JP2009245179A - Document retrieval support device - Google Patents

Abstract

A user's burden associated with document search is reduced.
A document search support apparatus extracts a document file including a designated search word as a candidate document. When a document file similar in content to the target document is selected as a similar document among the candidate documents, terms included in the similar document are extracted as candidate terms, and each of them is selected according to the rarity of candidate terms in the predetermined document file group. Rank candidate terms. Then, candidate terms positioned at a predetermined rank or higher are presented to the user as suitable search terms for searching for the target document.
[Selection] Figure 1

Description

  The present invention relates to a document retrieval technique.

With the spread of computers and the development of network technology, the exchange of electronic information via networks has become popular. Many business processes that have been conventionally performed on a paper basis are being replaced by network-based processes. Advances in digitalization and network technology have drastically reduced information acquisition costs. Under such circumstances, the importance of a technique for searching for a desired document file from a large number of document files is increasing.
JP 2002-015001 A

  In a general document search method, a user first inputs one or more words as “search words”. Usually, a word considered to be contained in a document file to be searched (hereinafter referred to as “target document”) is selected as a search word. The target document is specified by trying various search expressions combining a plurality of search terms. In the case of such a document retrieval method, what kind of retrieval formula is created, in other words, what kind of retrieval word is combined and how to decide the success or failure of the target document detection. However, the creation of a search expression is heavily dependent on personal skills such as experience and intuition, and thus places a heavy burden on the user. Also, it is difficult to create a suitable search formula if there is no image of words that are definitely included in the target document.

  As another document search method, there is a technique called concept search (see, for example, Patent Document 1). Concept search is a technique for searching a document file having contents related to a natural sentence input by a user. Since the target document can be searched by inputting the outline of the target document in a natural sentence format, the user is freed from the burden of creating a search expression. However, if the detection of the target document fails, it is difficult to make use of the failure of the search for the next search because it is not known where and how to correct the natural sentence that failed to be searched to obtain a more preferable search result.

  The present invention has been completed based on the above-described problems, and its main object is to provide a technique for reducing the burden on the user associated with document search and improving the efficiency of document search.

One embodiment of the present invention relates to a document search support apparatus.
This apparatus acquires a document file similar in content to the target document as a similar document, extracts terms included in the similar document as candidate terms, and sets each candidate term according to the rarity of candidate terms in a predetermined document file group. Ranking. Then, candidate terms positioned at a predetermined rank or higher are presented to the user as suitable search terms for searching for the target document.

The number of similar documents is not necessarily one, and may be a plurality. When extracting candidate terms from multiple similar documents, terms included in any of the multiple similar documents may be extracted as candidate terms, or included in common to any of multiple similar documents Terms that have been identified may be extracted as candidate terms.
The “term” may be a so-called “morpheme” or a combination of morphemes. For example, two terms “user” and “interface” may be further extracted from the term “user interface”.

  It should be noted that any combination of the above-described components, and the present invention expressed by a method, system, recording medium, and computer program are also effective as an aspect of the present invention.

  According to the present invention, it is possible to reduce the burden on the user in document search and realize efficient document search.

FIG. 1 is a schematic diagram of document search in this embodiment.
Assume a situation in which a desired published patent publication is detected as a “target document” from the patent database. The patent database may be an existing database such as “Patent Digital Library”. Hereinafter, a group of document files stored in a document database such as a patent database and to be searched is referred to as a “corpus”.

  The target document is an open patent publication of an invention related to “stereoscopic vision”. As terms included in the target document, “virtual reality”, “virtual environment”, “virtual space”, “coordinate system”, “stereoscopic view”, “three-dimensional”, “sight”, and so on can be considered. . Assume that the user first inputs “three-dimensional” that seems to be surely included in the target document as a search term. Hereinafter, a search term that is explicitly input by the user is referred to as a “designated search term”.

S1. Search process:
An open patent information including the designated search word “3D” is detected as “candidate document”. Usually, the number of candidate documents including the character string “3D” is enormous. Many of the candidate documents have contents related to technologies other than the virtual reality technology. The user selects, from these candidate document groups, candidate documents that are likely to be similar in content to the target document as “similar documents”. For example, a candidate document including a description related to virtual reality is selected. It is assumed that the user overviews about 20 to 30 candidate documents and selects three of them as similar documents.

S2. Extraction process:
Terms included in each similar document are extracted as “candidate terms”. The candidate terms included in the similar document are likely to be included in the target document.

