JP6173846B2 - Document analyzer - Google Patents

Description

  Embodiments described herein relate generally to a document analysis apparatus.

  In recent years, it has become possible to collect, store, and use electronic documents in large quantities using a computer system, as computer performance and storage capacity have increased, and computer networks have become popular. . There is a possibility that valuable knowledge is buried in a large amount of documents. For example, customer information and survey information received daily by customers for products and services provided by companies, defect information related to product design and manufacturing, reputation information on the Internet, etc. It often contains valuable clues to know. Therefore, technologies such as automatic document classification, clustering, and text mining have been developed as technologies for classifying, analyzing, and utilizing such atypical text information written in natural language according to its semantic content. Has been.

  For the purpose of analyzing text information written in a natural language such as Japanese, conventional analysis methods based only on the notation and frequency of words described in the sentence are appropriate to reflect the semantic content of the sentence. There is a problem that analysis results cannot be obtained. Therefore, in recent years, an analysis method based on the relationship between a plurality of words described in a sentence, such as a dependency relationship, has been devised (see, for example, Patent Document 1, Patent Document 2, and Patent Document 3). In such a technique, for example, when analyzing defect information of a certain product, from a sentence “a crack has occurred in a tank”, a phrase indicating a part “tank”, a phrase indicating a symptom “crack”, and The process of extracting the relationship is performed.

  On the other hand, cross tabulation is one method for presenting the results of analyzing text information of a plurality of documents in a form that is easy for the user to grasp. This is to obtain a subset of documents corresponding to a combination of analysis items based on a set of documents corresponding to a plurality of analysis items of each analysis axis for two or more analysis axes. This is a method of displaying numbers in a matrix. By using the result of the cross tabulation, the user can grasp the entire contents of the document set and can examine the correlation between the analysis items in detail. For example, when analyzing defect information of a certain product, the user selects the “component” axis as one analysis axis and the “symptom” axis as the other analysis axis. As a result, the analysis items on the “parts” axis, such as “tank”, “pipe”, and “wiring”, and the analysis items on the “symptom” axis, such as “crack”, “drop off”, “interference”, etc. Can understand the overall relationship. Of these, for example, a document describing the relationship between the “tank” and “crack” can be examined in detail. In Patent Document 1, it is devised to display a characteristic concept having a dependency relationship using this cross tabulation.

  If it is possible to extract a phrase corresponding to each analysis axis and a relationship between the phrases from a large amount of sentences described in a natural language with respect to a plurality of analysis axes desired by the user, the above-mentioned Using various methods such as cross tabulation, various knowledge can be obtained from the document. However, the number of combinations of phrases described in a large number of documents and relations between phrases (such as dependency relations) is enormous, and some of them are not related to the analysis axis desired by the user. There are many. For example, if you want to analyze the defect information of a product using the "parts" analysis axis and the "symptoms" analysis axis, all the words and phrases corresponding to "parts" and "symptoms" are automatically and comprehensively It is difficult to extract, and there is a problem that many words and phrases that are not related to “parts” and “symptoms” are erroneously extracted. Therefore, as described in Patent Document 1 and Patent Document 2, a dictionary describing words and phrases corresponding to “parts” and “symptoms” is prepared in advance, and the relationship between words only for known words and phrases. It is also possible to extract. However, it takes a lot of work to create a dictionary, and there is a problem that it is not possible to cope with new words and phrases described in documents that are increasing daily. On the other hand, as described in Patent Document 1, there may be a method using a rule for extracting only a desired relationship, for example, a predetermined dependency relationship, but natural language atypical sentences are described in various expressions. In addition, there are various combinations of analysis axes used by users for analysis. For this reason, it is difficult to prepare necessary rules in advance, and there is a problem that extraction errors and omissions often occur due to insufficient rules. Further, as described in Patent Document 3, a method of presenting only important expressions based on the importance of extracted expressions (phrases and their relationships) is also conceivable. However, in Patent Document 3, this importance level is simply “number of occurrences of expression ÷ number of documents in a subset of target documents”. Therefore, depending on this method, for example, the relationship between “part” and “symptom” Can not be extracted correctly.

JP 2001-75966 A JP 2001-101199 A JP 2004-102397 A

  The problem to be solved by the present invention is to provide a document analysis apparatus capable of extracting a phrase corresponding to an analysis axis selected as a target to be used for cross tabulation by a user from a large amount of sentences described in a natural language. It is.

  A document analysis apparatus according to an aspect of the present invention includes a document storage unit, an analysis axis storage unit, an analysis item generation unit, and a cross tabulation unit. The document storage unit stores a plurality of document data. The analysis axis storage unit stores a plurality of analysis axes, analysis items of the analysis axis, and words / phrases corresponding to the analysis items. The analysis item generation unit receives the input of the first analysis axis and the second analysis axis, and receives from the analysis axis storage unit the first word set and the second word set corresponding to the analysis item of the first analysis axis. A phrase that is a phrase that co-occurs with a phrase included in the first phrase set in the document data stored in the document storage unit while reading out the second phrase set that is a phrase corresponding to the analysis item of the analysis axis Candidates are extracted, and from these word candidates, words or phrases whose frequency or expression co-occurs with the words included in the first word set in the document data are similar to the words included in the second word set than the predetermined criteria A candidate is selected, and a new analysis item of the second analysis axis using the selected word candidate is written in the analysis axis storage unit. The cross tabulation unit reads the analysis items for each of the plurality of analysis axes and the words corresponding to the analysis items from the analysis axis storage unit, and stores them in the document storage unit for each combination of analysis items read for the plurality of analysis axes. The number of document data including words / phrases corresponding to the analysis items constituting the combination is counted, and the counting result is displayed.

It is a block diagram which shows the structure of the document analyzer which concerns on embodiment of this invention. It is a figure which shows the example of the document data memorize | stored in a document memory | storage part. It is a figure which shows the example of the analysis axis data memorize | stored in an analysis axis memory | storage part. It is a figure which shows the example of the analysis item data memorize | stored in an analysis axis memory | storage part. It is a flowchart showing the flow of the whole process of a document analyzer. It is a flowchart showing the flow of the cross tabulation process performed by the cross tabulation part. It is a flowchart showing the flow of the analysis item production | generation process performed by the analysis item production | generation part. It is a flowchart showing the flow of the evaluation process of the phrase relationship performed by the phrase relationship evaluation part of an analysis item production | generation part. It is a figure which shows the example of the relationship of the phrase extracted by the phrase relationship extraction part of an analysis item production | generation part. It is a figure which shows the example of the analysis item produced | generated by the result of the cross tabulation displayed by the cross tabulation part, the analysis axis and analysis item operated by the analysis axis operation part, and the analysis item generation part. It is a figure which shows the example of the analysis axis | shaft and analysis item which are operated by the analysis axis | shaft operation part, and the result of the cross tabulation displayed by the cross tabulation part.

Hereinafter, embodiments of the present invention will be described with reference to the drawings.
FIG. 1 is a block diagram showing a configuration of a document analysis apparatus 100 according to an embodiment of the present invention. As shown in FIG. 1, the document analysis apparatus 100 includes a document storage unit 1, an analysis axis storage unit 2, a cross tabulation unit 3, an analysis axis operation unit 4, and an analysis item generation unit 5.

  The document storage unit 1 stores a plurality of document data. The document data indicates a document to be analyzed by the document analysis apparatus 100. The analysis axis storage unit 2 stores analysis axis data and analysis item data. The analysis axis data and the analysis item data respectively indicate an analysis axis and an analysis item created for analyzing the document. The analysis axis has one or more analysis items. The document storage unit 1 and the analysis axis storage unit 2 can be realized using storage means of a conventional computer such as a file system or a database device.

