JP5404287B2 - Document analysis apparatus and method - Google Patents

Description

  The present invention relates to a document analysis apparatus and method, and more particularly to language records such as news, web news, blogs, newspapers, and magazines, as well as language records that increase in time series, such as interview records, statements, questionnaires, and novels. The present invention relates to a novel document analysis apparatus and method that can significantly analyze language materials that do not increase in time series.

  In disaster research, social surveys using questionnaires based on the mailing method, interview method, indwelling method, etc. are one of the means of investigating the actual situation of disaster-affected communities and the residents' awareness of risks and risk countermeasures. It is indispensable.

  In the questionnaire, there are question items in a pre-coded answer format including a face item for selecting one or a plurality of answers from options, and a question item in a free answer format in which a free description is entered. Of the latter, the question items to be filled in regarding the respondent's free opinions and ideas are customarily provided at the end of the questionnaire. Data obtained from such question items is called free answer or free answer description.

  In addition to social surveys on disasters, the analysis of free answer data obtained through questionnaire surveys is very difficult and has not been analyzed in many social surveys. Systematic analysis methods such as simple tabulation, cross tabulation, and multivariate analysis are prepared for the data obtained in the precoded answer format, but the free answer description obtained in the free answer format is difficult to tabulate Therefore, the analysis is often hindered, and the free answer descriptions themselves are often only enumerated when preparing reports and the like. For example, a survey of victims of the Miyakejima eruption disaster has provided free answers, but only a few cases are reported based on a broad classification.

  Free-answer questions can provide important informational material that cannot be investigated with pre-coded questions in the questionnaire. In particular, in the surveys of victims after a disaster, the contents of dissatisfaction and obstacles after the disaster are often described, and it is expected that concrete and detailed data on the actual situation of the disaster-affected area society will be obtained. Significantly, it is very important to analyze properly.

  As an example of analysis of free answer descriptions, in the field of disaster research, free answer descriptions that describe lessons and experiences of earthquake disasters in questionnaire surveys for households affected by the 1995 Hanshin-Awaji Earthquake were included in the seven elements of life reconstruction. Non-patent documents 1 and 2 are examples of classified examples and examples of classified and structured free answer descriptions that the victims asked about lessons they want to convey to people in other regions in similar surveys. ing. However, these are performed manually, and it can be imagined that a large amount of labor was required for analysis.

  On the other hand, several studies have been reported to develop an analysis method for free answer descriptions. These use natural language processing technology and text mining, and a method of extracting keywords using the frequency of words or the like (Non-Patent Document 3) and a method of automatically selecting important free answer descriptions (Non-Patent Document 3) Patent Literature 4), a method of automatically clustering free answer descriptions (Non-Patent Literature 5, etc.) and the like.

On the other hand, the inventors of the present invention proposed a new document analysis apparatus and method in Patent Document 1. In this background art, a linguistic material (corpus) relating to a disaster or crisis is regarded as a corpus that increases with time, and an index of an increased TFIDF with a modified TFIDF and a singular value is defined, and keywords are automatically extracted.
Haruo Hayashi (ed.): Survey on changes in residence after the earthquake and actual living conditions, Technical Report, Disaster Research Institute, Disaster Prevention Research Institute, Kyoto University, 1999 Kazuki Nakabayashi, Kunihiro Fukudome, Makiko Kawakami: Lessons from the victims of the Great Hanshin-Awaji Earthquake: From questionnaires and free answer analysis in Hyogo, Nagata, and Suma Wards. 9, pp146-149, 1999 Noboru Otsuki, Ludovic Lebart: Analysis of free answer data in the survey-Exploratory text data analysis by InforMiner-Statistical mathematics, Vol. 48, no. 2, pp 339-376, 2000 Masahiro Matsumura, Daisuke Kawahara, Masafumi Okamoto, Ikuo Kurohashi, Toyoaki Nishida: Extraction of “questions” lurking behind messages, Journal of Population Intelligence Society, Vol. 22, No.1, pp93-102, 2007 Inui Yuko, Tamura Maki, Uchimoto Kiyotaka, Isahara Hitoshi: Automatic classification based on intention of free answer questionnaire focusing on surface expression, natural language processing, Vol. 10, No2, pp14-102, 2007 WO 2008/062910 A1 [G06F 17/30]

  The background art of Patent Document 1 is intended to analyze a corpus that increases with the passage of time. On the other hand, free answer descriptions are data collected and formed at a certain point in time. Because it is not, it cannot be arranged in the order of time. Therefore, a corpus including a free answer description cannot be directly analyzed by the background art of Patent Document 1.