S3. Rank processing:
Rank candidate terms according to importance. The ranking based on importance is called “importance ranking”. The importance level of the candidate term is calculated based on the number of appearances and the appearance position in the similar document, the rarity seen from the entire corpus, and the like. A method of calculating the importance will be described in detail with reference to FIG. A candidate term with higher importance is more likely to be a useful search term for specifying a target document. Candidate terms whose importance rank is higher than a predetermined rank, for example, within the top 20 are presented as new search term candidates. Hereinafter, the search terms presented as a result of the ranking process are referred to as “suggested search terms”.

  For example, it is assumed that the candidate term with the highest priority is “parallax”. In this case, the user may perform a re-search by a search expression “designated search word * proposed search word (“ * ”indicates a logical product, ie, AND condition)”. In the case of this example, a search expression “stereoscopic * parallax” is input. By the search process (S1) based on this search formula, the candidate documents including both “solid” and “parallax” are narrowed down. The subsequent process is the same.

  According to such a processing method, it is possible to gradually narrow down the number of candidate documents by repeating the addition of designated search terms and the selection of similar documents. Since a new designated search word is selected from the top 20 suggested search words, it is not necessary to consider all of the designated search words to be included in the search formula from the beginning. As a result, the burden of creating search terms and search expressions is greatly reduced.

  In the conventional document search method, after creating a single search expression and executing a document search, it turns out that the search expression, in particular, the search word contained in the search expression is not appropriate. In many cases, a search expression is created from scratch and a document search is executed. Even in the case of conceptual search, after creating a natural sentence and executing a document search, it turns out that the natural sentence is not appropriate, and it is often the case that a completely different natural sentence is created from scratch and the document search is executed. .

  On the other hand, according to the document search method in the present embodiment, the number of candidate documents can be narrowed down gradually while increasing the search terms reasonably. This eliminates the need to re-create the search formula from the beginning. For example, if the search result when narrowed down by the search formula “stereo * parallax” is not suitable, “virtual” in the second priority ranking is adopted instead of “parallax” in the first ranking ranking, Re-searching may be performed using the search expression “ Alternatively, a search expression “3D * (parallax + virtual)” (“+” indicates a logical sum, that is, an OR condition) may be employed. By gradually narrowing the search range from a general designated search word such as “three-dimensional”, it becomes easier to reach the target document.

  It is not always necessary to select a similar document from candidate documents, and a predetermined similar document may be input as it is. For example, if you are thinking about applying for a patent for an idea contained in a manuscript or design document, if you enter these manuscript manuscripts as similar documents, whether or not an invention close in content to this idea has already been filed. It becomes easy to judge.

FIG. 2 is a functional block diagram of the document search support apparatus 100.
Each block shown here can be realized in hardware by an element such as a CPU of a computer or a mechanical device, and in software it is realized by a computer program or the like. Draw functional blocks. Therefore, those skilled in the art will understand that these functional blocks can be realized in various forms by a combination of hardware and software.

The document search support apparatus 100 includes an IF (interface) unit 110, a data processing unit 140, and a data holding unit 160.
The IF unit 110 is in charge of an interface with a user or a corpus. The data processing unit 140 executes various data processing based on data acquired from the IF unit 110 and the data holding unit 160. The data holding unit 160 also serves as an interface between the IF unit 110 and the data holding unit 160. The data holding unit 160 is a storage area for holding various data.

IF unit 110:
The IF unit 110 includes a document acquisition unit 112, a search expression acquisition unit 120, a presentation unit 126, and a search execution unit 132. The document acquisition unit 112 and the search execution unit 132 are mainly in charge of the interface with the corpus, and the search expression acquisition unit 120 and the presentation unit 126 are mainly in charge of the interface with the user.

  The document acquisition unit 112 acquires various document files. The document acquisition unit 112 includes a similar document acquisition unit 114, a dissimilar document acquisition unit 116, and a candidate document acquisition unit 118. The candidate document acquisition unit 118 acquires a candidate document specified as a result of the search process (S1) from the corpus. The similar document acquisition unit 114 acquires the similar document selected by the user from the candidate documents from the corpus. The similar document acquisition unit 114 may acquire a similar document directly from the user. The dissimilar document acquisition unit 116 acquires a dissimilar document file in terms of the target document and contents as a “dissimilar document”. The dissimilar document acquisition unit 116 may also acquire dissimilar documents directly from the user. The dissimilar document will be described in detail with reference to FIG. 5 and FIG.

