarity learning unit 143 sets the number of cases Ht in the m-th (m = 1, 2,..., M) processing for one set of word pairs as Ht = m × H (for example, H = 1000). And

  When the semantic similarity learning process in step S73 is completed, the synonymous compound noun specifying unit 140 next performs a determination threshold setting process (step S76) used for synonym determination for the word pair. The determination threshold value setting unit 146 performs the determination threshold value setting process in step S76. The determination threshold value setting unit 146 sets a determination threshold value based on the semantic similarity learning results for the plurality of synonymous word pairs in the similarity transition table 195 generated in step S73. Specifically, the determination threshold setting unit 146 calculates a correlation coefficient for the learning result of the semantic similarity between synonymous word pairs in the similarity transition table 195, and sets the average value of the correlation coefficient as the determination threshold. To do.

  When the process of step S76 is completed, the synonymous compound noun specifying unit 140 then performs a synonym determination process (step S74) for the unconfirmed word pair. The process of step S74 is performed by the word synonym determination unit 144 of the synonym compound noun specification unit 140. The word synonym determination unit 144 refers to the similarity transition table 195 to calculate a correlation coefficient between the transition of the semantic similarity of the synonym word pair and the transition of the semantic similarity of the undefined word pair, and the correlation coefficient Is equal to or greater than the threshold, it is determined that the words of the undefined word pair are synonymous. The word synonym determination unit 144 changes an unconfirmed word pair that has been determined to be synonymous to a synonym word pair. In the process of step S74 according to the present embodiment, the word synonym determination unit 144 determines whether the unconfirmed word pair is a synonym word pair based on the determination threshold set in step S76. That is, the word synonym determination unit 144 uses the determination threshold set in step S76 for the determination in step S7408 in FIG. 7B.

  Next, the synonymous compound noun specifying unit 140 performs synonym determination processing for a pair of compound nouns (step S75). The process of step S75 is performed by the compound noun synonym determination unit 145 of the synonym compound noun identification unit 140. The compound noun synonym determination unit 145 determines whether or not each of a plurality of word pairs in a compound noun pair having the possibility of synonymous expression is a synonym word pair. When all the word pairs in the compound noun pair are synonymous word pairs, the compound noun synonym determination unit 145 determines that the compound noun pair is a compound noun pair of synonymous expressions, and the compound noun pair is synonymously expressed. Register in the list 196.

  When the process of step S75 is completed, the synonymous compound noun specifying unit 140 ends the synonymous compound noun specifying process (step S7).

  Thus, in the synonymous compound noun identification process (step S7) according to the present embodiment, a determination threshold value used for determining whether or not a word pair is synonymous based on a learning result of semantic similarity in a plurality of word pairs. Set. The determination threshold value setting process (step S76) is performed by the determination threshold value setting unit 146 as described above. The determination threshold setting unit 146 performs, for example, the process illustrated in FIG. 16 as the determination threshold setting process.

FIG. 16 is a flowchart for explaining the contents of the determination threshold value setting process.
The determination threshold setting unit 146 extracts the learning result of the semantic similarity of the synonym word pair from the similarity transition table 195 (step S7601). In step S7601, the determination threshold value setting unit 146 acquires, for example, the array ID of the word pair having the same attribute from the word pair list 404 in FIG. 10B, and obtains the learning result of the semantic similarity of the array ID. Extracted from the similarity transition table 195.

  Next, the determination threshold value setting unit 146 performs a loop process (steps S7602 to S7604) for calculating a correlation coefficient between the semantic similarity of the synonym word pairs. One process in the loop process is a process of calculating a correlation coefficient between semantic similarities in each of two pairs of synonymous words (step S7603). At the loop end (steps S7602 and S7604) in the loop processing, the determination threshold value setting unit 146 selects a correlation coefficient for a combination of two synonym word pairs for calculating a correlation coefficient and all synonym word pairs. It is determined whether or not it has been calculated. When the correlation coefficient is calculated for all combinations in the case of extracting two sets of synonym word pairs from among a plurality of synonym word pairs to be processed, the determination threshold value setting unit 146 ends the loop process.

  When the loop processing (Steps S7602 to S7604) for calculating the correlation coefficient of the semantic similarity in the two pairs of synonymous words is finished, the determination threshold setting unit 146 next calculates the correlation coefficient between the calculated semantic similarities. An average value is calculated (step S7605). In step S7605, the determination threshold setting unit 146 calculates an average value of correlation coefficients from the plurality of correlation coefficients calculated in the loop process (steps S7602 to S7604).

  Thereafter, the determination threshold value setting unit 146 sets the calculated average value of the correlation coefficients as the determination threshold value of the synonymous expression for the word pair (step S7606), and ends the determination threshold value setting process.

  After completing the determination threshold value setting process (step S76), the synonymous compound noun identification unit 140 performs the synonym determination process (step S74) for the word pair as described above, and the unconfirmed word pair is a synonym word pair. It is determined whether or not there is. The word synonym determination unit 144 performs the process in step S74. The word synonym determination unit 144 performs, for example, the processes of steps S7401 to S7411 illustrated in FIGS. 7A and 7B. At this time, the word synonym determination unit 144 uses the determination threshold set in step S76 as the determination threshold in the determination in step S7408.

FIG. 17 is a diagram illustrating a specific example of a determination threshold setting method.
FIG. 17A shows the word pair list 404 exemplified in the first embodiment. As described above, the word pair list 404 is created by the word pair creation unit 141 by performing steps S71 and S72 in the synonymous compound noun identification process (step S7). In the word pair list 404, word pairs extracted from compound noun pairs and synonymous word pairs collected from the document set 191 are registered. Also, in the word pair list 404, each word pair is given an “unconfirmed” or “synonymous” attribute indicating whether or not they are synonymous, and an array ID for identifying the word pair.