  Therefore, a main object of the present invention is to provide a novel document analysis apparatus and method.

  Another object of the present invention is to provide a document analysis apparatus and method capable of analyzing linguistic material that does not increase in time series such as free answer descriptions.

  Another object of the present invention is to provide a document analysis apparatus and method capable of analyzing a language material that does not increase in time series such as a free answer description based on the concept of singular values.

  The present invention employs the following configuration in order to solve the above problems. Note that reference numerals in parentheses, supplementary explanations, and the like indicate correspondence with embodiments to be described later in order to help understanding of the present invention, and do not limit the present invention.

  A first invention is a document analysis device that analyzes, based on an increase type TFIDF, a language material that is artificially increased in accordance with an order criterion, and an estimated value based on a cumulative value of the increase type TFIDF up to the previous corpus and a current corpus A singular value for each morpheme is obtained by performing a residual analysis with the cumulative value of incremental TFIDF. In document analysis apparatus, ascending cumulative singular value for each morpheme when linguistic materials are arranged in ascending order according to order criteria Ascending cumulative singular value calculation means for calculating decimated cumulative singular value for each morpheme, and ascending cumulative singular value and descending cumulative singular value. An average cumulative singular value calculation means for calculating an average cumulative singular value on average is provided.

  In the first invention, the document analysis apparatus is typically constituted by a computer. In the background art, linguistic material (corpus) whose unit document increases with the passage of time is targeted. However, in the present invention, a free answer format description that can be arranged according to an arbitrary order criterion such as age, occurrence date and time, for example. Target the corpus based. However, after performing morphological analysis on each document (corpus text), the method of extracting singular words by performing residual analysis in the current corpus on the regression curve calculated up to the previous corpus for significant morphemes can be applied as it is. .

For example, in morphological analysis, for example, text data in a language system in which morphemes are not divided, such as Japanese, for example, tea bowl (http://chasen.naist.jp/hiki/Cha
Using a morphological analysis tool such as Sen /), the corpus text data is decomposed into morphemes, and part-of-speech information is added to each morpheme. However, when the morpheme in the text is already divided, for example, in the case of a language system such as English, the work of dividing the morpheme (tokenization) is not necessary, and this morpheme analysis means uses a stemming process (stemming). ) Is used to add the quality information to each morpheme constituting the text by, for example, tagging, after the utilization form is changed to the original form.

  Based on the part-of-speech information added to each morpheme, the morpheme of the part-of-speech type set as an unnecessary morpheme is removed. That is, at the time of morpheme analysis, based on the part-of-speech information given to each morpheme, whether or not to adopt the morpheme as a singular word and / or a common word candidate is selected. However, the type of part of speech of unnecessary morphemes can be set arbitrarily. In the case of English sentences, unnecessary morphemes are removed by referring to a list of words that are inappropriately used as keywords, which are composed of articles and prepositions called stop words.

  For each morpheme remaining in the corpus, the TF (Term Frequency), that is, the frequency (total number) of the occurrence of the keyword candidate in the unit document, is calculated, and the IDF (Inversed Document Frequency) considering the time parameter (order standard) That is, by calculating a unique value that does not appear elsewhere, an increased TFIDF (Term Frequency Inversed Document Frequency) of the morpheme in the corpus is calculated as “TF” × “IDF”.

  In the residual analysis, for example, between the estimated value of the cumulative TFIDF of the corresponding morpheme estimated in the previous corpus and the actual value of the cumulative value of the increased TFIDF in the current corpus. A residual analysis is performed to obtain a residual value (singular value) of the morpheme, and a morpheme from which a positive singular value is obtained is selected as a singular word in the corpus.

  The cumulative singular value (ΣD) of each morpheme is calculated using the residual analysis method in the background art. Ascending cumulative singular value calculation means (S31-S35) calculates the corpus text in ascending order on the basis of order. The ascending cumulative singular value for each morpheme obtained when analyzing in parallel is calculated, and the descending cumulative singular value calculation means (S37-S41) is a descending cumulative singularity for each morpheme when the corpus texts are arranged in descending order on the order basis. Calculate the value.