  The search expression acquisition unit 120 receives input of various search expressions for searching for a target document. The search expression includes one or more designated search terms. The search expression acquisition unit 120 includes a search word acquisition unit 122. The search term acquisition unit 122 extracts designated search terms included in the search formula.

The presentation unit 126 presents various information to the user. The presentation unit 126 includes a search word presentation unit 128 and a search expression presentation unit 130. The search term presenting unit 128 displays a suggested search term and “relevant search terms” and “excluded search terms” described later on the screen. The search formula presenting unit 130 further presents a suitable search formula including a suggested search term to the user. The search formula presented by the search formula presentation unit 130 will be described later with reference to FIGS. 4 and 7.
The search execution unit 132 executes document search according to the specified search expression. The function of the search execution unit 132 may be realized by a device other than the document search support device 100.

Data holding unit 160:
The data holding unit 160 includes a related table holding unit 162. The related table holding unit 162 holds a “related table”. In the association table, one term is associated with another term. For example, a term corresponding to a synonym is associated with a certain term. Details of the association table will be described later with reference to FIG.

Data processing unit 140:
The data processing unit 140 includes a term extraction unit 142, a related processing unit 148, and a rank setting unit 154. The term extraction unit 142 extracts terms from the document file. The term extraction unit 142 includes a candidate term extraction unit 144 and an inapplicable word extraction unit 146. The candidate term extraction unit 144 extracts terms in the similar document as candidate terms. The inapplicable word extraction unit 146 extracts the terms in the dissimilar document as “inapplicable words”.

  The rank setting unit 154 calculates the importance level of the candidate terms and sets the importance rank for each candidate term. Candidate terms within a predetermined rank, for example, within the 20th, are shown to the user as suggested search terms by the search term presentation unit 128. Similarly, the rank setting unit 154 calculates the importance of the inapplicable word and sets the rank for each inapplicable word. The ranking based on the importance of non-applicable words is called “unsuitable ranking”.

  The related processing unit 148 is in charge of processing related to related search terms. The related processing unit 148 includes a related search word specifying unit 150 and a related table update unit 152. When the suggested search word is specified, the related search word specifying unit 150 specifies a term associated with the suggested search word in the related table as the related search word. The related table update unit 152 updates the related table based on the search formula input by the user. For example, it is assumed that the user inputs a search expression including two designated search terms “context switch” and “task switch”. In this case, the related table update unit 152 associates “context switch” with “task switch” and registers them in the related table.

FIG. 3 is a conceptual diagram showing a processing process until a proposed search word is specified.
First, it is assumed that the user inputs a character string “natural language” as a designated search term. The search word acquisition unit 122 acquires the character string “natural language” as the designated search word, and the search execution unit 132 detects a published patent publication including the character string “natural language” as a candidate document. The candidate document acquisition unit 118 acquires a candidate document from the corpus. It is assumed that the user selects the published patent publications A to C as similar documents. The similar document acquisition unit 114 acquires a similar document from a corpus or a candidate document group that has already been acquired.

From the similar document A, various candidate terms such as “anomalous words”, “input sentence”, “corpus”, “learning data”, “classifier”, “feature vector” are extracted. Terms whose number of appearances in the similar document A is within a predetermined number of times, for example, three times or less, may be excluded from extraction. Similarly, various candidate terms are extracted from the similar documents B and C. However, general terms registered in the dictionary in advance are not subject to detection. For example, conjunctions such as “but”, “or”, and particles such as “ha”, “ni”, and “from” are not candidate terms. In addition, terms such as “above”, “apparatus”, “preferred”, and “variation” that are general in patent specifications may be excluded.
In the present embodiment, all similar documents, that is, only candidate terms appearing in any of the similar documents A to C are subjected to the subsequent analysis. For example, it is assumed that at least one “abnormal word” extracted from the similar document A appears in the similar document B and the similar document C. Even if the term “recall rate” appears frequently in the similar document A, if the “recall rate” is not extracted from the similar document B, the “recall rate” is a calculation target of the individual importance and the total importance described later. Get out.