  In the determination threshold value setting process according to the present embodiment, the determination threshold value setting unit 146 determines that the uncertain word pair is a synonym word pair based on the correlation coefficient between the learning results (transitions) of the semantic similarity of the synonym word pair. It is determined whether or not there is. Therefore, the determination threshold value setting unit 146 calculates the correlation coefficient of the semantic similarity for all combinations (sequence ID pairs) when selecting two sequence IDs from among the sequence IDs for the synonym word pairs. . For example, in the word pair list 404 of FIG. 17A, the attribute of the word pair whose array ID is 1 is “indeterminate”, and the attribute of the word pair whose array ID is 2 to 4 and N Is synonymous. Therefore, the determination threshold value setting unit 146 generates array ID pairs {2, 3}, {2, 4} and the like given to the synonym word pairs as shown in the table 406 of FIG. The correlation coefficient of the semantic similarity is calculated for each pair of array IDs (steps S7602 to S7604). The determination threshold value setting unit 146 calculates a correlation coefficient of semantic similarity according to the above equation (1). When the correlation coefficient of the semantic similarity is calculated for all the array ID pairs, the determination threshold setting unit 146 calculates an average value of the correlation coefficients from the calculated plurality of correlation coefficients (step S7605).

  As described above, in this embodiment, the determination threshold used for determining whether or not the uncertain word pair extracted from the compound noun pair is synonymous is the semantic similarity of the synonym word pairs collected from the document set 191. Automatically set based on correlation coefficient. As shown in FIG. 12 (a), the relationship between the number of cases and the semantic similarity in the learning result of the semantic similarity for each of a plurality of synonymous word pairs increases as the number of cases increases. However, there is a difference in the amount of change (straight line). For this reason, by setting the determination threshold based on the correlation coefficient of the semantic similarity of the synonym word pairs in the document set 191 including the document from which the compound nouns of synonymous expressions are extracted, the content of the documents in the document set 191 is set. It is possible to determine whether the word pair is synonymous with the threshold. Therefore, according to the present embodiment, it is possible to improve the determination accuracy of whether or not the unconfirmed word pair is synonymous.

  Note that the flowcharts of FIGS. 15 and 16 are only examples of synonymous compound noun identification processing according to the present embodiment. The synonymous compound noun identification processing according to the present embodiment is not limited to the processing illustrated in FIGS. 15 and 16, and the processing content can be appropriately changed without departing from the gist of the present embodiment.

[Third Embodiment]
FIG. 18 is a diagram showing a system configuration of a synonym dictionary creation system according to the third embodiment.

  As shown in FIG. 18, the synonym dictionary creation system 5 according to the present embodiment includes a synonym expression extraction device 1, a first storage device 6, and a second storage device 7. The synonymous expression extraction device 1 is connected to each of the first storage device 6 and the second storage device 7 by a transmission cable or the like, and can transmit and receive data between the devices. The synonymous expression extraction device 1, the first storage device 6, and the second storage device 7 are connected to a network 8 such as the Internet or a local area network (LAN), and the terminal device 9 is connected via the network 8. (9A to 9C) are communicably connected.

  As described in the first embodiment or the second embodiment, the synonym expression extraction apparatus 1 according to the present embodiment extracts synonym expressions for compound nouns included in a document set. The synonymous expression extraction device 1 according to the present embodiment includes a communication unit that performs communication (data transmission / reception) with each of the first storage device 6 and the second storage device 7.

  The first storage device 6 is a device that stores a document set for extracting synonymous expressions of compound nouns. For example, as shown in FIG. 18, the first storage device 6 includes a plurality of documents in which only document data in a predetermined field including a first field document set 601 and a second field document set 602 are stored. Remember a set of documents. In the plurality of document sets 601 and 602, for example, document data collected by the user of the synonym dictionary creation system 5 using the terminal device 9 and transferred from the terminal device 9 to the first storage device 6 is accumulated. The

  The second storage device 7 is a device that stores a synonym dictionary including a pair of compound nouns that are synonymous expressions. For example, as shown in FIG. 18, the second storage device 7 stores a plurality of synonym dictionaries including a synonym dictionary 701 in the first field and a synonym dictionary 702 in the second field. The synonym dictionary 701 in the first field is a dictionary in which synonyms for compound nouns extracted from the document set 601 in the first field are registered. The synonym dictionary 702 in the second field is a dictionary in which synonyms for compound nouns extracted from the document set 701 in the second field are registered. In the plurality of synonym dictionaries 701 and 702, for example, pairs of compound nouns that are synonymous expressions extracted from the document sets 601 and 602 by the synonym expression extracting apparatus 1 are registered. The user of the synonym dictionary creation system 5 uses the terminal device 9 to access the synonym dictionaries 701 and 702 of the second storage device 7 and uses other expressions (synonymous expressions) for compound nouns in the document. It is possible to investigate. The synonym dictionary stored in the second storage device 7 may include not only synonyms for compound nouns but also synonyms for various words.

  In the synonym dictionary creation system 5 according to the present embodiment, for example, a user of the system collects various document data using the terminal device 9 and accumulates them in the document sets 601 and 602 as needed. The synonym expression extraction device 1 in the synonym dictionary creation system 5 receives, for example, a command that the user of the system has transmitted to the synonym expression extraction device 1 using the terminal device 9, or a preset date and time. The process as shown in FIG. 4 is performed in response to the arrival. At this time, the synonym expression extracting device 1 acquires a document set of one field from the first storage device 6 and extracts a synonym compound noun pair. After extracting synonymous compound noun pairs, the synonym expression extracting device 1 registers the extracted compound noun pairs in a predetermined synonym dictionary of the second storage device 7. In addition, when registering the extracted compound noun pair in the synonym dictionary, the synonym expression extracting device 1 refers to the compound noun pair registered in the synonym dictionary and adds only the unregistered pair to the synonym dictionary. sign up.