  The cross tabulation unit 3 cross tabulates a set of document data stored in the document storage unit 1 using a plurality of analysis axes indicated by the analysis axis data stored in the analysis axis storage unit 2, and the cross tabulation result Is presented to the user by a display as shown in FIG. In this presentation, a general-purpose device such as a browser that communicates via the Internet and through a display of a personal computer is often used. That is, the cross tabulation unit 3 reads out analysis items included in each of the plurality of analysis axes based on the analysis axis data and the analysis item data stored in the analysis axis storage unit 2. The cross tabulation unit 3 counts the number of document data classified into the analysis items at the same time for each combination of analysis items read out, and presents information representing the counting result to the user. The analysis axis operation unit 4 allows the user to perform operations such as creation, deletion, and movement on the analysis axis and analysis items indicated by the analysis axis data and analysis item data stored in the analysis axis storage unit 2 To. The user can perform these operations on a screen displaying the result of cross tabulation as shown in FIG.

  The analysis item generation unit 5 automatically generates an analysis item for a predetermined analysis axis indicated by the analysis axis data stored in the analysis axis storage unit 2. In this embodiment, the analysis item generation unit 5 includes a phrase extraction unit 51, a phrase relationship extraction unit 52, and a phrase relationship evaluation unit 53. The phrase extraction unit 51 extracts a phrase from a sentence described in a natural language. The phrase extraction unit 51 can be realized using a technique such as conventional morphological analysis. The phrase relation extraction unit 52 extracts a relation such as a dependency relation between a plurality of phrases extracted from a sentence described in a natural language. The phrase relationship extraction unit 52 can be realized using a technique such as conventional syntax analysis, for example. The phrase relationship evaluation unit 53 evaluates whether or not the phrase should be used as an analysis item based on the relationship between a plurality of phrases extracted from the document data. Based on the evaluation result, the analysis item generation unit 5 generates an analysis item corresponding to the phrase, and registers the analysis item data in the analysis axis storage unit 2.

  FIG. 2 is a diagram illustrating an example of document data stored in the document storage unit 1. The document data 200 shown in FIG. 2A includes a document number 201, a report date 202, a target product 203, and a body 204, and the document data 210 shown in FIG. 2B includes a document number 211, a report date 212, The target product 213 and the text 214 are included. The document data 220 shown in FIG. 2C includes a document number 221, a report date 222, a target product 223, and a body 224. The document data 230 shown in FIG. 2D includes a document number 231, a report date 232, A target product 233 and a text 234 are included. The document data 240 shown in FIG. 2E includes a document number 241, a report date 242, a target product 243, and a text 244. The document data 250 shown in FIG. 2F includes a document number 251, a report date 252, The target product 253 and the text 254 are included.

  Document numbers 201, 211, 221, 231, 241, and 251 are unique data for specifying document data. Texts 204, 214, 224, 234, 244, and 254 are examples of text data corresponding to the type of document. This text is described in a natural language such as Japanese, and is a main analysis target of the document analysis apparatus 100. Report date 202, 212, 222, 232, 242, 252 and target product 203, 213, 223, 233, 243, 253, document data 200, 210, 220, 230, 240, 250 represents product defect information This is attribute data possessed because it is document data. Report dates 202, 212, 222, 232, 242, and 252 are data indicating defect report dates, and target products 203, 213, 223, 233, 243, and 253 are data indicating target products for which defects have been reported. It is.

  FIG. 3 is a diagram illustrating an example of analysis axis data stored in the analysis axis storage unit 2. The analysis axis data 300 illustrated in FIG. 3A includes an analysis axis number 301 and a name 302, and the analysis axis data 310 illustrated in FIG. 3B includes an analysis axis number 311 and a name 312. The analysis axis numbers 301 and 311 are unique data for the document analysis apparatus 100 to specify analysis axis data. Names 302 and 312 are data indicating the names of analysis axes. The analysis axis data 300 shown in FIG. 3A is an example of analysis axis data for analyzing the contents of the document data set of defect information shown in FIG. 2 in terms of “parts”. The analysis axis data 310 shown in (2) is an example of analysis axis data for analysis from the viewpoint of “symptoms”. Hereinafter, for example, the analysis axis number set in the analysis axis data for analyzing the contents of the document data set from the viewpoint of the analysis axis a is described as “analysis axis number of the analysis axis a”.

  FIG. 4 is a diagram illustrating an example of analysis item data stored in the analysis axis storage unit 2. The analysis item data 400 shown in FIG. 4A includes an analysis item number 401, an analysis axis 402, a name 403, a phrase 404, and a document 405. The analysis item data 410 shown in FIG. 411, analysis axis 412, name 413, word / phrase 414, and document 415. The analysis item numbers 401 and 411 are unique data for the document analysis apparatus 100 to specify analysis item data.

  The analysis axes 402 and 412 are data for specifying the analysis axis to which the analysis item belongs, and are indicated by the analysis axis number of the analysis axis. Since the value of the analysis axis 402 is “p01”, the analysis item data 400 of FIG. 4A corresponds to the analysis item of the analysis axis indicated by the analysis axis data 300 of FIG. Similarly, the analysis item data 410 in FIG. 4B corresponds to the analysis item of the analysis axis indicated by the analysis axis data 310 in FIG. 3B because the analysis axis 412 is “p02”.

  Names 403 and 413 are data indicating names of analysis items. The phrases 404 and 414 are data indicating phrases for determining a document data set corresponding to an analysis item, that is, classified into an analysis item, in processing such as cross tabulation described later. In the example of the document data shown in FIG. 2, a set of document data that includes the word / phrase indicated by the words / phrases 404 and 414 in the text, that is, the “body” is a document data set corresponding to the analysis item. In the example of the analysis item data 400 shown in FIG. 4A, document numbers “d01”, “d02”, “d03”, and the like of document data including “tank” set in the word / phrase 404 in the text are included in the document 405. Is set. Note that, in the word / phrase 414 of the analysis item data 410 shown in FIG. 4B, a plurality of words “crack” and “crack” are set. Thus, a plurality of different notations representing the same meaning are used. The phrase may be set as the phrase of the analysis item. In the document 415, the document number of the document data including “crack” or “crack” set in the word / phrase 414 in the body is set.

  FIG. 5 is a flowchart showing the overall flow of processing executed by the document analysis apparatus 100. The document analysis apparatus 100 accepts user operations for the analysis axis and analysis items in steps S501 to S515, and ends the process when receiving a termination request in step S515. The process shown in the figure is mainly executed by the analysis axis operation unit 4, but the process of step S512 is executed by the cross tabulation unit 3, and the process of step S514 is executed by the analysis item generation unit 5. Further, in the processing of steps S502, S504, S506, S508, S510, and S514, the analysis axis operation unit 4 or the analysis item generation unit 5 executes creation, deletion, movement, etc. of the analysis axis or analysis item, and as a result The analysis axis data and the analysis item data (data shown in FIGS. 3 and 4) stored in the analysis axis storage unit 2 are changed.

  When the analysis axis operation unit 4 receives an input of an operation for creating a new analysis axis from the user (step S501—YES), the analysis axis operation unit 4 creates the analysis axis data of the new analysis axis p and writes it to the analysis axis storage unit 2 (step S502). ). For example, the analysis axis operation unit 4 sets the newly assigned analysis axis number and the name of the analysis axis p received from the user in the generated analysis axis data of the analysis axis p. The analysis axis operation unit 4 repeats the processing from step S501.