  The rank setting unit 154 calculates the individual importance of each candidate term for each similar document. Here, the candidate terms appearing in “Claims” are highly likely to be effective terms in showing the contents of similar documents. On the other hand, the candidate terms appearing in the “application” are not so powerful. Therefore, points are assigned according to the appearance locations of candidate terms. In this example, 0 points are given to the candidate terms appearing in the “application”, 2.0 points to the “claims” and “abstract”, 1.8 points to the “title of the invention”, “ 1.0 point is assigned to “the best mode for carrying out the invention”. If the candidate term “anomalous word” appears five times in the “Claims” of the similar document A and 22 times in the “best mode for carrying out the invention”, the candidate in the similar document A The individual importance of the term “allophone” is 32.0 (= 2.0 × 5 + 1.0 × 22). In the case of the similar document A shown in the figure, the first place “learning data (42.8)”, the second place “abnormal word (32.0)”, the third place “corpus (29.8)”,. The ranking is made as follows. Candidate terms that appear more frequently in important parts such as “Claims” have higher individual importance. For similar documents B and C, the individual importance of candidate terms is calculated.

  Next, the individual importance of each candidate term is integrated, and the total importance of each candidate term is calculated. In calculating the total importance, the concept of “vector space method” is followed. That is, a rare candidate term that does not appear so much in a predetermined document file group is considered to be an effective term for indicating the content of a similar document. Therefore, adjustments are made so that the overall importance of rare candidate terms is relatively high. Conversely, ordinary candidate terms appearing in many document files in this document file group are unlikely to be so powerful terms in indicating the contents of similar documents. Therefore, calculation is performed so that the general importance of general terms is relatively low.

In this example,
Overall importance of candidate term Wi = (n ₁ + n ₂ + n ₃ ) · log {(Dn / Rn) +1}
To calculate the total importance of each candidate term. Dn indicates the total number of document files included in the corpus. For example, it may be the total number of document files included in the patent database. Rn is the number of document files including the candidate term Wi among the document files included in the corpus. log {(Dn / Rn) +1} increases as the number Rn of document files including the candidate term Wi is smaller than the total number Dn of document files. In other words, the smaller the candidate term Wi is, the larger the candidate term Wi becomes. n ₁ , n ₂ , and n ₃ indicate the individual importance levels of the candidate terms Wi in the similar documents A, B, and C, respectively. For example, the candidate term “feature vector” is included in all of the similar documents A to C. Since the candidate term “feature vector” is n ₁ = 18.4, n ₂ = 12.2, and n ₃ = 4.3, n ₁ + n ₂ + n ₃ = 34.9.

Candidate terms with higher individual importance have higher overall importance. In addition, a rare candidate term that appears only in some document files as viewed from the entire corpus has a higher overall importance. In this way, ranking is performed after the total importance of candidate terms extracted from each similar document is calculated. In the case of similar documents A to C in the figure, the first place “feature vector (91.4)”, the second place “learning data (84.8)”, the third place “similarity (81.2)”,... The ranking is made as follows. The search formula presenting unit 130 displays the candidate terms within the top 20 as “suggested search terms” on the screen. The user selects the next designated search term from these suggested search terms. For example, a search expression such as “natural language * feature vector” or “natural language * (feature vector + learning data)” can be considered as the next search expression. Hereinafter, the total importance is simply referred to as “importance”.
Note that the extraction of the candidate terms and selection, the calculation of the importance based on the number of appearances, the position, and the rarity need not be executed in the above order. For example, terms commonly appearing in similar documents A to C are extracted as candidate terms, general terms registered in the dictionary are removed therefrom, candidate terms are further narrowed down according to the number of appearances, and Dn / Rn The individual importance level and the total importance level may be calculated after narrowing down to rare candidate terms that are equal to or greater than a predetermined value.

FIG. 4 is a data structure diagram of the related table.
The related table is a table stored in the related table holding unit 162 and registered in association with a plurality of terms. The basic term column 170 shows each term. When searching for a published patent publication related to “spoon” in a general document search, not only “spoon” but also a term close to “spoon” is often used as a search term. For example, synonyms such as “Saji”, “Saji”, and “Samurai” are candidates, and a search expression “Spoon + Saji + Saji + Samurai” is created. In addition, in the case of the term “virtual reality”, synonyms such as “artificial reality” and “mixed reality”, foreign words such as “virtual reality” and “virtual reality”, and an abbreviation “VR” are also candidates for a search term. In order to prevent omission of search, it is necessary to consider not only one search term but also another search term related to the search term.