  As described above, the synonym dictionary creation system 5 according to the present embodiment extracts a synonym expression of compound nouns used in a field based on a document set in which only documents in a specific field are collected. Register (store) in the prepared synonym dictionary. Furthermore, in the synonym dictionary creation system 5 according to the present embodiment, as described in the first embodiment and the second embodiment, the compound noun pair is divided into a plurality of word pairs, and a plurality of word pairs are obtained. Compound noun pairs that are all synonymous word pairs are determined to be synonymous expressions. For this reason, in the synonym dictionary creation system 5 according to the present embodiment, it is possible to efficiently and accurately extract synonym expressions of compound nouns used only in a specific field and create a synonym dictionary. .

  Note that the system configuration in FIG. 18 is merely an example of the system configuration of the synonym dictionary creation system 5 according to the present embodiment. The synonym dictionary creation system 5 according to the present embodiment is not limited to the system configuration of FIG. 18 and can be changed as appropriate. For example, only the document set of one specific field may be stored in the first storage device 6 and only the synonym dictionary of the specific field may be stored in the second storage device 7. Also, one synonym dictionary is prepared in the second storage device 7, and synonym expressions of compound nouns extracted from each of a plurality of document sets are registered together in the one synonym dictionary. Good. Further, the synonym dictionary creation system 5 includes a single storage device that stores a document set and a synonym dictionary instead of the first storage device 6 and the second storage device 7. Also good.

[Fourth Embodiment]
FIG. 19 is a diagram showing a system configuration of a document rewriting system according to the fourth embodiment.

  As shown in FIG. 19, the document rewriting system 10 according to this embodiment includes a synonym expression extracting device 1, a first storage device 6, a second storage device 7, and a document data rewriting device 11. . The synonymous expression extraction device 1 is connected to each of the first storage device 6 and the second storage device 7 by a transmission cable or the like, and can transmit and receive data between the devices. The document data rewriting device 11 is connected to the second storage device 7 by a transmission cable or the like, and data can be transmitted and received between the devices. Furthermore, the synonymous expression extraction device 1, the first storage device 6, the second storage device 7, and the document data rewriting device 11 are connected to a network 8 such as the Internet or a local area network (LAN). The terminal device 9 (9A to 9C) is connected to be communicable with each other.

  As described in the first embodiment or the second embodiment, the synonym expression extraction apparatus 1 according to the present embodiment extracts synonym expressions for compound nouns included in a document set. The synonymous expression extraction device 1 according to the present embodiment includes a communication unit that performs communication (data transmission / reception) with each of the first storage device 6 and the second storage device 7.

  The first storage device 6 is a device that stores a document set for extracting synonymous expressions of compound nouns. For example, as shown in FIG. 18, the first storage device 6 includes a plurality of documents in which only document data in a predetermined field including a first field document set 601 and a second field document set 602 are stored. Remember a set of documents. For example, document data collected by the user of the document rewriting system 10 using the terminal device 9 and transferred from the terminal device 9 to the first storage device 6 is stored in the plurality of document sets 601 and 602.

  The second storage device 7 is a device that stores a synonym dictionary including a pair of compound nouns that are synonymous expressions. For example, as shown in FIG. 18, the second storage device 7 stores a plurality of synonym dictionaries including a synonym dictionary 701 in the first field and a synonym dictionary 702 in the second field. The synonym dictionary 701 in the first field is a dictionary in which synonyms for compound nouns extracted from the document set 601 in the first field are registered. The synonym dictionary 702 in the second field is a dictionary in which synonyms for compound nouns extracted from the document set 701 in the second field are registered. In the plurality of synonym dictionaries 701 and 702, for example, pairs of compound nouns that are synonymous expressions extracted from the document sets 601 and 602 by the synonym expression extracting apparatus 1 are registered. The user of the document rewriting system 10 accesses the synonym dictionaries 701 and 702 of the second storage device 7 by using the terminal device 9 and examines other expressions (synonymous expressions) of compound nouns in the document. Is possible. The synonym dictionary stored in the second storage device 7 may include not only synonyms for compound nouns but also synonyms for various words.

  The document data rewriting device 11 is a device that refers to the synonym dictionaries 701 and 702 stored in the second storage device 7 and rewrites compound nouns included in the document data into synonymous expressions. For example, the document data rewriting device 11 extracts compound nouns from document data transmitted to the document data rewriting device 11 by the user of the document rewriting system 11 using the terminal device 9. In addition, the document data rewriting device 11 searches the synonym dictionary of the second storage device 7 and acquires another expression (synonymous expression) of the compound noun extracted from the document data. When there is a synonym expression for the compound noun extracted from the document data, the document data rewriting device 11 determines whether or not to rewrite the compound noun with another synonym expression. For example, the document data rewriting device 11 determines whether to rewrite the compound noun based on the priority between the compound noun extracted from the document data and the synonym expression acquired from the synonym dictionary. The priority of synonymous expressions is set based on, for example, the frequency of appearance in a document set. When the synonym expression acquired from the synonym dictionary is an expression with higher priority, the document data rewriting device 11 rewrites the compound noun in the document data with the synonym expression acquired from the synonym dictionary. When the above processes are completed, the document rewriting data device 11 returns document data to the terminal device 9.