  Then, the average cumulative singular value calculation means (S43) calculates the arithmetic average of the cumulative singular value ΣD (i, ord, asc) in ascending order and the cumulative singular value ΣD (i, ord, dsc) in descending order, that is, the average cumulative singular value. Calculate aveΣD. By this averaging process, it is possible to obtain a more objective (or representative) analysis result that excludes the characteristic of the order reference that is focused on. For example, when age is used as the order criterion, morphemes (words) with a large ascending cumulative singular value are characteristic morphemes for older people, while morphemes (words) with a large descending cumulative singular value are characteristic for younger people. It is a morpheme. Therefore, the average cumulative singular value indicates the degree to which the morpheme (keyword) is prominent in, for example, both young and old people, that is, how representative the morpheme is in the entire corpus. Therefore, the average cumulative singular value can also be referred to as a representative keyword value, and this representative keyword value aveΣD is one index for evaluation of a free answer description (unit document).

  A second invention is dependent on the first invention, and comprises a cumulative singular value graph display means for displaying a cumulative singular value graph in which one of an ascending cumulative singular value and a descending cumulative singular value is one of two axes and the other is the other. A document analysis apparatus further provided.

  In the second invention, the cumulative singular value graph display means (S45) displays, for example, a cumulative singular value graph (FIG. 21) with the ascending cumulative singular value on the vertical axis and the descending cumulative singular value on the horizontal axis. If you look at this graph, keywords that are characteristic of the order criteria, for example, articles of the old age of incidents, are located in places where the value in the horizontal axis is large, and keywords that are characteristic in the new era are displayed in the vertical direction. It can be easily found where the value is large.

  3rd invention is dependent on 1st or 2nd invention, the ascending order accumulation singular value sum total calculation means which calculates the sum total of ascending order accumulation singular value about the morpheme which has the accumulation singular value of a specific high order, the accumulation singular value of a specific high order Descending cumulative singular value summation means for calculating the sum of descending cumulative singular values for morphemes, and average cumulative singular value summation that averages ascending cumulative singular value sums and descending cumulative singular value sums The document analysis apparatus further includes a calculation unit.

  The above ascending / descending order cumulative singular value ΣD is an index indicating how important the word (morpheme) is in ascending / descending order of the order reference. Therefore, the idea that an article having a larger cumulative singular value ΣD and containing more morphemes (words) having a higher weight is an important free description (article) holds. In the third aspect of the invention, ΣΣD (cumulative singular value sum) indicating how important morphemes are included in one corpus data is employed. However, if ΣΣD is adopted as it is, there is a problem that the numerical value is affected by the length of the description sentence (number of words), and the cumulative singular value ΣD is obtained by ascending the order of each document included in the corpus. Since it is affected by whether it is arranged in descending order or in descending order, the accumulated singular value summation ΣΣD, which is the sum total in each document (article) of the accumulated singular value ΣD of each morpheme, is also affected in ascending / descending order. In order to eliminate the influence of the corpus, the third invention adopts the specific higher order ascending cumulative singular value sum ΣΣD (j, ord, asc, rank) and the descending order cumulative singular value sum ΣΣD (j, ord, dsc, rank) After that, the average cumulative singular value sum calculating means (S63) calculates the arithmetic mean of the ascending cumulative singular value sum and the descending cumulative singular value sum to obtain the average cumulative singular value sum aveΣΣD (j, ord, rank).

  The ascending cumulative singular value ΣD (j, ord, asc, rank) according to the order criterion and the descending cumulative singular value ΣD (j, ord, dsc, rank) according to the order criterion have a higher morpheme (word) with a more specific meaning. Indicates the value. Therefore, if these words and their weights are adopted as they are, the order is as follows: ascending cumulative singular value sum ΣΣD (j, ord, asc, rank) and descending cumulative singular value sum ΣΣD (j, ord, dsc, rank). Strongly shows the characteristics of either one of the standards. On the other hand, a homogeneous representative value that does not require the effect of order criteria is also required. Therefore, in order to offset the effect of the order reference, by taking the average value of both indicators, the unit document (free answer description) is adopted as an indicator indicating how representative the article is.