  The synonym column 172 shows synonyms for the terms shown in the basic term column 170. Synonyms are registered in the dictionary in advance. For example, the term “spoon” is associated with the synonyms “spoof”, “saji”, and “匙”. When “spoon” is listed as a suggested search word, the related search word specifying unit 150 refers to the related table and specifies “spoons”, “saj”, and “匙” as “related search words”. The search term presentation unit 128 displays not only the proposed search terms but also related search terms on the screen.

  For example, it is assumed that “spoon” is specified as the suggested search term as a result of inputting “fork” as the designated search term. Since “spoon”, “sag”, and “匙” are related to each other in the related table, “spoon” is connected by an OR condition (+ symbol) to form a “search set”. The search expression presenting unit 130 generates a search expression “fork * search set”, that is, “fork * (spoon + spoon + saji + claw)” and displays it on the screen. When “fork” is registered as a synonym of “fork”, the search expression presenting unit 130 includes not only the suggested search word but also related search words for the designated search word “(fork + fork) * ( A search expression “spoon + spoon + saji + 匙)” is generated and displayed on the screen.

  Furthermore, as a result of inputting the search expression “(fork + fork) * (spoon + spouse + saji + 匙)”, “signal” is specified as the suggested search term, and “signal” Assume that a “message” is specified. At this time, a search set of “signal + signal + message” in which “signal”, “signal”, and “message” are combined with OR condition is generated, and “(fork + fork) * (spoon + spoon + sag + sag + 匙) A search expression “* (signal + signal + message)” is generated and displayed on the screen. The same applies thereafter.

  The related word column 174 shows related words. For example, it is assumed that a search expression including both “spoon” and “dish” is input, such as “spoon + dish” and “spoon * dish”. At this time, the related table update unit 152 registers “dish” as a related word of “spoon” or “spoon” as a related word of “dish” in the related table. The related table update unit 152 may record search formulas for a predetermined number of times in the past, and register terms that are particularly frequently combined in the related table. For example, in 1000 search formulas input in the past, if there are a predetermined number of search formulas including both “dish” and “spoon”, for example, 20 or more, “spoon” and “dish” are associated with each other. Also good. In addition, not only “dish” itself, but also “tablet” and “dish”, which are synonyms for “dish”, are included in the same search expression together with “spoon”. It may be attached. According to such related word registration, it is also possible to create a suitable related table from a search formula created by an experienced searcher. For example, when a veteran searcher inputs “virtual reality” as a designated search word, the synonym “artificial reality” is often connected in an OR condition. In this case, even an inexperienced user can know that the term “artificial reality” exists as a similar concept of “virtual reality”, so that it is easy to create a search expression with few search omissions.

FIG. 5 is a conceptual diagram for explaining a method of adjusting the overall importance based on the inapplicable word.
The user may select not only similar documents but also dissimilar documents from candidate documents. The dissimilar document is a document file whose contents are dissimilar from the target document. Similar to similar documents, it is not always necessary to select a dissimilar document from candidate documents, and a predetermined dissimilar document may be input as it is. In addition, candidate documents that are not selected as similar documents among candidate documents may be treated as dissimilar documents. The inapplicable word extraction unit 146 extracts terms included in dissimilar documents as “inapplicable words”. The rank setting unit 154 calculates the inappropriateness of non-applicable words using the same algorithm as the method for calculating the importance of candidate terms.

  Candidate terms that are also non-applicable words are highly likely to be closely related to the contents of dissimilar documents, and thus are not necessarily appropriate for showing the characteristics of similar documents. The rank setting unit 154 excludes non-applicable words within the 20th inappropriate rank from the suggested search words among the non-applicable words specified from dissimilar documents. In the case of the same figure, the candidate term “similarity” having the third highest priority in the similar document is also excluded from the suggested search word because it is also an inapplicable fourth-ranked word in the dissimilar document.

  As a modification, when a candidate term is also ranked as an inapplicable word, a predetermined value, for example, 10 points may be deducted from the importance of the candidate term. In the case of the figure, the importance of “similarity” is adjusted from 81.2 to 71.2. In this way, the rank setting unit 154 adjusts the importance of candidate terms so that the importance ranks of candidate terms extracted as non-applicable words are lowered, or excludes candidate terms from the proposed search terms. By adjusting the importance ranking based on dissimilar documents, only candidate terms that appropriately indicate the characteristics of similar documents are easily listed as suggested search terms.

