JP5078960B2 - Text processing apparatus and computer program - Google Patents

Description

  The present invention relates to a technique for text processing the contents of text information having time-series information represented by opinions received from persons related to goods or services.

  For example, a company's help desk receives various opinions such as complaints and opinions from customers for its products and services. In addition, opinions on such products and services can be sent in various forms other than help desks through market research questionnaires, reputations on the Web, daily sales reports and daily service reports from salesmen and servicemen.

  In the past, companies have individually dealt with the above-mentioned views, accumulate the contents of these views as texts, and analyze the contents to detect problems of their products and services.

  In this case, the problem of the product or service is detected in advance by setting a keyword used for the existing problem, etc., and is set from the contents of the accumulated opinions by the filtering technique using the keyword. It was done by extracting already used keywords.

  For this reason, in the above-described conventional technology, although an already-existing problem or the like can be automatically detected, an unknown problem or the like cannot be automatically detected.

  However, in order to prevent serious troubles with their products and services, it is desirable for companies to discover not only the existing problems but also unknown problems as soon as possible and take appropriate measures. It was. For this reason, a system that automatically detects a sign that a problem or the like may occur in the future has been desired.

  Accordingly, an object of the present invention is to enable automatic detection of a sign such as a problem with a product or service from text information typified by views received from persons related to the product or service.

  The text processing device according to one embodiment of the present invention is a storage means for storing a plurality of data blocks including a date of acceptance of an opinion on the product or service by a person related to the product or service, and a text indicating the content of the opinion. And text analysis means for comparing the texts of the plurality of data blocks by text analysis and calculating the similarity between them, and a plurality of data based on the similarity calculated by the text analysis means Group generation means for grouping blocks to generate a new group, and for the new group generated by the group generation means, a time series in a predetermined extraction period on the basis of reception dates of a plurality of data blocks in the new group And a time series analysis means for performing analysis.

  In a preferred embodiment, there is further provided group extracting means for extracting the new group in which a time series change in the number of appearances based on the reception date of the opinion obtained by analysis by the time series analyzing means matches a predetermined extraction condition. You may prepare.

  In a preferred embodiment, the extraction condition is determined based on any of the number of data blocks in the new group, the rate of change in the number of appearances based on the reception date, or the variation pattern of the number of appearances based on the reception date. The conditions may be different.

  In a preferred embodiment, the extraction period may include a most recent first extraction period and a second extraction period that includes the first extraction period and is longer than the first extraction period.

  In a preferred embodiment, an expression satisfying a predetermined condition is stored in a condition storage unit that is stored for each predetermined group, and the data block read from the storage unit is expressed as an expression in the condition storage unit. You may further provide the filtering process part classified into the 1st data block containing and the 2nd data block which does not contain the expression in the said condition memory | storage part.

  In a preferred embodiment, the text analysis unit calculates the similarity for the second data block, the group generation unit groups the second data block, and the filtering processing unit includes: The first data block is associated with an identifier of an already-existing group corresponding to the expression included in each first data block, and the time series analysis unit includes a new group generated by a group generation unit, and A time series analysis may be performed on at least one of the previously described groups.

  In a preferred embodiment, the original data storage means for storing the original data of the opinion, and when the original data includes a plurality of sentences, the plurality of sentences are divided into one sentence, and the text includes only one sentence. And a data block generating means for generating a plurality of data blocks.

  In a preferred embodiment, an already-existing condition storage unit in which predetermined appearance expressions are stored for a plurality of appearance groups, an unnecessary condition storage unit in which predetermined unnecessary expressions are stored, and the storage unit The data block read from the first data block including the above-described expression, the second data block including the unnecessary expression, and the third data block including neither the above-mentioned expression nor the unnecessary expression And a filtering processing unit that further classifies the first data block by the already-existing group, and the text analysis unit calculates the similarity for the third data block. The group generation means generates the new group by grouping the third data blocks, and the time series analysis unit. For at least one of the existing group sorted by the filtering processing unit and the new group generated by the grouping generation means, at the time of a predetermined extraction period based on the reception time of a plurality of data blocks in the group A series analysis may be performed.

  By the above processing, the predictor candidates can be appropriately narrowed down, and it is possible to greatly reduce the burden of predictor finding work.

  In a preferred embodiment, the existing condition storage is performed by using an expression extracted from the text in the data block belonging to the new group generated by the group generating means as a new existing expression or a new unnecessary expression of the new existing group. Or registration means for registering in the unnecessary condition storage unit.