  A fourth invention is a document analysis apparatus according to the third invention, further comprising cumulative singular value sum difference calculation means for calculating a difference between an ascending cumulative singular value sum and a descending cumulative singular value sum.

  In the fourth invention, the cumulative singular value total difference calculating means (S65) calculates the difference between the ascending cumulative singular value sum and the descending cumulative singular value sum. The free answer description showing the high values of the ascending cumulative singular value sum ΣΣD (j, ord, asc, rank) and the descending cumulative singular value sum ΣΣD (j, ord, dsc, rank) However, there is a possibility that there is a free answer description that has both characteristics, and the effect of the order criterion may not be expressed appropriately. Accordingly, in the fourth invention, the difference between the ascending order cumulative singular value sum and the descending order cumulative singular value sum is obtained, and weighting reflecting the ascending / descending order property of the order reference is made possible by the absolute value. This is called the difference accumulated singular value sum diffΣΣD (j, ord, rak), so that the effect of the order criterion can be more emphasized.

  A fifth invention is dependent on the fourth invention, and displays a cumulative singular value summation graph displaying a cumulative singular value summation graph in which one of the average cumulative singular value summation and the cumulative singular value summation difference is one of the two axes and the other is the other. The document analysis apparatus further includes a graph display unit.

  In the fifth invention, the cumulative singular value sum graph display means (S67) displays, for example, the cumulative singular value sum graph shown in FIG. The entire corpus is represented by displaying a graph showing the average cumulative singular value sum aveΣΣD (j, ord, rank) on one of the two axes and the difference cumulative singular value sum diffΣΣD (j, ord, rank) on the other axis. It is possible to easily find a representative article or example (with a large average cumulative singular value sum aveΣΣD). Further, the feature in the order reference in which articles in the corpus are arranged can be grasped more clearly by the magnitude of the difference accumulated singular value sum diffΣΣD (j, ord, rank).

According to a sixth aspect of the present invention, there is provided a document analysis apparatus that analyzes a language material that increases in accordance with an order criterion based on an increased TFIDF, and an estimated value based on a total value of the increased TFIDF up to the previous corpus and a total sum of the increased TFIDF in the current corpus a Rubun manual analysis method obtains the singular values of each morpheme by performing a residual analysis between the values, the computer of the document analysis device, each morpheme when arranged corpora in ascending in order criteria An ascending cumulative singular value calculation step for calculating the ascending cumulative singular value of the grammar, a descending cumulative singular value calculation step for calculating the descending cumulative singular value for each morpheme when the language materials are arranged in descending order on the order basis, and an ascending cumulative singular value The document analysis method executes an average cumulative singular value calculation step of calculating an average cumulative singular value by arithmetically averaging descending cumulative singular values.

  In the sixth invention, the same effect as in the first invention can be expected.

  A seventh invention is a document analysis device for analyzing a language material that increases in accordance with an order criterion based on an increased TFIDF, and an estimated value based on the sum of the increased TFIDF up to the previous corpus and an increased TFIDF in the current corpus A singular value for each morpheme is obtained by performing a residual analysis with the sum of the values. Ascending cumulative singular value calculation step to calculate, descending cumulative singular value calculation step to calculate descending cumulative singular value for each morpheme when linguistic materials are arranged in descending order in order criteria, and ascending cumulative singular value and descending cumulative singular value A document analysis program characterized by executing an average cumulative singular value calculation step of calculating an average average cumulative singular value by averaging. .

  In the seventh invention, the same effect as in the first invention can be expected.

  An eighth invention is a document analysis device that analyzes a language material that increases in accordance with an order criterion based on an increased TFIDF, and an estimated value based on a cumulative value of the increased TFIDF up to the previous corpus and an increased TFIDF in the current corpus Ascending order to calculate the ascending cumulative singular value for each morpheme when the linguistic materials are arranged in ascending order on the order basis in the document analysis device that obtains the singular value for each morpheme by performing residual analysis with the cumulative value of Cumulative singular value calculation means, descending cumulative singular value calculation means for calculating descending cumulative singular values for each morpheme when linguistic materials are arranged in descending order in order criteria, ascending cumulative singular values for morphemes with specific higher cumulative singular values Ascending order cumulative singular value summation calculation means for calculating the summation, descending order accumulation for calculating the summation of the descending order singular values for the morphemes with the cumulative singular value at the top A document analysis apparatus comprising: an abnormal sum total calculating means, and an average cumulative singular value sum calculating means for calculating an average cumulative singular value sum by averaging an ascending cumulative singular value sum and a descending cumulative singular value sum .