  The rank setting unit 154 may detect whether or not each candidate term within the 20th priority rank is not included in the inapplicable word group. For example, when the candidate term “corpus” ranked first in the priority rank appears in dissimilar documents, in other words, when it is included in the non-applicable word group, the candidate term “corpus” is selected as the suggested search term. Or the degree of importance of the candidate term “corpus” may be reduced. When the candidate term “corpus” is executed as the designated search term, not only similar documents but also dissimilar documents are detected. Therefore, the candidate term “corpus” determined to be prominent in expressing the characteristics of the similar documents. "May not necessarily be the best search term. From this point of view, adjustment may be made so that the rank of candidate terms extracted as non-applicable words is lower or out of ranking.

  Inapplicable words, in particular, inapplicable words positioned in the higher ranks of the inappropriate ranking may be specified as “excluded search words”. For example, the inapplicable word “execution time” is an effective term for expressing the characteristics of dissimilar documents, and the target document is likely to be a document file that does not include the inapplicable word “execution time”. Therefore, if a search expression such as “natural language * (− execution time)” (“(−): minus sign” means “exclusion”) is input, the term “natural language” is included, and , “Document file not including the term“ execution time ”” is a search target. By setting exclusion search terms, dissimilar documents can be easily removed from the search range.

  Further, when the inapplicable word “execution time” is also included in the candidate term group, if the inapplicable word “execution time” is an excluded search word, similar documents are also excluded. Therefore, the search word presenting unit 128 may present to the user only the inapplicable words that are not included in the candidate term group, in other words, the inapplicable words that are not included in the similar document as suitable “excluded search words”. .

FIG. 6 is a screen view of a search engine to which the document search method according to the present embodiment is applied.
In the above description, it is assumed that the published patent gazette is searched from the patent database. However, the document search method in this embodiment can be applied to general web search. The document search support apparatus 100 transmits a web page to the client terminal, and displays a search screen 200 on the client terminal. When a search expression is entered in the search expression input field 202 and the search button 204 is clicked, search results are displayed in a list in the result display area 208. In the figure, a character string “one click” is entered in the search expression input field 202. When the search button 204 is clicked, a list of URIs (Uniform Resource Identifiers) of web pages including the character string “one click” is displayed in the result display area 208. Up to this point, the user interface is the same as that of a general search engine.

A check box 206 is displayed beside each page. The user selects “A” for a page that is similar in content to the target document, “B” for a page that is somewhat similar, and “X” for a page that is not similar. Input to 206. Web pages in which “A” and “B” are set become similar documents, and candidate terms are extracted. The importance of each candidate term is calculated by weighting the individual importance of the candidate term extracted from the similar document of A with respect to the individual importance of the candidate term extracted from the similar document of B. Web pages with “X” set are dissimilar documents, and inapplicable words are extracted. Suppose that “authentication” is specified as the suggested search term with the highest priority. At this time, the search expression presenting unit 130 may display a search expression “one click * authentication” in the search expression input field 202.
When the search formula suggestion button 210 is clicked, a search formula proposal screen 212 for suggesting several search formulas is displayed corresponding to the input results such as “A”, “B”, “X”, and the like.

FIG. 7 is a screen diagram of the search formula proposal screen 212.
When the search formula suggestion button 210 is clicked, a search formula proposal screen 212 shown in the figure is displayed in a dialog box format. It is assumed that the proposed search word “authentication”, “patent”, and the excluded search word “fraud” are specified as a search result based on the designated search word “one click” and the result of setting similar documents and dissimilar documents. In order to simplify the explanation, the related search terms are not considered here. As shown in the figure, the search expression presentation unit 130 searches for various combinations of search expressions such as “one click * authentication”, “one click * (− fraud)”, and “one click * patent * (− fraud)”. Presented on the formula proposal screen 212. The user selects a search expression using a radio button next to each search expression, and clicks the re-search button 214. Then, the search execution unit 132 executes the web search again according to the selected search formula. According to such an aspect, after the user inputs “one click”, the search can be continued only with a mouse click.
Note that a ranking list of candidate terms and inapplicable words may be displayed instead of the search expression.