  The text processing apparatus according to the present invention can automatically detect a sign such as a problem with a product or service from text information typified by views received from a person related to the product or service.

It is a figure which shows the whole structure of the text processing apparatus which concerns on one Embodiment of this invention. It is a figure which shows an example of the data structure stored in an original data storage part. It is a figure which shows an example of the data structure stored in a data block memory | storage part. It is a conceptual diagram which shows an example of the existing expression extraction conditions stored in the existing condition storage part. It is a conceptual diagram which shows an example of the existing expression extraction conditions stored in an unnecessary condition memory | storage part. It is a figure which shows an example of the data structure after processed by the already-existing condition process part. It is a figure which shows an example of the data structure after processing by the unnecessary condition process part. It is a figure for demonstrating the method of grouping based on similarity. It is a figure which shows an example of a text analysis result table. It is a figure which shows an example of the data structure after grouping by the group production | generation means. It is a figure which shows an example of the histogram produced | generated by the time series analysis means. It is the figure which showed the example of the fluctuation pattern. It is a flowchart which shows the flow of a process of a time series analysis means and a group extraction means. It is a figure which shows an example of the condition setting screen in a time series analysis. It is a figure which shows an example of the detailed setting screen in a time series analysis and group extraction. It is a graph which shows the result of a time series analysis and group extraction. It is the figure which showed notion and the frequent appearance period of the problem of a time series base conceptually.

  Hereinafter, a text processing system that processes information including text input from a help desk or the like will be described as an example of a system that processes information including text according to an embodiment of the present invention with reference to the drawings. .

  FIG. 1 is a diagram showing an overall configuration of a text processing system according to the present embodiment.

  The system includes a text processing device 1, an input device 2, and an output device 3.

  Information including text is input from the input device 2. The text processing device 1 analyzes information including text input from the input device 2 and outputs the result of the analysis processing to the output device 3. Information including text may be directly input to the text processing device 1 from the other terminal device 2c, the portable terminal device 2b, or the like via the network 2a.

  The text processing device 1 is configured by, for example, a general-purpose computer system, and each component or function in the text processing device 1 described below is realized by executing a computer program, for example.

  For example, as shown in FIG. 1, the text processing device 1 includes an original data storage unit 4, a data block generation unit 5, a data block storage unit 6, a condition storage unit 7, a filtering processing unit 8, and similarity determination. Unit 9, grouping processing unit 10, time series analysis unit 11, and group extraction unit 12.

  Original data 40 is stored in the original data storage unit 4. The original data 40 is input from the input device 2 by, for example, an operator in charge, and is sequentially stored for each reception number 42. For example, the original data 40 input from the portable terminal device 2b or another terminal device 2c of a person related to the product or the like may be directly stored in the original data storage unit 4 via the network 2a.

  In the original data 40, an opinion on a product or the like by a person related to the product or service (hereinafter referred to as a product or the like) corresponding to the operator is stored. The person related to the product or the like is a person who is related to the product or the like in some form, such as a customer, a prospective customer, or a maintenance person. Further, the opinion on the product or the like is text information including a complaint, opinion or comment on the product or the like. In the following description, a case where an opinion received from a customer is processed will be described.

  FIG. 2 is a diagram showing an example of the data structure of the original data 40 stored in the original data storage unit 4.

  For example, in the original data 40 shown in FIG. 2, the date 41 when the customer's opinion was received, the reception number 42, the person in charge 43 as reception operator information, the customer name 44, the address 45, and the telephone number 46 as customer information. In addition, an e-mail address 47, a reception content 48 indicating the content of the customer's view of the product, and a response content 49 indicating the content of the operator's response to the customer's view are stored.

  In the received content 48, the customer's view on the product or the like is stored as text. The response content 49 is also stored as text.

  FIG. 3 is a diagram illustrating an example of the data structure of the data block group 60 stored in the data block storage unit 6.

  The data block storage unit 6 stores a plurality of data blocks 65 including a reception date 63 as an example of reception time information of views on products and the like by customers and a text 64 indicating the content of the views. The data block 65 is generated by the data block generation means 5 as follows.

  When the original data 40 includes a plurality of sentences, the data block generation unit 5 divides the plurality of sentences into one sentence and generates a plurality of data blocks 65 including only one sentence.