  When the analysis axis operation unit 4 receives an input of an operation for deleting the analysis axis from the user (step S501-NO, step S503-YES), the analysis axis data of the analysis axis p designated by the user and the analysis axis p of the analysis axis p are specified. The analysis item data of all analysis items is deleted from the analysis axis storage unit 2 (step S504). Specifically, when the analysis axis operation unit 4 specifies the analysis axis data of the analysis axis p and reads the analysis axis number, the analysis axis data of the specified analysis axis p is deleted from the analysis axis storage unit 2, All of the analysis item data in which the analysis axis number of the read analysis axis p is set is deleted from the analysis axis storage unit 2. The analysis axis operation unit 4 repeats the processing from step S501.

  When the analysis axis operation unit 4 receives an input of an operation for creating an analysis item from the user (step S501, step S503-NO, step S505-YES), the analysis axis operation unit 4 newly creates an analysis item c on the analysis axis p designated by the user. To do. Then, the analysis axis operation unit 4 sets a set of document data including the word t specified by the user as a document set corresponding to the analysis item c (step S506). Specifically, the analysis axis operation unit 4 newly creates the analysis item data of the analysis item c and writes it into the analysis axis storage unit 2, and newly assigns the analysis item number, the analysis axis number of the analysis axis p, Set. Further, the analysis axis operation unit 4 sets the name and the phrase t of the analysis item c input by the user and the document number of the document data including the phrase t in the body text in the analysis item data of the analysis item c. The analysis axis operation unit 4 repeats the processing from step S501.

  When the analysis axis operation unit 4 receives an input of an operation for deleting an analysis item from the user (step S501, step S503, step S505-NO, step S507-YES), the analysis axis data of the analysis item c designated by the user is obtained. It deletes from the analysis axis memory | storage part 2 (step S508). The analysis axis operation unit 4 repeats the processing from step S501.

  When the analysis axis operation unit 4 receives an input of an operation for moving the analysis item from the user (step S501, step S503, step S505, step S507-NO, step S509-YES), the analysis item c designated by the user is The original analysis axis p1 is moved to the analysis axis p2 designated by the user (step S510). Specifically, the analysis axis operation unit 4 rewrites the analysis axis number of the analysis axis p1 currently set in the analysis item data of the analysis item c to the analysis axis number of the analysis axis p2 that is the movement destination. The analysis axis operation unit 4 repeats the processing from step S501.

  When the analysis axis operation unit 4 receives a cross tabulation execution request input from the user (step S501, step S503, step S505, step S507, step S509-NO, step S511-YES), the cross tabulation unit 3 The cross tabulation is executed for the analysis axis p1 and the analysis axis p2 designated by (2), and the execution result is displayed (step S512). Details of this processing will be described later with reference to FIG. The analysis axis operation unit 4 repeats the processing from step S501.

  When the analysis axis operation unit 4 receives an input of an analysis item generation request from the user (step S501, step S503, step S505, step S507, step S509, step S511-NO, step S513-YES), the analysis item generation unit 5 generates an analysis item for the analysis axis p1 and the analysis axis p2 designated by the user (step S514). Details of this processing will be described later with reference to FIGS. The analysis axis operation unit 4 repeats the processing from step S501.

  The analysis axis operation unit 4 receives operations other than analysis user creation / deletion operations, analysis item creation / deletion / movement operations, cross tabulation execution requests, or analysis item generation requests from the user (step S501). Step S503, Step S505, Step S507, Step S509, Step S511, Step S513-NO). If the operation is other than an end request (step S515-NO), the analysis axis operation unit 4 repeats the process from step S501. If the operation is an end request (step S515-YES), the analysis axis operation unit 4 ends the process.

  FIG. 6 is a flowchart showing the flow of the cross tabulation process executed by the cross tabulation unit 3, and shows the detailed processing in step S512 of FIG. The analysis axes to be subjected to cross tabulation are the two analysis axes p1 and p2 specified by the user in step S512 in FIG.

  First, the cross tabulation unit 3 sets a set of analysis items on the analysis axis p1 as an analysis item set C1, and sets a set of analysis items on the analysis axis p2 as an analysis item set C2 (step S601). Specifically, the cross tabulation unit 3 identifies the analysis item data in which the analysis axis number of the analysis axis p1 received from the user is set, and the analysis item c1i (i represented by each identified analysis item data = 1, 2,...) Is an analysis item set C1. Further, the cross tabulation unit 3 identifies the analysis item data in which the analysis axis number of the analysis axis p2 received from the user is set, and the analysis item c2j (j = 1, 1) represented by each identified analysis item data 2,...) Is an analysis item set C2.

  The cross tabulation unit 3 repeatedly performs the processing from step S603 to step S607 for the selected analysis item c1i by sequentially increasing i by 1 from 1 until all the analysis items c1i in the analysis item set C1 are selected. Step S602—continuation is repeated).

  The cross tabulation unit 3 sets the word / phrase set in the analysis item data of the analysis item c1i as t1i (step S603). For example, in the case of the analysis item data 400 in FIG. 4A described above, the phrase “tank” set in the phrase 404 becomes the phrase t1i. The cross tabulation unit 3 repeatedly performs the processing from step S605 to step S607 on the selected analysis item c2j by sequentially incrementing j from 1 until it selects all the analysis items c2j in the analysis item set C2. Step S604-Repeatedly continue).

  The cross tabulation unit 3 sets the word / phrase set in the analysis item data of the analysis item c2j as t2j (step S605). The cross tabulation unit 3 obtains a document data set D (t1i, t2j) that includes both the phrase t1i and the phrase t2j (step S606). For example, when the word t1i is “tank” and the word t2j is “crack”, the cross tabulation unit 3 includes document data including both “tank” and “crack” in the text (for example, as shown in FIG. 2A). A set of document data 200) is set as a document data set D (t1i, t2j). The cross tabulation unit 3 sets the value of the i-th row and j-th column of the cross tabulation result as the number of documents | D (t1i, t2j) | that is the number of document data included in the document data set D (t1i, t2j). (Step S607).

  The cross tabulation unit 3 returns to the process of step S604, and if all the analysis items c2j in the analysis item set C2 are not selected, the current j value is incremented by 1 and the processes from step S650 to step S607 are performed. Repeat (step S604-repeat continuation). When all the analysis items c2j are selected, the cross tabulation unit 3 finishes the iterative process (step S604-repeat end), and returns to the process of step S602.

  When the cross tabulation unit 3 returns to the process of step S602, if all the analysis items c1i in the analysis item set C1 are not selected, 1 is added to the current value of i, and the process from step S603 to step S607 is performed. The process is repeated (step S602-repeat continuation). Then, when all the analysis items c1i in the analysis item set C1 are selected, the cross tabulation unit 3 ends the iterative process (step S602—end of repetition). The cross tabulation unit 3 presents the result of cross tabulation corresponding to the combination of the analysis items of the analysis item set C1 of the analysis axis p1 and the analysis item set C2 of the analysis axis p2 to the user by displaying it on the display (for example). Step S608). For example, the cross tabulation unit 3 sets i rows and j columns in a matrix in which the number of analysis items c1i included in the analysis item set C1 is the number of rows and the number of analysis items c2j included in the analysis item set C2 is the number of columns. Information representing | D (t1i, t2j) | is displayed. Further, the cross tabulation unit 3 displays the name set in the analysis item data of the analysis item c1i in the heading of the i row of this matrix, and is set in the analysis item data of the analysis item c2j in the heading of the j column. Display the name. In addition, the cross tabulation unit 3 displays the name set in the analysis axis data of the analysis axis p1 in the headline summarizing the analysis items of each row, and the analysis axis of the analysis axis p2 in the headline summarizing the analysis items of each column. Displays the name set in the data.