FIG. 20 is a diagram illustrating a functional configuration of the document data rewriting device.
As shown in FIG. 20, the document data rewriting device 11 includes a document data acquisition unit 1110, a character string extraction unit 1120, a morpheme analysis unit 1130, a compound noun extraction unit 1140, a synonym expression search unit 1150, and a synonym expression. A rewriting unit 1160 and a document data reply unit 1170 are provided.

  The document data acquisition unit 1110 receives input of document data transmitted from the terminal device 9 to the document data rewriting device 11. The document data return unit 1170 returns the document data that has been rewritten by the document data rewriting device 11 to the terminal device 9.

  The character string extraction unit 1120 and the morpheme analysis unit 1130 have the same functions as the character string extraction unit 110 and the morpheme analysis unit 120 in the synonym expression extraction device 1 according to the first embodiment, respectively. The character string extraction unit 1120 extracts a character string from document data that is a target of rewriting processing. The morpheme analysis unit 1130 refers to the analysis dictionary 1190 and performs morpheme analysis on the extracted character string.

  The compound noun extraction unit 1140 extracts a compound noun included in the character string of the document data based on the result of the morphological analysis.

  The synonym expression search unit 1150 searches the synonym dictionary of the second storage device 7 and acquires a synonym expression for the compound noun extracted from the character string.

  The synonym expression rewriting unit 1160 rewrites the compound noun included in the character string of the document data with the synonym expression acquired from the synonym dictionary. The synonym expression rewriting unit 1160 rewrites the compound noun to the synonym expression when the synonym expression acquired from the synonym dictionary has a higher priority than the compound noun included in the character string of the document data.

  When a command for instructing rewriting of compound nouns in document data is input from the terminal device 9 or the like, the document data rewriting device 11 in the document rewriting system 10 according to the present embodiment acquires the document data specified by the command. Then, the process shown in FIG.

FIG. 21 is a flowchart for explaining processing performed by the document data rewriting device.
The document data rewriting device 11 that has acquired the document data to be rewritten first extracts a character string from the document data (step S11). The character string extraction unit 1120 performs the process in step S11.

  Next, the document data rewriting device 11 performs morphological analysis on the extracted character string (step S12). The processing in step S12 is performed by the morphological analysis unit 1130. The morpheme analyzer 1130 divides the character string into morphemes with reference to the analysis dictionary 1190 according to a known analysis method.

  Next, the document data rewriting device 11 extracts a compound noun from the character string based on the result of the morphological analysis (step S13). The compound noun extraction unit 1140 performs the process in step S13. The compound noun extraction unit 1140 extracts a compound noun from the character string in accordance with the compound noun extraction condition. For example, the compound noun extraction condition includes a condition that a word string in which a plurality of words (morphemes) that are nouns are continuous is extracted as a compound noun.

  Next, the document data rewriting device 11 performs a loop process (steps S14 to S19) that searches the synonym dictionary and rewrites the compound noun extracted from the document data into a synonym expression. The loop processing is performed by the synonym expression search unit 1150 and the synonym expression rewrite unit 1160. One process (steps S15 to S18) in the loop process is a process of searching for and rewriting synonymous expressions for one type of compound noun extracted from the document data. At the loop end (steps S14 and S19) in the loop processing, the synonym expression search unit 1150 selects one type of compound nouns extracted from the document data and synonyms for all compound nouns extracted from the document data. It is determined whether the expression has been searched. When the synonym expression is searched for in all of the extracted one or more types of compound nouns and rewritten as necessary, the synonym expression search unit 1150 ends the loop processing.

  In one processing in the loop processing (steps S14 to S19), first, the synonym expression search unit 1150 searches the synonym dictionary of the second storage device 7 using the compound noun selected as the processing target as a keyword ( Step S15).

  Next, the synonym expression search unit 1150 determines whether another synonym expression for the compound noun to be processed is registered in the synonym dictionary (step S16). When another synonym expression is not registered in the synonym dictionary (step S16; NO), the synonym expression search unit 1150 determines whether or not to end the loop process at the loop end (step S19) of the loop process. .

  When another synonym expression for the compound noun to be processed is registered in the synonym dictionary (step S16; YES), the synonym expression search unit 1150 then synonyms other than the compound noun extracted from the document data. It is determined whether or not the expression has a higher priority (step S17). If the compound noun extracted from the document data has a higher priority expression (step S17; NO), the synonym expression search unit 1150 determines whether or not to end the loop process at the loop end of the loop process (step S19). Determine.

  On the other hand, if the other synonym expression is a higher priority expression (step S17; YES), the synonym expression search unit 1150 then causes the synonym expression rewriting unit 1160 to transfer the compound noun in the document data to another synonym. The expression is rewritten (step S18). Thereafter, the synonym expression search unit 1150 determines whether or not to end the loop processing at the loop end of the loop processing (step S19).

  When it is determined at the loop end of the loop processing (step S19) that the loop processing is to be ended, the synonym expression search unit 1150 ends the loop processing. After completing the loop processing, for example, the synonym expression search unit 1150 causes the synonym expression rewrite unit 1160 to return the document data in which the compound noun is rewritten to another synonym expression to the terminal device 9 (step S20).

  When the document data is returned to the terminal device 9, the document data rewriting device 11 finishes the rewriting process for one piece of document data.

  As described above, the document rewriting system 10 according to the present embodiment refers to the synonym dictionary created by the method described in the first embodiment or the second embodiment on the basis of the document set for each field, and the document data Rewrite the compound noun with another synonymous expression. Further, in the document rewriting system 10, as described above, when other synonymous expressions of the compound noun have higher priority than compound nouns extracted from the document data, Rewrite it as a synonymous expression. For this reason, according to the present embodiment, synonymous compound nouns existing in document data can be unified into one notation.