  The data block generator 5 reads the original data 40 from the original data storage unit 4 for each reception number 42. If the received content 48 of the read original data 40 includes a plurality of sentences, the data block generation means 5 detects, for example, a punctuation mark, divides the plurality of sentences one sentence at a time, A plurality of only texts 64 are created. The data block generation means 5 describes the reception date 63 and the reception number 61 in which the date 41 and the reception number 42 of the original data 40 are associated with the text 64 of only one sentence, respectively, and assigns a block ID 62 thereto. A plurality of data blocks 65 are generated.

  The data block 65 generated by the data block generation unit 5 is stored in the data block storage unit 6. For example, the data block generation unit 5 may appropriately generate the data block 65 from the original data 40 stored in the original data storage unit 4.

  In the condition storage unit 7, expressions satisfying a predetermined condition are accumulated for each predetermined group. The expression here is, for example, a character string.

  For example, as shown in FIG. 1, the condition storage unit 7 includes an already-existing condition storage unit 71 in which an existing expression extraction condition 71a is set as a predetermined condition, and an unnecessary condition storage unit in which an unnecessary expression extraction condition 72a is set as a predetermined condition. 72.

  FIG. 4 shows an example of a plurality of previously-explained expression extraction conditions 71 a stored in the already-existing condition storage unit 71.

  The existing expression extraction conditions 71a are grouped, for example, for each similar existing expression extraction condition 71a. An identifier 71b for identifying a group is associated with each of the previously described expression extraction conditions 71a. The identifier 71b uniquely identifies each of the existing expression extraction conditions 71a, and can also specify the group to which each of the existing expression extraction conditions 71a belongs and the existing expression extraction condition 71a. In the illustrated example, the identifier 71b is composed of “e” indicating the existing expression extraction condition 71a, a 2-digit number indicating the group, and a 4-digit number identifying the condition in the group. Yes. The already-existing condition storage unit 71 may have a tree structure as shown in FIG.

  The already-explained expression extraction condition 71a is an expression corresponding to a problem that has already been recognized from the customer's view of a product or the like sent in the past. At this time, an identifier corresponding to the previously described expression extraction condition 71a may be input together.

  In the existing condition storage unit 71, a group (new group) newly generated by the grouping processing unit 10 to be described later may be additionally registered. For example, the text processing apparatus 1 uses the expression extracted from the text in the data block belonging to the new group generated by the grouping processing unit 10 as a new appearance expression (existing expression extraction condition) of a new appearance group or a new expression. As an unnecessary expression (unnecessary expression extraction condition), means (not shown) for registering in the already-existing condition storage unit 71 or the unnecessary expression condition storage unit 72 may be provided.

  FIG. 5 shows an example of a plurality of unnecessary expression extraction conditions 72 a stored in the unnecessary condition storage unit 72.

  The unnecessary expression extraction conditions 72a are grouped for each similar unnecessary expression extraction condition 72a, for example. Each unnecessary expression extraction condition 72a is associated with an identifier 72b for identifying a group. The identifier 72b uniquely identifies each unnecessary expression extraction condition 72a, and can also identify the group to which each unnecessary expression extraction condition 72a belongs and the unnecessary expression extraction condition 72a. In the illustrated example, the identifier 72b is composed of “u” indicating the unnecessary expression extraction condition 72a, a 2-digit number indicating the group, and a 4-digit number identifying the condition in the group. Yes. The unnecessary condition storage unit 71 may have a tree structure as shown in FIG.

  The unnecessary expression extraction condition 72a is an expression that is used as a fixed phrase such as “Thank you for your help” or “Please respond”, and is an expression that is not directly related to a problem with a product or the like. . At this time, an identifier associated with the unnecessary expression extraction condition 72a may be input together.

  The unnecessary expression extraction condition 72a stored in the unnecessary condition storage unit 72 is written from the input device 2 as needed by an operator or the like. At this time, an identifier associated with the unnecessary expression extraction condition 72a may be input together.

  The filtering processing unit 8 includes the first data block including the expression in the condition storage unit 7 and the second data not including the expression in the condition storage unit 7 for the data block 65 read from the data block storage unit 6. Sort into blocks.

  The filtering processing unit 8 includes, for example, an existing condition processing unit 81 and an unnecessary condition processing unit 82 as shown in FIG.

  FIG. 6 is a diagram illustrating an example of the data block group 60 after the filtering process is performed by the already-existing condition processing unit 81.