  In the process of step S606, a document data set including both of two words / phrases is simply obtained. However, as shown in the word / phrase 414 of the analysis item data 410 shown in FIG. 4 (b), it corresponds to one analysis item. There are cases where there are a plurality of words (for example, there are two cases of “crack” and “crack”). In this case, the cross tabulation unit 3 obtains a document data set including at least one of the plurality of words and including the word of the analysis item of the other analysis axis. For example, when the word t1i is “tank” and the word t2j is “crack” or “crack”, the cross tabulation unit 3 includes document data including both “tank” and “crack” in the text, and “ A set of document data including both “tank” and “crack” in the text is a document data set D (t1i, t2j). If both the phrase t1i and the phrase t2j are a plurality of phrases, the cross tabulation unit 3 generates all combinations of the phrase t1i and the phrase t2j, and both the generated phrase t1i and the phrase t2j are included in the text. A set of document data to be included is a document data set D (t1i, t2j).

  Further, the processing of step S606 is modified so that the cross tabulation unit 3 limits the document data set D (t1i, 2) so that the two words are limited to document data having a predetermined relationship (such as dependency relationship) described later. t2j) may be determined. In step S607, the i-th row and j-th column of the cross tabulation result are set as the number of document data in the document data set. However, this value is not limited to the number of document data, and may be, for example, a ratio (percentage) to the total document data set. As the display format on the screen, for example, a bubble chart may be used. For example, the cross tabulation unit 3 displays the result of cross tabulation in this way in the form shown in FIG. 10 and presents it to the user.

  FIG. 10 shows examples of cross tabulation results displayed by the cross tabulation unit 3, analysis axes and analysis items operated by the analysis axis operation unit 4, and analysis items generated by the analysis item generation unit 5. FIG. FIG. 10A shows the result of cross tabulation for the analysis axis “component” 1001 and the analysis axis “symptom” 1002. The analysis item of the analysis axis “component” 1001 is “tank” 1003, and the analysis items of the analysis axis “symptom” 1002 are “crack” 1004 and “dropout” 1005. The number of documents in the document data set corresponding to the two analysis items on different axes is expressed by the area of the bubble chart circle. For example, the number of documents in the document data set corresponding to the two analysis items of the analysis item “tank” 1003 of the analysis axis “component” 1001 and the analysis item “crack” 1004 of the analysis axis “symptom” 1002 is the circle of the bubble chart. It is represented by an area of 906.

  FIG. 7 is a flowchart showing the flow of analysis item generation processing executed by the analysis item generation unit 5, and shows detailed processing in step S514 of FIG. 5 described above. The analysis axes to be processed are the two analysis axes p1 and p2 specified by the user in step S514 of FIG. 5, and the analysis axis p1 is an axis having a predetermined analysis item specified by the user. The analysis axis p2 is an axis that is an analysis item generation target.

  When the analysis item generation unit 5 sets the analysis axis p1 and the analysis axis p2 instructed with respect to the analysis item indicated by the cross tabulation result displayed by the cross tabulation unit 3 by the user, the set of analysis items of the analysis axis p1 Is an analysis item set C1, and a set of analysis items on the analysis axis p2 is an analysis item set C2. Specifically, the analysis item generation unit 5 identifies analysis item data in which the analysis axis number of the analysis axis p1 is set, and a set of analysis items represented by each identified analysis item data is set as an analysis item set C1. . Further, the analysis item generation unit 5 specifies analysis item data in which the analysis axis number of the analysis axis p2 is set, and sets a set of analysis items represented by the specified analysis item data as an analysis item set C2.

  Further, the analysis item generation unit 5 sets a set of words and phrases corresponding to each analysis item in the analysis item set C1 as a phrase set T1, and sets a set of words and phrases corresponding to each analysis item in the analysis item set C2 as a phrase set T2. The phrase corresponding to the analysis item is a phrase set in the analysis item data of the analysis item. That is, the analysis item generation unit 5 sets a phrase set read from the analysis item data of each analysis item included in the analysis item set C1 as a phrase set T1, and uses the analysis item data of each analysis item included in the analysis item set C2. A set of read phrases is referred to as a phrase set T2.

  Further, as described in the flowchart of FIG. 5, the analysis item of the analysis axis may be deleted or moved by a user operation. Therefore, the analysis item generation unit 5 sets a set of words / phrases corresponding to each analysis item that has been deleted from the analysis item set C2 in step S503 or step S509 in FIG. 5 as a deleted word / phrase set T2rmv (step S701). That is, the analysis item generation unit 5 deletes a set of words / phrases set in the analysis item data of each analysis item deleted from the analysis axis p2 or moved from the analysis axis p2 to another analysis axis. And

  The analysis item generation unit 5 obtains a phrase candidate set T2new, which is a set of phrase candidates corresponding to the analysis item newly generated on the analysis axis p2 (step S702). The phrase t2new included in the phrase candidate set T2new is a phrase that is not included in the phrase set T1, the phrase set T2, or the deleted phrase set T2rmv among the phrases extracted by the phrase extraction unit 51 from the text of the document data. Subject to being. Further, when a document data set D (T1, t2new) is a document data set that includes any one of the phrases included in the phrase set T1 and the phrase t2new in the body, this document data set D (T1, t2new) Is also a condition for the phrase t2new. In other words, the phrase t2new is a phrase extracted from the document data set D (T1) including any phrase in the phrase set T1, that is, a phrase that co-occurs with any phrase in the phrase set T1 in the document data. is there.

  Subsequently, the analysis item generation unit 5 selects unselected words t2new one by one until all the words t2new in the word candidate set T2new are selected, and the processing from step S704 to step S706 below for the selected word t2new. Is repeatedly executed (step S703—continuation of repetition).

  First, the analysis item generation unit 5 obtains a score scr (T1, T2, t2new) based on the similarity between the phrase t2new and the phrase included in the phrase set T2 (step S704). The score calculation process will be described in detail with reference to FIG. 8 to be described later. The larger the score, the more appropriate the analysis item to be newly generated. If the score calculated in step S704 is less than the predetermined threshold (YES in step S705), the analysis item generation unit 5 determines that the analysis item corresponding to this word is not generated, and extracts the phrase t2new from the phrase candidate set T2new. Except (step S706), the processing from step S703 is repeated. On the other hand, if the score calculated in step S704 is equal to or greater than a predetermined threshold (step S705—NO), the analysis item generation unit 5 repeats the processing from step S703 as it is.

  When the analysis item generation unit 5 selects all the words t2new in the word candidate set T2new, the analysis item generation unit 5 finishes the repetition process (step S703—end of repetition). The analysis item generation unit 5 has a large score scr (T1, T2, t2new) obtained in step S704 among the words remaining in the word candidate set T2new without being deleted at the stage where the repetition processing in step S703 is completed. A predetermined number of phrases t2new are selected in order, and the unselected phrases t2new are removed from the phrase candidate set T2new (step S707).