  Note that the flowchart of FIG. 21 is merely an example of processing performed by the document data rewriting device 11 according to the present embodiment. The processing performed by the document data rewriting device 11 according to the present embodiment is not limited to the processing illustrated in FIG. 21, and the processing content can be appropriately changed without departing from the gist of the present embodiment.

  Further, the system configuration of FIG. 19 is only an example of the system configuration of the document rewriting system 10 according to the present embodiment. The document rewriting system 10 according to the present embodiment is not limited to the configuration illustrated in FIG. 19. For example, the synonym expression extracting device 1 and the document data rewriting device 11 may be integrated. Further, the document set and the synonym dictionary may be stored in one storage device.

  In addition, the synonymous expression extraction apparatus 1 according to each of the above embodiments can be realized by a computer and a program executed by the computer. Hereinafter, the synonymous expression extraction apparatus 1 realized by a computer and a program will be described with reference to FIG.

FIG. 22 is a diagram illustrating a hardware configuration of a computer.
As shown in FIG. 22, the computer 15 includes a processor 1501, a main storage device 1502, an auxiliary storage device 1503, an input device 1504, an output device 1505, an input / output interface 1506, a communication control device 1507, and a medium. A driving device 1508. These elements 1501 to 1508 in the computer 15 are connected to each other by a bus 1510 so that data can be exchanged between the elements.

  The processor 1501 is a central processing unit (CPU), a micro processing unit (MPU), or the like. The processor 1501 controls the overall operation of the computer 15 by executing various programs including an operating system. In addition, the processor 1501 executes a synonym expression extraction program including a process of extracting a pair of synonym compound nouns illustrated in FIGS. 4 to 8, for example.

  The main storage device 1502 includes a read only memory (ROM) and a random access memory (RAM) not shown. In the ROM of the main storage device 1502, for example, a predetermined basic control program read by the processor 1501 when the computer 15 is started is recorded in advance. On the other hand, the RAM of the main storage device 1502 is used as a working storage area as necessary when the processor 1501 executes various programs. The RAM of the main storage device 1502 can be used for storing, for example, a character string extracted from the document set 191, an analysis result corpus 193, a list of unconfirmed word pairs and synonym word pairs, a similarity transition table 195, and the like. The RAM of the main storage device 1502 can be used for storing, for example, the analysis dictionary 192, the synonym word dictionary 194, the synonym expression list 196, and the like.

  The auxiliary storage device 1503 is a storage device having a larger capacity than the RAM of the main storage device 1502, and includes, for example, a hard disk drive (HDD) and a non-volatile memory (Solid State Drive (SSD)) such as a flash memory. ) Etc. The auxiliary storage device 1503 can be used to store various programs executed by the processor 1501 and various data. The auxiliary storage device 1503 can be used, for example, for storing a synonym expression extraction program including a process of extracting a synonym compound noun pair shown in FIGS. The auxiliary storage device 1503 can be used for storing, for example, a character string extracted from the document set 191, an analysis result corpus 193, a list of unconfirmed word pairs and synonym word pairs, a similarity transition table 195, and the like. Further, the auxiliary storage device 1503 can be used for storing, for example, the analysis dictionary 192, the synonym word dictionary 194, the synonym expression list 196, and the like.

  The input device 1504 is, for example, a keyboard device or a touch panel device. When an operator (user) of the computer 15 performs a predetermined operation on the input device 1504, the input device 1504 transmits input information associated with the operation content to the processor 1501. The input device 1504 can be used, for example, for inputting a command for starting synonymous expression extraction processing for compound nouns, commands for other processing that can be executed by the computer 15, inputting various setting values, and the like.

  The output device 1505 is, for example, a display device such as a liquid crystal display device or a printing device such as a printer. The output device 1505 can be used to output the results of synonym expression extraction processing for compound nouns and the contents of the synonym expression list.

  The input / output interface 1506 connects the computer 15 and other electronic devices. The input / output interface 1506 includes, for example, a universal serial bus (USB) standard connector. The input / output interface 1506 can be used, for example, for connection between the computer 15 and the storage devices 6 and 7.

  The communication control device 1507 is a device that connects the computer 15 to a network such as the Internet and controls various communications between the computer 15 and other communication devices via the network. The communication control device 1507 can be used for communication between the computer 15 and the terminal device 9, for example.

  The medium driving device 1508 reads programs and data recorded in the portable storage medium 16 and writes data stored in the auxiliary storage device 1503 to the portable storage medium 16. For the medium driving device 1508, for example, a memory card reader / writer corresponding to one type or a plurality of types of standards can be used. When a memory card reader / writer is used as the medium driving device 1508, the portable storage medium 16 is a memory card (flash memory) conforming to a standard supported by the memory card reader / writer, for example, Secure Digital (SD) standard. ) Etc. can be used. As the portable recording medium 16, for example, a flash memory provided with a USB standard connector can be used. Further, when the computer 15 is equipped with an optical disk drive that can be used as the medium driving device 1508, various optical disks that can be recognized by the optical disk drive can be used as the portable recording medium 16. Examples of the optical disc that can be used as the portable recording medium 16 include a Compact Disc (CD), a Digital Versatile Disc (DVD), and a Blu-ray Disc (Blu-ray is a registered trademark). The portable recording medium 16 can be used, for example, for storing a synonym expression extraction program including a process of extracting a synonym compound noun pair shown in FIGS. The portable recording medium 16 can be used for storing, for example, a character string extracted from the document set 191, an analysis result corpus 193, a list of unconfirmed word pairs and synonym word pairs, a similarity transition table 195, and the like. Furthermore, the portable recording medium 16 can be used for storing, for example, an analysis dictionary 192, a synonym word dictionary 194, a synonym expression list 196, and the like.