  The already-existing condition processing unit 81 reads the data block group 60 of a predetermined extraction period from the data block storage unit 6, and refers to the already-existing condition storage unit 71, and the already-explained expression extraction condition 71 a is included in the text 64 in the data block 65. Determine whether it is. For example, referring to FIGS. 4 and 6, when the text expression extraction condition 71 a is included in the text, the identifier 71 b starting with “e” corresponding to the text expression extraction condition 71 a is used to identify the text 64. It is given to the group column 66 corresponding to the included data block 65. If the text 64 does not include the previously-explained expression extraction condition 71a, “other” is set in the group column 66 corresponding to the data block 65 including the text 64.

  After the filtering process is performed by the existing condition processing unit 81, the data block group 60 is delivered to the unnecessary condition processing unit 82. On the other hand, for example, in FIG. 6, among the data blocks 65 processed by the existing condition processing unit 81, the data block 65 to which the identifier 71 b starting with “e” is assigned to the group column 66 is stored in the existing data block storage unit 13. Is done.

  FIG. 7 is a diagram illustrating an example of the data block group 60 after being filtered by the unnecessary condition processing unit 82.

  The unnecessary condition processing unit 82 refers to the unnecessary condition storage unit 72 for the data block group 60 received from the already-existing condition processing unit 81, and determines whether or not the unnecessary expression extraction condition 72 a is included in the text 64 in the data block 65. Judging. The determination whether or not the unnecessary expression extraction condition 72a is included is made, for example, for the data block 65 in which “other” is set in the group column 66. For example, referring to FIGS. 5 and 7, if the text 64 includes an unnecessary expression extraction condition 72a, an identifier 72b starting from “u” corresponding to the unnecessary expression extraction condition 72a is displayed in the data block. It is set in the group column 66 corresponding to 65 by replacing it with “Other”. If the text 64 does not include the unnecessary expression extraction condition 72a, the group column 66 corresponding to the data block 65 including the text 64 remains “Other”.

  After the filtering process is performed by the unnecessary condition processing unit 82, the data block group 60 is delivered to the grouping processing unit 10 described later. On the other hand, for example, in FIG. 7, among the data blocks 65 processed by the unnecessary condition processing unit 82, the data block 65 to which the identifier 72 b starting with “u” is assigned to the group column 66 is recorded in the unnecessary data block storage unit 14. Billion.

  The similarity determination unit 9 determines the similarity between two texts. For example, the two target texts are each decomposed into morphemes, compared with each other, and their dependency is analyzed.

  For example, the grouping processing unit 10 extracts all the data blocks 65 in which “other” is set in the group column 66 from the data block group 60 received from the unnecessary condition processing unit 82. Then, the grouping processing unit 10 passes any two data blocks 65 of the extracted data blocks 65 to the similarity determination unit 9.

  The grouping processing unit 10 performs all matching between a plurality of target data blocks 65. That is, the similarity determination unit 9 is repeatedly requested for similarity determination for all combinations when two are extracted from the plurality of data blocks 65.

  Therefore, the similarity determination unit 9 and the grouping processing unit 10 compare the texts 64 of the plurality of data blocks 65 by text analysis, and calculate the similarity between each other.

  When performing all-matching, the grouping processing unit 10 stores the two data blocks 65 sent to the similarity determination unit 9 and the similarities returned from the similarity determination unit 9 in the similarity calculation table 100. . The similarity calculation table 100 is temporarily stored in the grouping processing unit 10.

  The degree of similarity is a comparison of the whole text 64 of a sentence including a word and a word change, and the text 64 included in one selected data block 65 and the text included in one data block 65 to be compared. This is a quantitative calculation of the degree of similarity to 64.

  FIG. 8 shows an example of the similarity calculation table 100.

  For example, as shown in FIG. 8, the similarity calculation table 100 includes, for each selected data block 65, the similarity 102 between the one data block 65 and all the data blocks 65 to be compared. It is remembered. In this case, the data is stored in the similarity calculation table 100 from the rank 101 “1” in order from the data block 65 having the highest similarity 102.

  In FIG. 8, the block ID 103 (hereinafter referred to as the reference ID) of the selected one data block 65 is “1114”. Regarding the similarity between the reference ID “1114” and the block ID 103 to be compared (hereinafter referred to as “target ID”), the similarity 102 with the target ID “1114” is the highest at 100%. The similarity 102 with the data block 65 having the target ID “5403” is the next highest at 85.7%.

  When all matching is completed, the grouping processing unit 10 groups the plurality of data blocks 65 based on the similarity 102.

  First, the grouping processing unit 10 generates a group by performing an analysis as illustrated in FIG. 9 based on the similarity calculation table 100, for example.