  Then, the analysis item generation unit 5 selects the unselected words t2new one by one until the selection of all the words t2new in the word candidate set T2new, and repeatedly executes the processing of the following step S709 (step S708-). Repeat continuously). That is, the analysis item generation unit 5 generates an analysis item c2new on the analysis axis p2, and if the selected word / phrase t2new is a word / phrase corresponding to the analysis item c2new, document data including the word / phrase t2new is a document set corresponding to the analysis item c2new. (Step S709). Specifically, the analysis item generation unit 5 generates analysis item data of a new analysis item c2new and registers it in the analysis axis storage unit 2. The analysis item generation unit 5 adds to this analysis item data the new analysis item number, the analysis axis number of the analysis axis p2, the name and phrase indicating the phrase t2new, and the document number of the document data including the phrase t2new in the text. Set a set. When the analysis item generation unit 5 selects all the phrases t2new in the phrase candidate set T2new (step S708—end repeatedly), the analysis item generation process of FIG. 7 ends.

  FIG. 10B shows an example of the result of the analysis item generation process described in FIG. For example, when generating an analysis item for the analysis axis “component” 1001 in FIG. 10A, the document analysis apparatus 100 receives an input of an analysis item generation request, and the analysis axis p1 is the analysis axis “symptom”. ”1002, and an input indicating that the analysis axis p 2 is the analysis axis“ component ”1001 is received. This input may be performed, for example, by the user selecting the analysis axis “symptom” 1002 as the analysis axis p1 and selecting the analysis axis “component” 1001 as the analysis axis p2 in the display of FIG. . As a result, the analysis item generation unit 5 performs the analysis item generation processing of FIG. 7 to generate the analysis items “pipe”, “pedal”, and “weld” on the analysis axis “component” 1001, Register in the analysis axis storage unit 2. Thereafter, when the document analysis apparatus 100 receives a cross tabulation execution request from the user, the cross tabulation unit 3 performs a cross tabulation process shown in FIG. As shown in FIG. 10B, the cross tabulation unit 3 displays the cross tabulation result in which the analysis items “pipe” 1011, “pedal” 1012, and “weld” 1013 are added to the analysis axis “component” 1001. .

  Note that the document analysis apparatus 100 does not receive the input of the cross tabulation execution request from the user after the execution of the processing of step S514 in FIG. 5 (analysis item generation processing in FIG. 7), and performs step S512 (cross tabulation in FIG. 6). Processing) may be performed.

  FIG. 8 is a flowchart showing the flow of the phrase relationship evaluation process executed by the phrase relationship evaluation unit 53 of the analysis item generation unit 5. In step S704 of FIG. 7, the process for obtaining the score of the phrase t2new is performed. Detailed processing is shown. The processing in FIG. 8 is executed by the word / phrase relationship evaluation unit 53, but step S 803 and step S 807 are processing using the relationship between words, so the word / phrase relationship extraction unit 52 executes.

  First, the phrase relationship evaluation unit 53 obtains a score fscr (T1, T2, t2new) focusing on the frequency with which the phrase is included in the document data set (step S801). The score fscr (T1, T2, t2new) is based on the similarity between the appearance frequency of the word t2new and the appearance frequency of the word in the word set T2. In the present embodiment, the phrase relationship evaluation unit 53 uses the score fscr (T1, T2, t2new) as the three characteristics regarding the frequency as shown by the following expressions (1), (2), and (3). Obtained in a comprehensive manner. That is, the phrase relationship evaluation unit 53 determines the appearance frequency of the phrase t2 in the phrase set T2 that is a phrase corresponding to the existing analysis item of the analysis axis p2 and the phrase t2new that is a candidate for the newly generated analysis item. A value representing the noticed similarity is obtained by Expression (1), Expression (2), and Expression (3), respectively. Then, the phrase relationship evaluation unit 53 sets the value obtained by Expression (1), Expression (2), and Expression (3) as a score of the phrase t2new, as shown in Expression (4). However, the present embodiment is not limited to this method.

  A set of document data stored in the document storage unit 1 is referred to as a document data set D. First, in Expression (1), the ratio of | D (t) | and | D (T1, t) | is x (T1, t). | D (t) | is the number of document data including a certain word t in the text in the document data set D. | D (T1, t) | is any of the word sets T1 in the document data set D. This is the number of document data including both the phrase and the phrase t in the text. The phrase relationship evaluation unit 53 obtains a ratio x (T1, t2new) when the phrase t is the phrase t2new and a ratio x (T1, t2) when the phrase t is each phrase t2 in the phrase set T2. . The phrase relationship evaluation unit 53 calculates the similarity between the ratio x (T1, t2new) and the ratio x (T1, t2) as fsim1 (T1, t2, t2new) for each phrase t2 in the phrase set T2. As a result, the score of the word t2new is increased as the value of fsim1 (T1, t2, t2new) is increased.

  Next, in the expression (2), for a certain phrase t, the phrase t1 having the largest number of document data sets included in the text together with the phrase t is selected from the phrases in the phrase set T1. The ratio of | D (t1, t) | and | D (T1, t) | is y (T1, t). | D (t1, t) | is the number of document data including the selected word t1 and word t in the text in the document data set D. As described above, | D (T1, t) | is the number of document data in the text data set D that includes any word / phrase in the word / phrase set T1 and the word / phrase t in the text. The phrase relationship evaluation unit 53 obtains a ratio y (T1, t2new) when the phrase t is the phrase t2new and a ratio y (T1, t2) when the phrase t is each phrase t2 in the phrase set T2. . The phrase relationship evaluation unit 53 calculates the similarity between the ratio y (T1, t2new) and the ratio y (T1, t2) as fsim2 (T1, t2, t2new) for each phrase t2 in the phrase set T2. Accordingly, the score of the word t2new is increased as the value of fsim2 (T1, t2, t2new) is increased.

  Subsequently, in Expression (3), a vector v (T1, t) represents the degree of frequency that a certain word t and each word t1 in the word set T1 appear in the text. Each element of the vector v (T1, t) corresponds to the phrase t1, and the value w (t1, t) of the element is a ratio of | D (t1, t) | and | D (t) | is there. | D (t1, t) | is the number of document data including both the phrase t1 and the phrase t in the text in the document data set D, and | D (t) | is the document data as described above. This is the number of document data that includes a certain word t in the text in the set D. The phrase relationship evaluation unit 53 calculates a vector v (T1, t2new) when the phrase t is the phrase t2new and a vector v (T1, t2) when the phrase t is each phrase t2 in the phrase set T2. To do. The phrase relationship evaluation unit 53 calculates the cosine similarity fsim3 (T1, t2, t2new) between the vector v (T1, t2new) and the vector v (T1, t2) for each phrase t2 in the phrase set T2. Thus, the larger the value of fsim3 (T1, t2, t2new), the higher the score of the word t2new.

  In the expression (4), the phrase relationship evaluation unit 53 determines a predetermined positive constant α1, α2 for each of the three similarities calculated by the expressions (1) to (3) for each phrase t2 in the phrase set T2. , Α3 are respectively multiplied and calculated, and the maximum value among them is set as a score fscr (T1, T2, t2new). The large value of this score means that a phrase that is very similar to the phrase t2new exists as a phrase corresponding to the existing analysis item of the analysis axis p2 in terms of the frequency of appearance together with the phrase in the phrase set T1. means.

  Therefore, when the phrase relationship evaluation unit 53 determines that the score fscr (T1, T2, t2new) obtained in step S801 is less than a predetermined threshold (step S802-YES), the score scr (T1) of the phrase t2new , T2, t2new) is set to 0, and the process of FIG. 8 is terminated (step S809). By doing so, the phrase t2new is removed from the candidates in the above-described step S706 of FIG.