  In the eighth invention, the same effect as in the third invention can be expected.

A ninth aspect of the document analysis device for analyzing based corpus that increasing the order criteria increasing type TFIDF, cumulative increasing type TFIDF in the estimate and the current corpus based on the cumulative value of the increase-type TFIDF and before Corpus A document analysis method for obtaining a singular value for each morpheme by performing a residual analysis with respect to values , wherein the document analysis device computer arranges the language materials in ascending order according to the order reference, and the ascending order for each morpheme Ascending cumulative singular value calculation step for calculating cumulative singular value, descending cumulative singular value calculation step for calculating descending cumulative singular value for each morpheme when linguistic materials are arranged in descending order according to order criteria, and having a specific higher cumulative singular value Ascending cumulative singular value summation calculation step that calculates the sum of ascending cumulative singular values for morphemes. Te descending cumulative singular value sum calculating step of calculating the sum of descending cumulative singular value, and the mean cumulative singular value sum calculating step of ascending cumulative singular value sum and average the descending cumulative singular value sum calculating a mean cumulative singular sum This is a document analysis method to be executed .

  In the ninth invention, the same effect as in the third invention can be expected.

  A tenth aspect of the invention is a document analysis apparatus that analyzes a language material that increases in accordance with an order criterion based on an increased TFIDF, and an estimated value based on the sum of the increased TFIDF up to the previous corpus and an increased TFIDF in the current corpus Calculates the ascending cumulative singular value for each morpheme when the linguistic materials are arranged in ascending order according to the order criteria on the computer of the document analyzer that calculates the singular value for each morpheme by performing residual analysis with the sum of Ascending cumulative singular value calculation step, descending cumulative singular value calculation step for calculating descending cumulative singular value for each morpheme when linguistic materials are arranged in descending order in order criteria, ascending cumulative singularity for morphemes with specific upper cumulative singular values Ascending cumulative singular value summation calculation step to calculate the sum of the values, the sum of the descending cumulative singular values for the morphemes with the cumulative singular value at the top Descending cumulative singular value summation calculating step, and ascending cumulative singular value summation and descending cumulative singular value summation are averaged to calculate the average cumulative singular value summation step. Document analysis program.

  In the tenth invention, the same effect as in the third invention can be expected.

  According to the present invention, by creating a corpus by arranging free answer descriptions (unit documents) according to an appropriate order criterion, it is possible to analyze a free answer description and select a representative keyword or the like.

  The above object, other objects, features, and advantages of the present invention will become more apparent from the following detailed description of embodiments with reference to the drawings.