  The grouping processing unit 10 extracts the reference ID, the target ID, and their similarity 102 from the similarity calculation table 100, and sets a pair of the reference ID and the target ID for each reference ID as shown in FIG. They are arranged in descending order of similarity 102.

  Next, the grouping processing unit 10 groups a plurality of data blocks 65 based on the text analysis result table 90. That is, on the basis of the text analysis result table 90, those in which the similarity 102 between the reference ID and the target ID is equal to or greater than a predetermined value are set as the same group.

  The similarity 102 between the reference ID and the target ID is equal to or greater than a predetermined value. The similarity 102 between the data blocks 65 having the two block IDs is equal to or greater than a threshold value, and the data block 65 related to the two block IDs. As a result, the concept includes a case where the similarity 102 of three or more data blocks 65 is equal to or greater than a threshold value. The threshold value is set in advance as 85%, for example.

  In the text analysis result table 90, the grouping processing unit 10 identifies a pair having a similarity 102 equal to or higher than a threshold and a pair having a threshold less than the threshold among the pairs of the reference ID and the target ID. In the illustrated example, in order to identify a group having a similarity 102 that is equal to or greater than a threshold value, the group that is less than the threshold value is hatched. In the illustrated example, the text analysis results for the block IDs 1 to 7 are shown for simplicity.

  In the text analysis result table 90 of FIG. 9, there are two sets of reference IDs and target IDs whose similarity 102 is equal to or higher than a threshold value, “1-7” and “1-2” for the reference ID “1”. The reference ID “2” is one set of “2-5”, and the reference ID “3” is two sets of “3-6” and “3-9”. The grouping processing unit 10 identifies and groups the above sets. First, since the similarity of “1-7” and “1-2” is equal to or greater than the threshold value, the data blocks 65 having the block ID “1-2-7” are grouped. Next, since the similarity 102 of “2-5” is equal to or greater than the threshold value, the data blocks 65 having the block ID of “1-2-5-7” are grouped. In this case, for example, it does not matter whether the similarity of “2-7” or the similarity of “5-7” is greater than or equal to a threshold value. Further, since “3-6” and “3-9” are equal to or greater than the threshold value, the data blocks 65 having the block ID “3-6-9” are grouped. Also in this case, it does not matter whether the similarity of “6-9” is equal to or higher than the threshold value.

  FIG. 10 is a diagram illustrating an example of a data structure after grouping by the grouping processing unit 10.

  The grouping processing unit 10 assigns an identifier for identifying the group to the data block 65 of the group newly generated by grouping. For example, in FIG. 10, the identifier identifies “n” indicating that there is a data block 65 of a new group grouped by the grouping processing unit 10, a 2-digit number indicating the group, and a condition within the group 4 It consists of digits.

  The identifier assigned by the grouping processing unit 10 is set in the group column 66 of the data block group 60 in place of “Other”.

  Subsequently, for example, in the data block group 60 shown in FIG. 10, the grouping processing unit 10 stores, in the new data block storage unit 15, the data block 65 in which the identifier starting with “n” is assigned to the group column 66. To do.

  Each processing by the filtering processing unit 8, the grouping processing unit 10, and the similarity determination unit 9 described above may be performed by batch processing for a data block 65 of a preset period with the reception date 63 as a reference. Good.

  The time series analysis unit 11 performs time series analysis in a predetermined extraction period of the group with reference to the reception date 64 of the plurality of data blocks 65 in the grouped group. The extraction period when performing the time series analysis may include the latest first extraction period and the second extraction period including the first extraction period and longer than the first extraction period.

  For example, the time series analysis unit 11 generates a histogram for one group with the predetermined period based on the reception date 63 as the horizontal axis and the number of data blocks for the predetermined period based on the reception date 63 as the vertical axis. To do. Furthermore, the time-series analysis unit 11 calculates the fluctuation rate of the number of data blocks in a predetermined period with the reception date 63 as a reference based on the generated histogram, for example.

  FIG. 11 shows an example of a histogram generated by the time series analysis unit 11. In FIG. 11A, the number of appearances of data blocks increases with the passage of days. FIG. 11B returns to the original state even if the number of appearances of the data block temporarily increases.

  The variation rate is an increase rate from the minimum value to the maximum value or a decrease rate from the maximum value to the minimum value in the predetermined period. The increase rate and the decrease rate are values obtained by dividing the maximum value of the number of data blocks by the minimum value and expressed as a percentage, both of which are expressed as pluses. When the minimum value is 0, for the sake of convenience, the increase rate and the decrease rate are treated as a percentage value obtained by dividing the maximum value of the number of data blocks 65 by 1.