  For example, when an operator inputs an instruction to start synonym expression extraction processing using the input device 1504 or the like to the computer 15, the processor 1501 stores the synonym expression stored in a non-temporary recording medium such as the auxiliary storage device 1503. Read and execute the extraction program. In this process, the processor 1501 functions (operates) as the character string extraction unit 110, the morpheme analysis unit 120, the compound noun extraction unit 130, and the synonym compound noun identification unit 140 in the synonym expression extraction device 1. Further, while the processor 1501 is executing the synonymous expression extraction program, the RAM of the main storage device 1502, the auxiliary storage device 1503, and the like function as a storage unit (not shown) of the synonymous expression extraction device 1. That is, the RAM of the main storage device 1502, the auxiliary storage device 1503, and the like store the document set 191, the analysis dictionary 192, the analysis result corpus 193, the synonym word dictionary 194, the similarity transition table 195, the synonym expression list 196, and the like. Functions as a storage unit.

  Note that the computer 15 that operates as the synonymous expression extraction apparatus 1 does not need to include all the elements 1501 to 1508 illustrated in FIG. 22, and some elements may be omitted depending on applications and conditions. For example, the computer 15 may be one in which the communication control device 1507 and the medium driving device 1508 are omitted.

  The storage device such as the auxiliary storage device 1503 in the computer 15 can also be used as the first storage device 6 and the second storage device 7 in the synonym dictionary creation system shown in the third embodiment, for example. It is.

  Further, the computer 15 can be operated not only as the synonym expression extracting device 1 but also as the document data rewriting device 11 of the document rewriting system 10 shown in the fourth embodiment. In the document rewriting system 10, one computer 15 can be operated as the synonym expression extracting device 1 and can also be operated as the document data rewriting device 11.