FIG. 1 is a block diagram showing a document analysis apparatus as a background of the present invention. FIG. 2 is an illustrative view showing one example of a text data table used in the document analysis apparatus. FIG. 3 is a flowchart showing an example of the operation of the computer of the document analysis apparatus of FIG. FIG. 4 is an illustrative view showing an example of a corpus that increases with time. FIG. 5 is a table showing an example of the analysis result of the appearance frequency of each article and morpheme. FIG. 6 is a table showing the number N of unit documents for each article and morpheme. FIG. 6 (A) shows a general case where the number of linguistic material bodies is a constant amount (when the linguistic material body does not increase with time), and FIG. Show. FIG. 6A shows the number N of unit documents for each morpheme (t1, t2, t3...) Of the display example in order to unify the notation with other figures (FIGS. 5 to 8). FIG. 7 is a table showing the DF for each article and morpheme. FIG. 7A shows a general case where the number of language material bodies is constant (a case where the number of language material bodies does not increase with time), and FIG. 7B shows a case where the language material body increases in time series. FIG. 8 is a table showing TFIDF (A) and incremental TFIDF (B) for each article and morpheme. FIG. 8A shows a general case where the number of language material bodies is a constant amount (when the language material body does not increase with time), and FIG. 8B shows a case where the language material body increases in time series. FIG. 9 is an illustrative view showing one example of a regression curve. FIG. 10 is a graph showing a regression curve and a residual (positive / negative) with respect to the regression curve. The horizontal axis represents the sum of TFs, and the vertical axis represents the sum of the increased TFIDFs. FIG. 11 is an illustrative view showing one display example displayed on the monitor of the document analysis apparatus shown in FIG. FIG. 12 is an illustrative view showing another display example displayed on the monitor of the document analysis apparatus shown in FIG. FIG. 13 is an illustrative view showing a relationship between a corpus and a regression curve. FIG. 14 is a block diagram showing a document analysis apparatus according to an embodiment of the present invention. FIG. 15 is an illustrative view showing one example of a survey form in a free answer format that can be analyzed in the embodiment of FIG. FIG. 16 is an illustrative view showing one example of a GUI displayed on the monitor in the document analysis apparatus of FIG. 14 embodiment. FIG. 16 is a flowchart showing the operation of the computer for obtaining the representative keyword value (average cumulative singular value: ave ΣD) in the embodiment of FIG. FIG. 18 is an illustrative view showing one example of a graph displaying an average cumulative singular value (ave ΣD) of each singular word in the FIG. 17 embodiment. FIG. 19 is an illustrative view showing one example of a graph displaying ascending order cumulative singular values (ΣD (occurrence date, asc)) of each singular word when corpus are arranged in ascending order in the embodiment of FIG. 17. FIG. 20 is an illustrative view showing one example of a graph displaying a descending order cumulative singular value (ΣD (occurrence date, dsc)) of each singular word when the corpus is arranged in descending order in the embodiment of FIG. 17. In FIG. 21, the vertical axis is the ascending cumulative singular value (ΣD (occurrence date, asc)) of each singular word, and the horizontal axis is the descending cumulative singular value (ΣD (occurrence year) of each singular word in the embodiment of FIG. 17. It is an illustration figure which shows an example of the graph which is a month day, dsc)). FIG. 22 is an illustrative view showing one example of a graph in which the average cumulative singular value (ave ΣD) of each singular word is taken on the vertical axis and the TFIDF (i) of each singular word is taken on the horizontal axis in the embodiment of FIG. FIG. 23 is a flowchart showing the operation of the computer for obtaining the representative article value (average cumulative singular value sum: ave ΣΣ D) and cumulative singular value sum difference (diff ΣΣ D) in the embodiment of FIG. FIG. 24 is an illustrative view showing one example of a graph in which the vertical axis is the average cumulative singular value summation (aveΣΣD) and the horizontal axis is the cumulative singular value summation difference (diffΣΣD) in the embodiment of FIG. FIG. 25 shows the case frequency corresponding to each class in the Z-axis, the class of the average cumulative singular value sum (aveΣΣD) on the Y axis, the class of the cumulative singular value sum difference (diffΣΣD) on the X axis in FIG. It is an illustration figure which shows an example of a three-dimensional graph.

  In the following description, the document analysis apparatus that is the background of the present invention will be described with reference to FIGS. 1 to 13 within the scope necessary for understanding the present invention, and then the present invention will be described with reference to FIGS. Examples will be described.

  A document analysis apparatus 10 as a background of the present invention shown in FIG. 1 includes a computer 14 coupled to a communication network (network) 12 such as the Internet by wire or wirelessly. The computer 14 is basically provided with operating means 15A such as a keyboard and a mouse and a monitor 15B such as a liquid crystal display. The computer 14 is further provided with a text database 16 and an analysis database 18. Is done. Naturally, the computer 14 has an internal memory, and the internal memory (not shown) is used as a working memory or the like, and results data obtained by developing or calculating necessary programs, analysis result data, Temporarily stores various data during the analysis.

  The text database 16 sequentially stores time-sequential web news text data acquired by the computer 14 through the network 12, for example, and the computer 14 sequentially analyzes or analyzes the web news text data to obtain a time series. Singular words and common words (keywords) that change periodically.

  An example of the text data table 20 stored in the text database 16 is shown in FIG. Specifically, the text data table 20 is a table having, in one record, text data of “unit document” having an arbitrary fixed size from a language material composed of text data.

  As an example of the unit document, in the case of web news, there are an article within a predetermined period, an article for a day, an article, a paragraph, a sentence, and the like. Taking a newspaper as an example, there are one paper, one article, one paragraph, one sentence, and the like. In the case of literary works (novels), there are one work, one chapter, one paragraph, one sentence, and the like. In addition, when analyzing blogs on the web, an arbitrary unit can be used for language materials such as one diary as a unit document or one inquiry or complaint to the call center as a unit document. The database 20 is created by defining it as “unit document”.