  Therefore, in FIG. 11A, the number of data blocks of “July 2008” is the maximum value, and the number of data blocks of “April 2008” is the minimum value. In this case, since the minimum value is 0, the increase rate is a value obtained by dividing the number of data blocks of “July 2008”, which is the maximum value, by 1 as a percentage.

  In the case of FIG. 11B, the number of data blocks of “June 2008” is the maximum, and the numbers of data blocks of “May 2008” and “July 2008” are the same. Minimum value.

  The time series analysis unit 11 may perform time series analysis on the generated new group and at least one of the existing groups corresponding to the expression included in each first data block by the grouping processing unit 10. Good.

  The group subjected to time series analysis by the time series analysis unit 11 is delivered to the group extraction unit 12 as shown in FIG.

  The group extraction unit 12 extracts a group in which the time series change of the number of appearances based on the reception date 64 of the opinion obtained by the time series analysis by the time series analysis unit 11 matches a predetermined extraction condition. The extraction condition may be determined based on any of the number of data blocks in the new group, the rate of change in the number of appearances based on the reception date of the opinion, or the variation pattern of the number of appearances based on the reception date.

  The group extraction unit 12 determines a variation pattern based on the analysis result of the time series analysis unit 11. For example, you may decide according to the fluctuation | variation of the number of data blocks when a histogram like FIG. 11 is produced | generated.

  FIG. 12 shows an example of the variation pattern.

  The time series pattern 182 of 1 is “monotonically increasing”. At this time, the increasing rate is calculated as the variation rate.

  The second time-series pattern 182 is “increase after decrease”. At this time, an increase rate is calculated as a variation rate.

  The time series pattern 182 of “3” is “irregularly fluctuating”. At this time, the fluctuation rate is “?”.

  The time series pattern 182 of “4” is “decrease after increase”. At this time, the decrease rate is calculated as the variation rate.

  The time series pattern 182 of “5” is “no change”. When the number of comment data blocks 63 on the vertical axis is 0, it is excluded.

  The time series pattern 182 of 6 is “monotonically decreasing”. At this time, the decrease rate is calculated as the variation rate.

  FIG. 13 is a flowchart showing a processing flow of the time series analysis unit 11 and the group extraction unit 12 of FIG. FIG. 14 shows a condition setting screen 16 for time series analysis displayed on the output device 3. FIG. 15 is a detailed setting screen 17 for time series analysis and group extraction displayed on the output device 3. FIG. 16 is a diagram illustrating an example of a result display screen 19 displayed on the output device 3.

  The process flow of the flowchart of FIG. 13 will be described with reference to FIGS.

  First, conditions for selecting an object to be analyzed are set (S1). At this time, the condition setting screen 16 of FIG. 14 is displayed on the input device 2. In the condition setting screen 16 of FIG. 14, a check box 161 for selecting a new data block stored in the new data block storage unit 15 and a check box for selecting an existing data block stored in the existing data block storage unit 13 are displayed. 162 and a check box 163 for selecting an unnecessary data block stored in the unnecessary data block storage unit 14, an execution button 164, and a cancel button are displayed.

  When performing the time series analysis, for example, the operator selects which storage unit 13 to 15 to perform the time series analysis from the condition setting screen 16 of the input device 2 and executes it. In FIG. 14, for example, a check is input to a check box 161 for selecting a new data block storage unit 15.

  When the new data block 161 is selected, the time series analysis unit 11 refers to the data block 65 in the new data block storage unit 15 and sets the time series for the group whose identifier in the group column 66 starts from “n”. Perform analysis.

  When the existing data block 162 is selected, the time series analysis unit 11 refers to the data block 65 in the existing data block storage unit 13 and sets the time series for the group whose identifier in the group column 66 starts from “e”. Perform analysis.

  When the unnecessary data block 163 is selected, the time series analysis unit 11 refers to the data block 65 in the unnecessary data block storage unit 14 and sets the time series for the group whose identifier in the group column 66 starts from “u”. Perform analysis.

  When the execution button 164 is pressed on the condition setting screen 16, the detailed setting screen 17 of FIG. 15 is displayed on the output device 3. The detailed setting screen 17 is a screen for instructing or selecting an extraction period for performing time series analysis and an extraction condition in the group extraction means.