  When the phrase relationship evaluation unit 53 determines that the score fscr (T1, T2, t2new) is equal to or greater than a predetermined threshold (step S802-NO), the phrase relationship extraction unit 52 determines whether the phrase t2new and the phrase in the phrase set T2 A score escr (T1, T2, t2new) based on the similarity to is calculated (step S803). This score escr (T1, T2, t2new) is a score that focuses on expressions in which a phrase is described in a natural language sentence. The phrase relationship extraction unit 52 obtains the score escr (T1, T2, t2new) by a method in which two features relating to the expression of natural language are integrated by the following equations (5) and (6). That is, the phrase relationship extraction unit 52 focuses on the expression of the phrase t2 in T2 that is a phrase corresponding to the existing analysis item of the analysis axis p2, and the phrase t2new that is a candidate phrase of the analysis item to be newly generated. A value representing the similarity is obtained by each of these formulas (5) and (6). The phrase relationship extraction unit 52 obtains a value obtained by combining the values obtained by the expressions (5) and (6) according to the expressions (7) and (8) from the expressions (7) and (6), and the score escr (T1, T2new). T2, t2new). However, the present embodiment is not limited to this method.

  FIG. 9 is a diagram illustrating an example of a phrase relationship extracted by the phrase relationship extraction unit 52 of the analysis item generation unit 5. The figure shows an example of the result of the phrase relationship extraction unit 52 extracting the relationship between phrases from a sentence including both the phrase t1 in the phrase set T1 and the phrase t2 in the phrase set T2. In the present embodiment, the processing of the phrase relationship extraction unit 52 is realized by using a conventional syntax analysis technique. In this case, there are a plurality of clauses constituting the syntax tree and dependency relationships between the clauses. Extracted. Each phrase is composed of morphemes. The phrase relation extraction unit 52 selects the phrase including the phrase t1 or the phrase t2, the least relation connecting these phrases, and the phrase including the phrase t1 based on this relation from the dependency relation between the phrases and the phrase. And a partial syntax tree s (t1, t2) composed of other clauses connected to the clause including the word t2.

  FIG. 9A shows an example in which the phrase relationship extraction unit 52 extracts phrases and relationships from the text 204 of the document data 200 shown in FIG. When the phrase t1 is “crack” and the phrase t2 is “tank”, the phrase 902 includes the phrase t1, and the phrase 903 includes the phrase t2. Accordingly, the phrase relationship extraction unit 52 is a rectangle composed of the clause 902 and the clause 903, the relationships 905 and 906 that connect the clause 902 and the clause 903, and the clause 904 that is connected to the clause 902 and the clause 903 by the relationships 905 and 906. A partial syntax tree 901 of the part is obtained as a partial syntax tree s (t1, t2). On the other hand, the phrase relation extraction unit 52 similarly obtains a partial syntax tree s (t1, t2new) of a sentence including the phrase t2new and the phrase t1 for the candidate phrase t2new. FIG. 9B is an example in which clauses and relationships are extracted from the text 234 of the document data 230 shown in FIG. 2D. When the phrase t2new is “pipe”, the partial syntax tree 911 of the rectangular portion is The partial syntax tree s (t1, t2new) is obtained. The partial syntax tree 911 includes a clause 912 including the phrase t2new “pipe”, a clause 913 including the phrase t1 “crack”, relations 915 and 916 connecting the clause 912 and the phrase 913, It consists of a clause 914 connected to the clause 913.

  FIGS. 9C and 9D are examples in which the word t1 is “crack”, the word t2 is “tank”, and the word t2new is “weld”, as in FIGS. 9A and 9B. is there. The partial syntax tree 921 shown in FIG. 9C includes a phrase 922 including the phrase t2 “tank”, a phrase 923 including the phrase t1 “crack”, and relations 925 and 926 connecting the phrases 922 and 923. It consists of a clause 922 connected to a clause 922 and a clause 923 by 925 and 926, and becomes a partial syntax tree s (t1, t2). The partial syntax tree 931 shown in FIG. 9D includes a phrase 932 including the phrase t2new “weld”, a phrase 933 including the phrase t1 “crack”, and relations 935 and 936 that connect the phrase 932 and the phrase 933. 935 and 936 are composed of a clause 932 and a clause 934 connected to the clause 933, and becomes a partial syntax tree s (t1, t2new).

  When the phrase t2 and the phrase t2new are regarded as the same phrase and the clauses and relationships of the partial syntax tree s (t1, t2) and the partial syntax tree s (t1, t2new) are associated with each other, they can be associated with each other. Based on the number of morphemes that overlap each other or the number of relationships that can be associated with each other, a mathematical expression for calculating the similarity between the partial syntax tree s (t1, t2) and the partial syntax tree s (t1, t2new) Equations (5) and (6). In equation (5), count (mεs) indicates the number of morphemes m included in the partial syntax tree s. In equation (6), count (rεs) is a relationship included in the partial syntax tree s. The number of r is shown. From the document data, the phrase relation extraction unit 52 obtains zero or more partial syntax trees s (t1, t2) for each combination of each phrase t1 in the phrase set T1 and each phrase t2 in the phrase set T2. For each combination of each word / phrase t1 and word / phrase t2new in the set T1, zero or more partial syntax trees s (t1, t2new) are obtained.

  First, in Expression (5), the number mn (s (t1, t)) of morphemes m included in the partial syntax tree s (t1, t), two partial syntax trees s (t1, ti), and the partial syntax tree s. Consider the number mc (s (t1, ti), s (t1, tj)) of morphemes m included in (t1, tj). When counting the number mc (s (t1, ti), s (t1, tj)), the word ti and the morpheme corresponding to the word tj are regarded as being equal and counted. The phrase relationship extraction unit 52 calculates the mn (s (t1, t2)) of each partial syntax tree s (t1, t2) and the mn (s (t1, t2 new)) of each partial syntax tree s (t1, t2). Count. Furthermore, the phrase relationship extraction unit 52 uses mc (s (t1, t2), s () for all combinations of the partial syntax tree s (t1, t2) and the partial syntax tree s (t1, t2new) that share the phrase t1. t1, t2new)), and the similarity esim1 (s) between the word t2 and the word t2new is calculated according to the ratio of mn (s (t1, t2new)) and mn (s (t1, t2new)), whichever is larger. (T1, t2), s (t1, t2new)) are calculated.

  On the other hand, in Expression (6), the number rn (s (t1, t)) of the relation r included in the partial syntax tree s (t1, t), two partial syntax trees s (t1, ti), and the partial syntax tree. Consider the number rc (s (t1, ti), s (t1, tj)) of the relation r included in s (t1, tj). The phrase relationship extraction unit 52 calculates the rn (s (t1, t2)) of each partial syntax tree s (t1, t2) and the rn (s (t1, t2new)) of each partial syntax tree s (t1, t2). Count. Further, the phrase relationship extraction unit 52 determines rc (s (t1, t2), s for each combination of the partial syntax tree s (t1, t2) and the partial syntax tree s (t1, t2new) that share the phrase t1. (T1, t2new)) is counted, and the similarity esim2 () between the word t2 and the word t2new is determined by the ratio of rn (s (t1, t2new)) and rn (s (t1, t2new)), whichever is larger. s (t1, t2), s (t1, t2new)) are calculated.