  As shown in FIG. 2, in addition to an identifier (ID number) 22 and text data 24 formed by several degrees or alphabets, time information (time stamp) 26 is given as metadata to one record. . The time information 26 corresponds to a transmission date and time for a web news article and an inquiry time for an inquiry to a call center. The document analysis apparatus 10 of this background art is intended for language information whose number of characters increases with time, such as news and blogs. However, even linguistic materials that are not constantly updated, such as literary works, etc., because linguistic materials have linearity, people who read linguistic materials understand linguistic information over time. It will be. Therefore, for language materials such as novels and literary works that are static at first glance and do not have time information, in the field of time information 26 shown in FIG. The first paragraph, the second paragraph, the first sentence, the second sentence, etc.) may be added as metadata. In addition, an arbitrary field, for example, a title 26 is provided as necessary to create the database table 20. Further, in the case of a free answer description that can be analyzed according to the embodiment of the present invention described later, as this time (order) information, for example, in the social survey of disaster, the age of the respondent, the degree of damage to the house, the amount of damage, If an order standard such as household income is used and each free answer description (unit document) is arranged according to the order standard, the same technique as the document analysis apparatus of this background art can be applied. Further, when a free answer form survey form regarding commuting conditions is held, for example, commuting time and transportation costs related to commuting can be adopted as the “order standard”.

  If the text data table 20 is created by the computer 14, the text data table 20 is obtained through, for example, the network 12 using an application such as a DBMS (Data Base Management System) installed in the computer 14. Text data table can be created from web news.

  2 includes one unit document text data 24 (FIG. 2) identified by one identification symbol (ID) 22 shown in FIG. 2 and attached with time-series (order) information 26. Call. A language material (corpus) means a set of such records.

  In the explanation of the background art described later, web news is used as a linguistic material that increases the time series in which keywords (singular words, common words) should be detected. Data including arbitrary time elements (order reference elements) such as newspapers, magazines, blogs, interview records, statement records, questionnaires, novels, and free answer descriptions can be assumed.

  The analysis database 18 stores in advance all dictionaries and grammar rules necessary for document analysis in this embodiment, such as a part-of-speech dictionary for morphological analysis described later, and also accumulates analysis results. However, the analysis database 18 may be constituted by the internal memory of the computer 14 although the above-described text database 16 is the same.

  The computer 14 extracts or detects keywords by analyzing the document according to the document analysis program shown in FIG.

  Referring to FIG. 3, in first step S1, computer 14 determines whether or not a set time has elapsed. The “set time” is a delimiter time (Δt) for defining each corpus having a time-series order from language materials that increase in time-series. This “set time” can be freely set by the user. For example, a short set time (Δt) may be set when analyzing a language material in which a situation change occurs in a short time, and in the case of a reverse language material, the set time Δt may be increased. Examples of Δt include 1 hour, 10 hours, 100 hours, 1 day, 1 week, and 1 month. It is also conceivable to change this Δt with the passage of time. As an example, for example, Δt is set to “1 hour” until 24 hours have passed since the occurrence of the disaster, and after that, for example, Δt is set to “10 hours” until the third day after the disaster. For example, Δt is set as “1 day”.

  When an arbitrary set time is set by the user, the set time is stored in an appropriate memory area (register) of the computer 14, so that the computer 14 sets the internal clock data in the register. It is possible to determine whether or not the set time has elapsed in step S1.

  If “YES” is determined in the step S1, the computer 14 subsequently executes a corpus creation process in a step S3, and the text data of the unit document increased during the set time (Δt), for example, the text shown in FIG. Read from the data table 20 to create the current text corpus C (t).

  The corpus C (t) shown in FIG. 4 indicates the corpus at the current time, and this corpus C (t) is a corpus formed after a set time Δt from the corpus C (t−Δt) whose time series order is earlier. It is. That is, the corpus C (t) is the sum of the immediately preceding corpus C (t−Δt) and the increased amount of corpus CΔt.