  In the detailed setting screen 17 of FIG. 15, the most recent period and the entire analysis period can be selected. The detailed setting screen 17 in FIG. 15 includes an analysis period specification area 171, a category selection area 172, a latest specification area 173, a variation pattern selection area 174, a variation rate specification area 175, and a document number specification area 176. Is displayed. The operator designates or selects necessary items on the detailed setting screen 17 (S2).

  In the analysis period specification area 171, the start date and the end date of the period to be extracted can be specified. The analysis period is the longest period that is extracted and includes the most recent period.

  In the category selection area 172, a period based on the reception date 63, which is a horizontal axis in the time series analysis, can be selected. In this case, four periods of “year”, “month”, “week”, and “day” can be selected.

  In the latest designated area 173, it is possible to designate how many sections are output from the most recent period specified in the section selection area 172. For example, as shown in FIG. 15, when “week” is selected in the selection area 172 of the category and “3 data” is selected in the latest designated area 173, data for the last 3 weeks is output in units of weeks. Is done.

  In the variation pattern selection area 174, a variation pattern to be extracted can be designated from among the variation patterns of the group analyzed by the time series analysis. Variation patterns can be specified for the most recent period and the entire analysis period, respectively. In the variation pattern selection area 174, in addition to “no designation” shown in FIG. 15, as exemplified in FIG. 12, several types of variation patterns (time series pattern 182) may be selected.

  The variation rate designation area 175 can designate what percentage of the group with the variation rate to be output.

  In the document number designation area 176, for example, in FIG. 15, it is possible to designate the number of data blocks to be output in the entire analysis period from the group analyzed in time series.

  It should be noted that the rate of change designation area 175 and the number of documents designation area 176 can specify the latest data number, or the change rate and the number of documents for both the latest data number and the entire analysis period. Good.

  Returning to FIG. 13, the time-series analysis unit 11 extracts the data block 65 of the analysis period input to the designated area 171 from the storage units 13 to 15 checked in the check boxes 161 to 163 in FIG. 14 ( S3).

  For the data block 65 extracted in step S3, the time series analysis unit 11 performs time series analysis in the manner described above for each of the latest and the entire analysis period according to the category selected in the category selection area 172. (S4).

  Based on the time series analysis result by the time series analysis stage 11, the group extraction means 12 calculates the fluctuation rate (S5).

  The group extraction unit 12 extracts groups corresponding to the ranges of the change rate specified in the specified area 175 and the number of documents specified in the specified area 176 on the detailed setting screen 17 in FIG. 15 (S6).

  The group extraction unit 12 extracts a group corresponding to the variation pattern selected in the selection area 174 on the detailed setting screen 17 in FIG. 15 from the groups extracted in step S6 (S7). The variation pattern is searched for each of the entire analysis period and the most recent period selected in the area 174, and a pattern that matches either or both is extracted. Note that when “undesignated” is selected in the selection area 174, the extraction period is not extracted.

  The results of the time series analysis process and the group extraction process based on the conditions specified or selected on the condition setting screen 16 in FIG. 14 and the detailed setting screen 17 in FIG. 15 are displayed on the output device 3 (S8).

  FIG. 16 is a diagram illustrating an example of the result display screen 19.

  The result display screen 19 is output to the output device 3. For example, as shown in FIG. 16, the output number 191, the entire period 192, the latest period 193, the change rate 194, the number of documents 195, and the group 196 are output on the result display screen 19. Note that the overall period 192 shows the fluctuation pattern of the whole analysis period, and the latest period 193 shows the fluctuation pattern of the latest period. Further, the change rate 194 and the number of documents 195 indicate the change rate and the number of documents in the entire period 192.

  Clicking on the entire period 192 for a predetermined output number 191 displays details of the variation pattern of the entire analysis period. For example, when the analysis period 192 in the output number 191 “1” is clicked, a detailed result as shown in FIG. 11A is displayed. For example, when the latest period 193 is clicked in the output number 191 “4”, three data of the latest period 173 specified in FIG. 15 are output as shown in FIG.

  Although not shown, if the group 196 is clicked, all the data blocks 65 in the group are displayed. By displaying the individual data blocks 65, the data block storage unit 6 and the data satisfying each condition are displayed. You may enable it to refer to the blocks 13-15 suitably.

  As described above, the text processing apparatus 1 according to the present embodiment is based on the similarity calculated by the text analysis without performing the filtering process with a preset keyword or the like as in the prior art. Thus, a problem can be automatically detected by grouping a plurality of data blocks 65 and analyzing the data blocks 65 in the group in time series.