  In the expression (7), the phrase relation extraction unit 52 uses the two expressions obtained by the expressions (5) and (6) for each of the partial syntax trees s (t1, t2new) including the phrase t1 and the phrase t2new. The maximum value obtained by multiplying the similarity by multiplying each of the predetermined positive constants β1 and β2 is selected for each partial syntax tree s (t1, t2) including the phrase t1 and the phrase t2. The phrase relationship extraction unit 52 divides the result of adding the selected maximum values by the number of partial syntax trees s (t1, t2new) | s (t1, t2new) |, and the phrase t2 and the phrase t2new Similarity esim (t1, t2, t2new).

  In Expression (8), the phrase relationship extraction unit 52 uses the maximum of the similarities esim (t1, t2, t2new) calculated in Expression (7) for all combinations of the phrase t1 in T1 and the phrase t2 in T2. The value is a score escr (T1, T2, t2new). A high score value means that a phrase that is very similar to the phrase t2new exists as a phrase corresponding to the existing analysis item of the analysis axis p2 in terms of an expression that appears with the phrase in the phrase set T1. means.

  For example, the morphemes of the partial syntax tree 901 in FIG. 9A are “tank” “ni” “crack” “ga” “occurred” “done” “.”, And the partial syntax tree 911 in FIG. The morphemes are “pipe” “to” “crack” “ga” “occurrence” “done” “.”. Therefore, in the partial syntax tree 901 in FIG. 9A and the partial syntax tree 911 in FIG. 9B, when the phrase t2 “tank” and the phrase t2new “pipe” are regarded as equal, all the morphemes and the relationships are related. Match. Accordingly, the similarity esim1 (s (t1, t2), s (t1, t2new)) = 7/7 = 1, the similarity esim2 (s (t1, t2), s (t1, t2new)) = 2/2 = 1 and the maximum similarity.

  Further, in the partial syntax tree 921 in FIG. 9C and the partial syntax tree 931 in FIG. 9D, when the phrase t2 “tank” and the phrase t2new “weld” are regarded as equal, six out of seven. Morphemes match (only the “bottom” and “location” do not match), and all the relationships match. Therefore, the similarity degree esim1 (s (t1, t2), s (t1, t2new)) = 6/7 and the similarity degree esim2 (s (t1, t2), s (t1, t2new)) = 2/2 = 1. In this case, the similarity is considerably large.

  In FIG. 8, when the phrase relationship evaluation unit 53 determines that the score escr (T1, T2, t2new) obtained in step S803 is less than a predetermined threshold (step S804—YES), the score of the phrase t2new Scr (T1, T2, t2new) is set to 0, and the process of FIG. 8 is terminated (step S809). By doing so, the phrase t2new is removed from the candidates in the above-described step S706 of FIG.

  When the phrase relationship evaluation unit 53 determines that the score escr (T1, T2, t2new) is equal to or greater than a predetermined threshold (step S804-NO), the word / phrase relationship evaluation unit 53 performs the processing from step S805 to S808, and step S701 in FIG. The phrase similar to the deleted phrase set T2rmv described in (2) is excluded. That is, the phrase relationship evaluation unit 53 calculates the score fscr (T1, T2rmv, t2new) for the deleted phrase set T2rmv, and the phrase relationship extraction unit 52 calculates the score escr (T1, T2rmv, t2new). The larger these values are, the more the word t2new is similar to the word corresponding to the previously deleted analysis item, so that the score scr (T1) of t2new is more easily removed from the analysis item candidates. , T2, t2new) is set to 0.

  Specifically, the word / phrase relationship evaluation unit 53 calculates the score fscr (T1, T2rmv, t2new) by the same process as in step S801 described above (step S805). If the phrase relationship evaluation unit 53 determines that the calculated score fscr (T1, T2rmv, t2new) is greater than or equal to a predetermined threshold (step S806—YES), the phrase relationship extraction unit 52 is the same as step S803 described above. A score escr (T1, T2rmv, t2new) is obtained by the processing (step S807). If the phrase relationship evaluation unit 53 determines that the calculated score escr (T1, T2rmv, t2new) is equal to or greater than a predetermined threshold (YES in step S808), the phrase relationship evaluation unit 53 sets the score scr (T1, T2, t2new) of the phrase t2new. 0 is set, and the process of FIG. 8 is terminated (step S809).

  The phrase relationship evaluation unit 53 determines that the score fscr (T1, T2rmv, t2new) calculated in step S805 is less than a predetermined threshold (step S806-NO), or the score escr (step S807 calculated). When it is determined that (T1, T2rmv, t2new) is less than the predetermined threshold (step S808-NO), it is determined that the word t2new is appropriate as a candidate for the analysis item. The phrase relationship evaluation unit 53 multiplies the scores fscr (T1, T2, t2new) calculated in step S801 and the score escr (T1, T2, t2new) calculated in step S803 by positive constants α and β, respectively. The combined value is calculated as the final score scr (T1, T2, t2new), and the process ends (step S810).

  The document analysis apparatus 100 performs cross tabulation while executing the processing described with reference to FIGS. 7 and 8 by alternately switching the target analysis axes, thereby performing each analysis as shown in FIG. Axis analysis items can be created in stages. For example, as described above, FIG. 10B illustrates an example in which the document analysis apparatus 100 generates an analysis item for the analysis axis “part” 1001 from the state illustrated in FIG. However, the automatically generated analysis items are not necessarily all appropriate particularly in the initial stage of the work. For example, the analysis axis “welding” 1013 in FIG. An analysis item may be generated by a phrase that does not correspond to “part” 1001. In such a case, the user can delete an inappropriate analysis item using the analysis axis operation unit 4 as described above. That is, the analysis axis operation unit 4 of the document analysis apparatus 100 deletes the analysis item “welding” 1013 designated as inappropriate by the user through the process of step S508 in FIG. Alternatively, the analysis axis operation unit 4 can also use the analysis item “welding” 1013 as an analysis item of another analysis axis by the processing in step S510 of FIG. An example is shown in FIG.

FIG. 11 is a diagram illustrating an example of a cross tabulation result displayed by the cross tabulation unit 3 after the analysis item is moved, an analysis axis and an analysis item operated by the analysis axis operation unit 4.
The user moves the analysis item “welding” 1101 inappropriate as the analysis axis “component” 1100 in the cross tabulation result shown in FIG. 11A to another analysis axis “process”. Accordingly, the cross tabulation unit 3 can display the cross tabulation result by using the analysis item “welding” 1112 for the analysis axis “process” 1111 as shown in FIG. The destination analysis axis may be specified in another cross tabulation result displayed by the cross tabulation unit 3, for example. When the analysis item is moved in this way, the analysis is performed on the subsequent analysis axis (“part” 1100 in the example of FIG. 11) by the processing described in steps S702 of FIG. 7 and steps S805 to S808 of FIG. In the process of generating the item, the moved phrase and the similar phrase are difficult to be extracted. On the other hand, in the process of generating an analysis item for the analysis axis of the movement destination (“process” in the example of FIG. 11), a phrase similar to the phrase is easily extracted.

  As described above, in the state of FIG. 10B, the user deletes or moves the analysis item “weld” 1013 from the analysis axis “component” 1001, and then selects the analysis item for the analysis axis “symptom” 1002. FIG. 10C shows the cross tabulation result after generation. In this example, words related to the symptom of the malfunction such as “broken” 1021 and “interference” 1022 are extracted on the analysis axis “symptom” 1002 and an analysis item corresponding to this is created.