The following additional notes are disclosed for each of the embodiments described above.
(Appendix 1)
The compound noun pair extracted from the document data is divided into a plurality of word pairs, and the synonym word dictionary in which the synonym word pairs are registered is referred to, and the plurality of word pairs are synonymous with the synonym word pairs. A word pair setting unit for identifying an unconfirmed word pair that is not confirmed,
A process of learning the semantic similarity between words is performed a plurality of times for each of the plurality of synonym word pairs in the document data including the synonym word pair in the compound noun pair and the uncertain word pair. A semantic similarity learning unit;
Whether or not the words of the undefined word pair are synonymous based on the learning result of the semantic similarity of the undefined word pair and the learning result of the semantic similarity in each of the plurality of synonymous word pairs. A word synonym determination unit for determining;
A compound noun synonym determination unit that determines that the compound noun pair is a synonymous expression when the plurality of word pairs in the compound noun pair are all synonymous word pairs;
Each time the semantic similarity learning unit performs a semantic similarity learning process for each of a plurality of word pairs to be processed, a case used for the learning process is added.
The word synonym determination unit includes a relationship between the number of learning processes and the semantic similarity in the learning result of the semantic similarity of the undefined word pair, and the number of learning processes in the learning result of the semantic similarity of the synonym word pair. Calculating a correlation coefficient with respect to the relationship with the semantic similarity, and determining whether or not the words of the undefined word pair are synonymous based on the correlation coefficient;
A synonymous expression extraction device characterized by that.
(Appendix 2)
The word synonym determination unit
In each of the plurality of synonymous word pairs, the relationship between the number of learning processes for the uncertain word pair and the semantic similarity, and the number of learning processes for the synonymous word pair and the semantic similarity. Calculate the correlation coefficient for the relationship,
When the average value of the plurality of calculated correlation coefficients is equal to or greater than a threshold, it is determined that the words of the undefined word pair are synonymous.
The synonym expression extraction device according to Supplementary Note 1, wherein
(Appendix 3)
In the synonym expression extraction device, the words of the undefined word pair are synonymous based on the relationship between the number of learning processes and the semantic similarity in the learning result of the semantic similarity in each of the plurality of synonymous word pairs. A determination threshold value setting unit that sets a determination threshold value of whether or not
The word synonym determination unit determines that the words of the undefined word pair are synonymous when the calculated average value of the correlation coefficients is equal to or greater than the determination threshold set by the determination threshold setting unit. ,
The synonymous expression extraction apparatus according to Supplementary Note 2, wherein
(Appendix 4)
The synonym expression extraction device further includes a storage unit that stores a synonym expression list in which the compound noun pairs that are synonym expressions are registered.
The synonym expression extraction device according to Supplementary Note 1, wherein
(Appendix 5)
The synonym expression extraction device further includes a synonym expression rewriting unit that rewrites a compound noun extracted from document data into a synonym expression based on the synonym expression of the compound noun registered in the synonym expression list.
The synonym expression extraction device according to Supplementary Note 1, wherein
(Appendix 6)
Computer
The compound noun pair extracted from the document data is divided into a plurality of word pairs, and the synonym word dictionary in which the synonym word pairs are registered is referred to, and the plurality of word pairs are synonymous with the synonym word pairs. Identified as unconfirmed word pairs that have not been confirmed,
Collecting the synonym word pairs registered in the synonym word dictionary from the document data,
For each of the plurality of synonym word pairs in the document data including the undefined word pair and the synonym word pair in the compound noun pair, a process of learning the semantic similarity between words is executed a plurality of times. And
Whether or not the words of the undefined word pair are synonymous based on the learning result of the semantic similarity of the undefined word pair and the learning result of the semantic similarity in each of the plurality of synonymous word pairs. Judgment,
When the plurality of word pairs in the compound noun pair are all synonymous word pairs, the compound noun pair is determined to be a synonym expression, and a process is executed.
In the process of learning the semantic similarity, each time the computer performs a semantic similarity learning process for each of a plurality of word pairs to be processed, it adds a case to be used for the learning process,
In the process of determining whether or not the words of the uncertain word pair are synonymous, the computer includes a relationship between the number of learning processes and the semantic similarity in the learning result of the semantic similarity of the uncertain word pair. , Calculating a correlation coefficient for the relationship between the number of learning processes and the semantic similarity in the learning result of the semantic similarity of the synonymous word pair, and based on the correlation coefficient, the words of the undefined word pair Whether or not is synonymous,
A synonymous expression extraction method characterized by that.
(Appendix 7)
In the process of determining whether or not the words of the undefined word pair are synonymous,
In each of the plurality of synonymous word pairs, the relationship between the number of learning processes for the uncertain word pair and the semantic similarity, and the number of learning processes for the synonymous word pair and the semantic similarity. Calculate the correlation coefficient for the relationship,
When the average value of the plurality of calculated correlation coefficients is equal to or greater than a threshold, it is determined that the words of the undefined word pair are synonymous.
The synonymous expression extraction method according to Supplementary Note 6, wherein
(Appendix 8)
Before performing the process of determining whether or not the words of the uncertain word pair are synonymous, the computer counts the number and meaning of learning processes in the learning result of the semantic similarity in each of the plurality of synonymous word pairs. Based on the relationship with the degree of similarity, further includes a process of setting a determination threshold value as to whether or not the words of the undefined word pair are synonymous,
In the process of determining whether or not the words of the uncertain word pair are synonymous, when the average value of the correlation coefficient is equal to or greater than the determination threshold, the computer Is determined to be synonymous,
The synonymous expression extraction method according to Supplementary Note 7, wherein
(Appendix 9)
The compound noun pair extracted from the document data is divided into a plurality of word pairs, and the synonym word dictionary in which the synonym word pairs are registered is referred to, and the plurality of word pairs are synonymous with the synonym word pairs. Identified as unconfirmed word pairs that have not been confirmed,
Collecting the synonym word pairs registered in the synonym word dictionary from the document data,
For each of the plurality of synonym word pairs in the document data including the undefined word pair and the synonym word pair in the compound noun pair, a process of learning the semantic similarity between words is executed a plurality of times. And
Whether or not the words of the undefined word pair are synonymous based on the learning result of the semantic similarity of the undefined word pair and the learning result of the semantic similarity in each of the plurality of synonymous word pairs. Judgment,
When the plurality of word pairs in the compound noun pair are all synonymous word pairs, the compound noun pair is determined to be a synonym expression, a synonym expression extraction program for causing a computer to execute a process,
The process of learning the semantic similarity includes a process of adding a case to be used for the learning process every time the semantic similarity learning process is performed for each of a plurality of word pairs to be processed.
The process of determining whether or not the words of the undefined word pair are synonymous is the relationship between the number of learning processes and the semantic similarity in the learning result of the semantic similarity of the undefined word pair, and the synonymous word A correlation coefficient is calculated for the relationship between the number of learning processes in the learning result of the semantic similarity of the pair and the relationship between the semantic similarity, and the words of the undefined word pair are synonymous based on the correlation coefficient Including processing to determine whether or not
A synonym expression extraction program characterized by that.
(Appendix 10)
The process of determining whether or not the words of the undefined word pair are synonymous,
In each of the plurality of synonymous word pairs, the relationship between the number of learning processes for the uncertain word pair and the semantic similarity, and the number of learning processes for the synonymous word pair and the semantic similarity. Calculate the correlation coefficient for the relationship,
Including a process of determining that the words of the unconfirmed word pair are synonymous when an average value of the plurality of calculated correlation coefficients is equal to or greater than a threshold value,
The synonymous expression extraction program according to supplementary note 9, characterized by:
(Appendix 11)
The number of learning processes and the semantic similarity in the learning result of the semantic similarity in each of the plurality of synonymous word pairs, executed before the process of determining whether or not the words of the undefined word pair are synonymous. Further including a process of setting a determination threshold for determining whether or not the words of the undefined word pair are synonymous based on the relationship of
The process of determining whether or not the words of the undefined word pair are synonymous is that the words of the undefined word pair are synonymous when the average value of the correlation coefficient is equal to or greater than the determination threshold. To determine,
The synonymous expression extraction program according to supplementary note 10, characterized by:

DESCRIPTION OF SYMBOLS 1 Synonymous expression extraction apparatus 2,3 Compound noun 401-403,405,406 Table 404 Word pair list 5 Synonym dictionary creation system 6,7 Storage apparatus 8 Network 9 (9A-9C) Terminal apparatus 10 Document rewriting system 11 Document data Rewriting device 15 Computer 16 Portable recording medium 110, 1120 Character string extraction unit 120, 1130 Morphological analysis unit 130, 1140 Compound noun extraction unit 140 Synonym compound noun specification unit 141 Word pair setting unit 142 Similarity transition table creation unit 143 Meaning similarity Degree learning unit 144 word synonym determining unit 145 compound noun synonym determining unit 146 determination threshold setting unit 191, 601, 602 document set 192, 1190 analysis dictionary 103 analysis result corpus 194 synonym word dictionary 195 similarity transition table 196 synonym expression list 701 702 Synonym dictionary 1110 Document data acquisition unit 1150 Synonym expression search unit 1160 Synonym expression rewrite unit 1170 Document data return unit 1501 Processor 1502 Main storage device 1503 Auxiliary storage device 1504 Input device 1505 Output device 1506 Input / output interface 1507 Communication control device 1508 Auxiliary storage apparatus