FIG. 8 is a flowchart showing the search process.
When a search expression is input, the search word acquisition unit 122 acquires a designated search word (S10). When a plurality of designated search terms are included in the search expression, the related table update unit 152 updates the related table (S12). Here, it is assumed that a search expression including only the designated search word “virtual reality” is input. The search execution unit 132 searches the corpus for a candidate document that matches the search expression, in other words, a candidate document that includes the character string “virtual reality”, and the candidate document acquisition unit 118 extracts the candidate document from the corpus (S14). . When the user completes the search (Y in S16), the search result is displayed on the screen, and the process ends (S32).

  If the search is not completed (N in S16), the similar document acquisition unit 114 receives selection of a similar document by the user and acquires a similar document from the candidate document group (S18). The candidate term extraction unit 144 extracts candidate terms from the similar document (S20). The dissimilar document acquisition unit 116 receives selection of dissimilar documents by the user and acquires dissimilar documents from the candidate document group (S22). The inapplicable word extraction unit 146 extracts inapplicable words from dissimilar documents (S24). The rank setting unit 154 calculates the individual importance of the candidate terms, and further calculates the (total) importance, and then determines the importance rank of the candidate terms with reference to the non-applicable word group (S26). The search word presentation unit 128 displays the suggested search word on the screen (S28). The search formula presenting unit 130 displays the search formula including the proposed search term on the screen (S30).

  For example, if the designated search term is “virtual reality”, the suggested search term with the first priority ranking is “spoon”, and the suggested search term with the second highest ranking ranking is “fork”, “(virtual reality + artificial reality + mixed reality + ...) * (spoon + spoon + saji + 匙 + ...) "or" (virtual reality + artificial reality + mixed reality + ...) * (spoon + spoon + saji + shark + ...) * (Fork + Fork + ...) "or" (Virtual Reality + Artificial Reality + Mixed Reality + ...) * (Spoon + Spoon + Saji + Spear + ... + Fork + Fork + ...) "Is generated. Not only the suggested search terms but also excluded search terms, or search expressions including the suggested search terms and the excluded search terms may be displayed on the screen.

  The user selects a designated search term from the proposed search terms while referring to the proposed search formula. At this time, an arbitrary related search word may be selected as the designated search word. A new search formula is created based on the selected suggested search term and related search terms. Alternatively, the search formula proposed by the document search support apparatus 100 may be employed as it is. Thereafter, the processes after S10 are repeated.

The document search support apparatus 100 has been described above based on the embodiment.
According to the document search support apparatus 100, an appropriate search word is selected as a suggested search word or a related search word through selection of a similar document or a dissimilar document without inputting a search expression that can sufficiently narrow down the number of candidate documents. Can be identified from The number of candidate documents can be gradually narrowed down while adding suggested search terms and related search terms to the search formula.

  It is also a promising proposal for identifying the target document by calculating the importance of the candidate term and the importance ranking from the appearance position of the candidate term in the similar document, the rarity in the corpus, the appearance frequency in the dissimilar document, etc. Search terms are easily identified. Furthermore, the occurrence of a search omission is prevented by setting the relationship between terms by using a relation table. According to the document search support apparatus 100, not only the search terms but also the related terms can be specified as the related search terms by the related table.

  The present invention has been described based on the embodiments. The embodiments are exemplifications, and it will be understood by those skilled in the art that various modifications can be made to combinations of the respective constituent elements and processing processes, and such modifications are within the scope of the present invention. .

It is a schematic diagram of the document search in a present Example. It is a functional block diagram of a document search support apparatus. It is a conceptual diagram which shows the process until it specifies a proposal search word. It is a data structure figure of a related table. It is a conceptual diagram for demonstrating the adjustment method of the comprehensive importance based on a non-application word. It is a screen figure of the search engine which applied the document search method in a present Example. It is a screen figure of a search type proposal screen. It is a flowchart which shows a search process.

Explanation of symbols

  DESCRIPTION OF SYMBOLS 100 Document search assistance apparatus, 110 IF part, 112 Document acquisition part, 114 Similar document acquisition part, 116 Dissimilar document acquisition part, 118 Candidate document acquisition part, 120 Search expression acquisition part, 122 Search term acquisition part, 126 Presentation part, 128 Search term presentation unit, 130 Search expression presentation unit, 132 Search execution unit, 140 Data processing unit, 142 Term extraction unit, 144 Candidate term extraction unit, 146 Inapplicable word extraction unit, 148 Related processing unit, 150 Related search word specifying unit 152 related table update unit, 154 rank setting unit, 160 data holding unit, 162 related table holding unit, 170 basic term field, 172 synonym field, 174 related word field, 200 search screen, 202 search expression input field, 204 search button , 208 Result display area, 212 Search formula Screen, 214 re-search button.