  As described above, the analysis items can be generated by alternately switching the analysis axes in the form of FIG. 10A to FIG. 10B and FIG. 10C. For example, FIG. From the state, analysis items may be generated at once for both the analysis axis “part” 1001 and the analysis axis “symptom” 1002. In the example shown in FIG. 10D, “cable” 1031 and “wiring” 1032 are generated as analysis items corresponding to the analysis axis “component” 1001, and simultaneously as analysis items corresponding to the analysis axis “symptom” 1002. A “break” 1033 is generated. Thus, it is possible to easily generate analysis items for a plurality of analysis axes at once by adding a small modification to the processing flow of FIG. For example, after finishing the processing of FIG. 7 for the analysis axis p1 and the analysis axis p2 instructed by the user, the analysis item generation unit 5 switches the analysis axis p1 and the analysis axis p2 and performs the processing of FIG. 7 again.

  According to the above-described embodiment, in the first stage of the analysis work, the analysis items using known words and phrases are reduced manually by the user for each of the first analysis axis and the second analysis axis, for example. Create it. After that, the analysis item generation unit 5 automatically extracts a word corresponding to the second analysis axis to generate a new analysis item, and the user confirms the result using the cross tabulation unit 3. . Then, if necessary, this time, the phrase corresponding to the first analysis axis is automatically extracted to generate a new analysis item, and the user confirms the result again using the cross tabulation unit 3. The analysis work can be carried out in the following procedure. By repeating such operations, a plurality of analysis items corresponding to the first analysis axis and the second analysis axis are sequentially created. Using the cross tabulation results of these analysis items, the user can grasp the overall contents of the document data set, discover unknown words, that is, analysis items, and correlate these with other analysis items. Etc. can be examined in detail.

  In addition, when an analysis item that is inappropriate as an analysis item of the first analysis axis or the second analysis axis is automatically generated by the analysis item generation unit 5, the user immediately confirms this using the cross tabulation unit 3. can do. In addition, the analysis axis operation unit 4 can easily delete an inappropriate analysis item and use it as an analysis item of another analysis axis. In addition, when the analysis axis operation unit 4 deletes a word / phrase as inappropriate as an analysis item of a certain analysis axis, the analysis item generation unit 5 thereafter resembles the deleted word / phrase and the deleted word / phrase. The extracted phrase is not extracted as a phrase corresponding to the analysis axis. Therefore, by repeating such operations, appropriate words for each analysis axis can be extracted with high accuracy.

  According to the document analysis apparatus 100 of at least one embodiment described above, the analysis item generation unit 5 is provided, so that analysis items for each analysis axis used when analyzing a large amount of documents with a plurality of analysis axes are created. can do. The user does not have to manually search for a sentence corresponding to an analysis item by examining individual sentences described in a large amount of documents, and prepares a phrase corresponding to each analysis item in the form of a dictionary in advance. There is no need to specify in advance a word or phrase to be extracted in the form of a rule or the like, or a relationship between words. Therefore, the user's labor required for the analysis work is greatly reduced, and even an atypical sentence in which an unknown phrase or relational expression is described can be analyzed reflecting its semantic content.

  Further, according to the document analysis apparatus 100 of at least one embodiment described above, the analysis axis operation unit 4 is included, so that it is inappropriate as a target analysis axis in the process of generating an analysis item, but a document When an analysis item that is useful in analyzing the above is obtained, it can be used as an analysis item of another analysis axis. As a result, the user can also perform analysis using an analysis axis different from the analysis axis that was initially targeted by the user. Then, the analysis item generation unit 5 does not generate an analysis item using a phrase deleted from the analysis axis and a phrase similar to the deleted phrase on the analysis axis. Accordingly, it is possible to accurately generate analysis items with appropriate words and phrases for each analysis axis so that the analysis items are not generated with words that are highly likely to be inappropriate.

  Note that a program for realizing the function of the document analysis apparatus 100 in FIG. 1 in each of the above-described embodiments is recorded on a computer-readable recording medium, and the program recorded on the recording medium is read into a computer system. By executing this, the document analysis apparatus 100 may be operated. Here, the “computer system” includes an OS and hardware such as peripheral devices. The “computer system” includes a WWW system having a homepage providing environment (or display environment). The “computer-readable recording medium” refers to a storage device such as a flexible medium, a magneto-optical disk, a portable medium such as a ROM and a CD-ROM, and a hard disk incorporated in a computer system. Further, the “computer-readable recording medium” refers to a volatile memory (RAM) in a computer system that becomes a server or a client when a program is transmitted via a network such as the Internet or a communication line such as a telephone line. In addition, those holding programs for a certain period of time are also included.

  The program may be transmitted from a computer system storing the program in a storage device or the like to another computer system via a transmission medium or by a transmission wave in the transmission medium. Here, the “transmission medium” for transmitting the program refers to a medium having a function of transmitting information, such as a network (communication network) such as the Internet or a communication line (communication line) such as a telephone line. The program may be for realizing a part of the functions described above. Furthermore, what can implement | achieve the function mentioned above in combination with the program already recorded on the computer system, what is called a difference file (difference program) may be sufficient.

  As mentioned above, although embodiment of this invention was described, this embodiment is shown as an example and is not intending limiting the range of invention. This embodiment can be implemented in various other forms, and various omissions, replacements, and changes can be made without departing from the spirit of the invention. This embodiment and its modifications are included in the scope of the present invention and the gist thereof, and are also included in the invention described in the claims and the equivalent scope thereof.

DESCRIPTION OF SYMBOLS 1 ... Document storage part 2 ... Analysis axis storage part 3 ... Cross tabulation part 4 ... Analysis axis operation part 5 ... Analysis item generation part 51 ... Phrase extraction part 52 ... Phrase relation extraction part 53 ... Phrase relation evaluation part 100 ... Document analyzer

Claims (3)

A document storage unit for storing a plurality of document data;
An analysis axis storage unit that stores a plurality of analysis axes, analysis items of the analysis axis, and words corresponding to the analysis items;
The first analysis set and the second analysis axis are input, and the first set of words and phrases corresponding to the analysis item of the first analysis axis from the analysis axis storage unit and the second analysis set A second phrase set that is the phrase corresponding to the analysis item on the analysis axis is read out, and co-occurs with the phrase included in the first phrase set in the document data stored in the document storage unit A phrase that is included in the second phrase set, and a frequency or expression that co-occurs with the phrase included in the first phrase set in the document data from the word candidates. An analysis item generating unit that selects the word candidates that are more similar than a predetermined criterion, and writes a new analysis item of the second analysis axis using the selected word candidates to the analysis axis storage unit;
An analysis item for each of a plurality of analysis axes and a word corresponding to the analysis item are read from the analysis axis storage unit, and each combination of the analysis items read for the plurality of analysis axes is stored in the document storage unit. A cross tabulation unit that counts the number of the document data including words corresponding to the analysis items constituting the combination of the document data, and displays the counting result;
A document analysis apparatus comprising: Based on the input instruction to delete or move the analysis item, the analysis axis storage unit performs a process of deleting the analysis item or a process of rewriting the analysis axis to which the analysis item belongs to another analysis axis. An analysis axis operation unit,
The analysis item generation unit includes the phrase included in the deleted phrase set corresponding to the analysis item deleted or moved from the second analysis axis from the phrase candidate, and the first phrase set in the document data. Excluding the word candidate whose frequency or expression co-occurring with the included word is similar to the word included in the deleted word set than a predetermined criterion;
The document analysis apparatus according to claim 1, wherein: The analysis item generation unit is a first analysis axis having a predetermined analysis item selected from the analysis items for which the cross tabulation unit displays the count result, and a first analysis item to be generated. Set 2 analysis axes,
The document analysis apparatus according to claim 1, wherein the document analysis apparatus is a document analysis apparatus.

Images