Claims (6)

The compound noun pair extracted from the document data is divided into a plurality of word pairs, and the synonym word dictionary in which the synonym word pairs are registered is referred to, and the plurality of word pairs are synonymous with the synonym word pairs. A word pair setting unit for identifying an unconfirmed word pair that is not confirmed,
A process of learning the semantic similarity between words is performed a plurality of times for each of the plurality of synonym word pairs in the document data including the synonym word pair in the compound noun pair and the uncertain word pair. A semantic similarity learning unit;
Whether or not the words of the undefined word pair are synonymous based on the learning result of the semantic similarity of the undefined word pair and the learning result of the semantic similarity in each of the plurality of synonymous word pairs. A word synonym determination unit for determining;
A compound noun synonym determination unit that determines that the compound noun pair is a synonymous expression when the plurality of word pairs in the compound noun pair are all synonymous word pairs;
Each time the semantic similarity learning unit performs a semantic similarity learning process for each of a plurality of word pairs to be processed, a case used for the learning process is added.
The word synonym determination unit includes a relationship between the number of learning processes and the semantic similarity in the learning result of the semantic similarity of the undefined word pair, and the number of learning processes in the learning result of the semantic similarity of the synonym word pair. Calculating a correlation coefficient with respect to the relationship with the semantic similarity, and determining whether or not the words of the undefined word pair are synonymous based on the correlation coefficient;
A synonymous expression extraction device characterized by that. The word synonym determination unit
In each of the plurality of synonymous word pairs, the relationship between the number of learning processes for the uncertain word pair and the semantic similarity, and the number of learning processes for the synonymous word pair and the semantic similarity. Calculate the correlation coefficient for the relationship,
When the average value of the plurality of calculated correlation coefficients is equal to or greater than a threshold, it is determined that the words of the undefined word pair are synonymous.
The synonymous expression extraction apparatus according to claim 1, wherein: In the synonym expression extraction device, the words of the undefined word pair are synonymous based on the relationship between the number of learning processes and the semantic similarity in the learning result of the semantic similarity in each of the plurality of synonymous word pairs. A determination threshold value setting unit that sets a determination threshold value of whether or not
The word synonym determination unit determines that the words of the undefined word pair are synonymous when the calculated average value of the correlation coefficients is equal to or greater than the determination threshold set by the determination threshold setting unit. ,
The synonymous expression extraction apparatus according to claim 2, wherein The synonym expression extraction device further includes a storage unit that stores a synonym expression list in which the compound noun pairs that are synonym expressions are registered.
The synonymous expression extraction apparatus according to claim 1, wherein: Computer
The compound noun pair extracted from the document data is divided into a plurality of word pairs, and the synonym word dictionary in which the synonym word pairs are registered is referred to, and the plurality of word pairs are synonymous with the synonym word pairs. Identified as unconfirmed word pairs that have not been confirmed,
Collecting the synonym word pairs registered in the synonym word dictionary from the document data,
For each of the plurality of synonym word pairs in the document data including the undefined word pair and the synonym word pair in the compound noun pair, a process of learning the semantic similarity between words is executed a plurality of times. And
Whether or not the words of the undefined word pair are synonymous based on the learning result of the semantic similarity of the undefined word pair and the learning result of the semantic similarity in each of the plurality of synonymous word pairs. Judgment,
When the plurality of word pairs in the compound noun pair are all synonymous word pairs, the compound noun pair is determined to be a synonym expression, and a process is executed.
In the process of learning the semantic similarity, each time the computer performs a semantic similarity learning process for each of a plurality of word pairs to be processed, it adds a case to be used for the learning process,
In the process of determining whether or not the words of the uncertain word pair are synonymous, the computer includes a relationship between the number of learning processes and the semantic similarity in the learning result of the semantic similarity of the uncertain word pair. , Calculating a correlation coefficient for the relationship between the number of learning processes and the semantic similarity in the learning result of the semantic similarity of the synonymous word pair, and based on the correlation coefficient, the words of the undefined word pair Whether or not is synonymous,
A synonymous expression extraction method characterized by that. The compound noun pair extracted from the document data is divided into a plurality of word pairs, and the synonym word dictionary in which the synonym word pairs are registered is referred to, and the plurality of word pairs are synonymous with the synonym word pairs. Identified as unconfirmed word pairs that have not been confirmed,
Collecting the synonym word pairs registered in the synonym word dictionary from the document data,
For each of the plurality of synonym word pairs in the document data including the undefined word pair and the synonym word pair in the compound noun pair, a process of learning the semantic similarity between words is executed a plurality of times. And
Whether or not the words of the undefined word pair are synonymous based on the learning result of the semantic similarity of the undefined word pair and the learning result of the semantic similarity in each of the plurality of synonymous word pairs. Judgment,
When the plurality of word pairs in the compound noun pair are all synonymous word pairs, the compound noun pair is determined to be a synonym expression, a synonym expression extraction program for causing a computer to execute a process,
The process of learning the semantic similarity includes a process of adding a case to be used for the learning process every time the semantic similarity learning process is performed for each of a plurality of word pairs to be processed.
The process of determining whether or not the words of the undefined word pair are synonymous is the relationship between the number of learning processes and the semantic similarity in the learning result of the semantic similarity of the undefined word pair, and the synonymous word A correlation coefficient is calculated for the relationship between the number of learning processes in the learning result of the semantic similarity of the pair and the relationship between the semantic similarity, and the words of the undefined word pair are synonymous based on the correlation coefficient Including processing to determine whether or not
A synonym expression extraction program characterized by that.

Images