  In particular, when a problem is detected in a product or the like, as shown in FIG. 17, for example, as shown in FIG. There is a characteristic that there is a predictive period. The text processing apparatus 1 according to the present embodiment can automatically detect a point that will be a problem even in the sign period before the frequent appearance period by using this feature.

  Moreover, in this embodiment, the tendency of the time series change of the appearance number of views to output can be set in advance.

  The above-described embodiments of the present invention are examples for explaining the present invention, and are not intended to limit the scope of the present invention only to those embodiments. Those skilled in the art can implement the present invention in various other modes without departing from the gist of the present invention.

DESCRIPTION OF SYMBOLS 1 Text processing device 2 Input device 2a Network 2b Portable terminal device 2c Terminal device 3 Output device 4 Original data memory | storage part 5 Data block production | generation means 6 Data block memory | storage part 7 Condition memory | storage part 8 Filtering process part 9 Similarity determination part 10 Grouping Processing unit 11 Time series analysis unit 12 Group extraction unit 13 Existing data block storage unit 14 Unnecessary data block storage unit 15 New data block storage unit 63 Reception date 64 Text 65 Data block

Claims (5)

Storage means for storing a plurality of data blocks including text reception time and a text reception time represented by an opinion on the product or service by a person related to the product or service;
Text analysis means for comparing each text of the plurality of data blocks by text analysis and calculating the degree of similarity with each other;
Group generation means for grouping a plurality of data blocks to generate a new group based on the similarity calculated by the text analysis means;
For the new group generated by the group generation means , the first extraction period including the most recent first extraction period and the first extraction period on the basis of reception times of a plurality of data blocks in the new group, and the first A time series analysis means for performing a time series analysis in a predetermined extraction period including a second extraction period longer than the extraction period ;
A group extracting unit for extracting the new group obtained by the analysis by the time series analyzing unit, wherein the time series change in the number of appearances based on the reception time of the opinion matches a predetermined extraction condition;
With
The extraction condition is a condition determined based on one or more parameters including at least one of a variation rate of the number of appearances based on the reception time and a variation pattern of the number of appearances based on the reception time.
Text processing apparatus. The extraction conditions are:
(*) The number of data blocks in the new group in at least one of the most recent first extraction period and the second extraction period,
(*) The rate of change in the number of appearances based on the reception time in at least one of the most recent first extraction period and the second extraction period,
(*) Fluctuation pattern of the number of appearances based on the reception time in each of the latest first extraction period and the second extraction period
It is a condition determined based on
The text processing apparatus according to claim 1. A pre-determined pre-existing expression stored in a plurality of pre-existing groups;
An unnecessary condition storage unit that stores predetermined unnecessary expressions;
The data block read from the storage means includes a first data block including the above-described expression, a second data block including the above-mentioned unnecessary expression, and a third including neither the above-described expression nor the unnecessary expression. And a filtering processing unit for further classifying the first data block by the already-existing group,
The text analysis means calculates the similarity for the third data block,
The group generation means groups the third data block to generate the new group,
The time series analysis means, with respect to at least one group of the existing group sorted by the filtering processing unit and the new group generated by the grouping generation means, based on the reception time of a plurality of data blocks in the group, performing time series analysis in the predetermined extraction period, the text processing device according to claim 1 or 2. The expression extracted from the text in the data block belonging to the new group generated by the group generation means is used as the new existing expression or new unnecessary expression of the new existing group, and the existing condition storage unit or the unnecessary condition storage The text processing apparatus according to claim 3 , further comprising registration means for registering in the section. A computer program for text processing,
Storage means for storing a plurality of data blocks including text reception time and a text reception time represented by an opinion on the product or service by a person related to the product or service;
Text analysis means for comparing each text of the plurality of data blocks by text analysis and calculating the degree of similarity with each other;
Group generation means for grouping a plurality of data blocks to generate a new group based on the similarity calculated by the text analysis means;
For the new group generated by the group generation means , the first extraction period including the most recent first extraction period and the first extraction period on the basis of reception times of a plurality of data blocks in the new group, and the first A time series analysis means for performing a time series analysis in a predetermined extraction period including a second extraction period longer than the extraction period ;
A group extraction means for extracting the new group whose time series change in the number of appearances based on the reception time base of the opinion obtained by the analysis by the time series analysis means matches a predetermined extraction condition ; is realized,
The extraction condition is a condition determined based on one or more parameters including at least one of a variation rate of the number of appearances based on the reception time and a variation pattern of the number of appearances based on the reception time. <br/> Computer program.

Images