Claims (13)

A device for supporting retrieval of a document file,
A similar document acquisition unit that acquires a similar document selected by the user as a document file similar in content to a desired document file;
A candidate term extraction unit that extracts terms included in the similar document as candidate terms;
A ranking setting unit that ranks candidate terms extracted from similar documents so that a candidate term having a lower appearance frequency in a predetermined document file group has a higher rank;
A search word presenting unit that presents candidate terms positioned at a predetermined rank or higher as search words suitable for searching for the desired document file;
A document search support apparatus comprising: A designated search term acquisition unit that acquires a search term specified by the user as a designated search term;
A candidate document acquisition unit that acquires, as a candidate document, a document file including the designated search word from a predetermined document file group,
The document search support apparatus according to claim 1, wherein the similar document acquisition unit acquires a candidate document designated by a user among one or more candidate documents as a similar document.   The document search support apparatus according to claim 1, wherein the rank setting unit adjusts ranks of candidate terms based on one or both of an appearance position and an appearance frequency in a similar document.   4. The ranking setting unit according to any one of claims 1 to 3, wherein among the extracted candidate terms, candidate terms whose number of appearances in the similar document is equal to or greater than a predetermined threshold are targeted for ranking. Document retrieval support device described in 1. A dissimilar document obtaining unit that obtains a dissimilar document selected by the user as a dissimilar document file in content with the desired document file;
5. The search term presenting unit presents a candidate term that is not included in the dissimilar document among the candidate terms positioned at the predetermined order or more, to the user. Document search support device. A dissimilar document acquisition unit that acquires a dissimilar document selected by a user as a dissimilar document file in content with the desired document file;
A non-applicable word extraction unit that extracts terms included in the dissimilar document as non-applicable words; and
The rank setting unit also sets ranks of non-applicable words so that ranks of non-applicable words with a lower appearance frequency in a predetermined document file group have higher ranks, and ranks of candidate terms that are also non-applicable words of a predetermined rank or higher are lowered. 5. The document search support apparatus according to claim 1, wherein the ranking of candidate terms is adjusted. A dissimilar document acquisition unit that acquires a dissimilar document selected by a user as a dissimilar document file in content with the desired document file;
A non-applicable word extraction unit that extracts terms included in the dissimilar document as non-applicable words; and
The rank setting unit also sets ranks of non-applicable words extracted from dissimilar documents so that ranks of non-applicable words with a lower appearance frequency in a predetermined document file group are higher.
5. The search word presenting unit presents an inapplicable word of a predetermined rank or higher to a user as an excluded search word that is a search word that should not be included in the desired document file. Document retrieval support device described in 1.   8. The document search support device according to claim 7, wherein the search word presentation unit presents, as the excluded search words, non-applicable words that are not included in the similar document among the non-applicable words of the predetermined order or higher. . An association table holding unit that holds an association table that associates one term with another term;
A related search term identifying unit that identifies candidate terms positioned above the predetermined rank as suggested search terms, and identifies terms associated with the proposed search terms in the related table as related search terms,
9. The search word presenting unit presents the related search word to the user as a search word suitable for searching the desired document file in addition to the suggested search word. The document search support apparatus according to any one of the above.   The document search support apparatus according to claim 9, wherein the related table holding unit holds the related table as a table in which terms having synonym relations are registered in association with each other.   10. A related table update unit that associates the plurality of terms and registers them in the related table when a plurality of terms are included as search terms in a search expression designated by a user. 10. The document search support device according to 10.   For each of a plurality of proposed search terms, a search term set in which a suggested search term and a related search term are linked by an OR condition is generated, and a plurality of the search term sets are used as a search expression suitable for searching the desired document file. The document search support device according to claim 9, further comprising a search formula presenting unit that presents a search formula linked according to an AND condition to a user. A computer program for supporting retrieval of a document file,
Processing for acquiring a similar document selected by the user as a document file similar in content to a desired document file;
Processing for extracting terms included in the similar document as candidate terms;
A process for ranking candidate terms extracted from similar documents so that a candidate term having a lower appearance frequency in a predetermined document file group has a higher rank;
A process of presenting candidate terms positioned above a predetermined rank as a search term suitable for searching for the desired document file;
Search support program that causes a computer to execute

Images