  A “corpus” is defined as a collection of written language or spoken language materials for linguistic analysis, especially those constructed with electronic text. This refers to a collection of original text groups. Here, the above definition is taken in a broad sense. For convenience, morpheme groups having information on increased TFIDF and TF (both described later) are referred to as corpus for convenience. I will call it. Therefore, the text corpus here is to be understood as meaning a language material including text data of at least one record, that is, at least one unit document.

  Subsequently, in step S5, the text data 24 (FIG. 2) included in the corpus is divided into morphemes and part-of-speech information is added. Here, morpheme analysis is a language process in which sentences written in a natural language are divided into morpheme (or Morpheme, roughly speaking, the smallest unit having meaning in a language) to distinguish parts of speech. As the information source to be referred to, grammar knowledge of the target language (here, a collection of grammar rules) and a dictionary (a word list with information such as parts of speech) are used. The analysis database 18 is prepared in advance.

  In the embodiment, as an example, free morphological analysis software called “tea bowl” (http://chasen.naist.jp/hiki/ChaSen/) is introduced into the computer 14 and used.

  When the document is in Japanese, in the embodiment, a tool such as “tea bowl” is used so that the morpheme is first divided and extracted, and the part of speech is given to the extracted morpheme. However, for example, in a language system such as English, the word that is the smallest unit has already been divided, and therefore, the tokenization process (tokenization) is unnecessary. However, in this step S5, it is necessary to change the utilization form to the original form. Since it is necessary to identify the part of speech by performing a stemming process (stemming), tagging processing (tagging) is performed.

  Further, the morpheme (group) and the part of speech information analyzed in step S5 are stored in the text database 16.

  In the subsequent step S7, the computer 14 executes an unnecessary morpheme removal process for removing the morpheme of the part of speech type set as an unnecessary word based on the above-mentioned part of speech information.

  That is, in the morpheme analysis, whether or not to adopt the morpheme as a keyword candidate is selected based on the “part of speech information” given to each morpheme. The type of part of speech of a morpheme (single word (keyword) / common word candidate) to be an unnecessary word differs depending on the part of speech system output by the morphological analysis system and the user's intention of analysis. The type of part of speech that is recognized as an unnecessary morpheme is arbitrarily determined by the user. In the experiments actually conducted by the inventors, unnecessary nouns, verbs, adverbs, and adjectives that did not take the form of independence or suffix as a result of analysis using “tea bowl” were regarded as unnecessary morphemes. However, an unnecessary word removal rule for determining what part of speech morpheme is an unnecessary word may be set in the analysis database 18 in advance. In the case of English sentences, as described above, unnecessary morphemes are removed by referring to a list of words inappropriate as frequently used keywords called stop words.

  After executing step S7, for example, one or more necessary morphemes remain in the corpus stored in the text database 16, for example. Therefore, the process of step S9 is executed for all remaining morphemes that are not removed from the corpus. That is, in step S9, the computer 14 obtains an increased TFIDF for each of all remaining morphemes.

  Here, “TF” is Term Frequency, that is, the frequency (total number) (appearance frequency) of the occurrence of the keyword candidate in the unit document, and “IDF” in consideration of the time parameter is Inverse Document Frequency (reverse appearance document) Number), that is, uniqueness that does not appear elsewhere. Therefore, “increased TFIDF” means “TF” × “IDF”, which is referred to as “Term Frequency Inversed Document Frequency”, and may be expressed as TF * IDF, but here it is expressed as increased TFIDF. The incremental TFIDF is a kind of weighting index. In the background art, since the corpus including articles that increase in time series is targeted, the term “time-increasing TFIDF” is used. However, the present invention increases in accordance with an arbitrary order criterion as described below. Since the purpose is to analyze or analyze a corpus including documents, for example, a survey form in a free answer format, the word “increasing TFIDF” is simply used.

  Even if the number of articles changes sequentially as shown in FIG. 5, in the case of general analysis, since a fixed number N of unit documents are finally accumulated, the total number of unit documents N is a fixed number as shown in FIG. Therefore, the DF (Document Frequency) of TFIDF and the number of documents in which the morphemes appear when analyzing such general text data are a fixed number as shown in FIG. Therefore, TFIDF in the case of a general analysis method is as shown in FIG.

  In contrast, since one record handled in the background art system has time information or order information 26 (FIG. 2), each record (text data) can be arranged in time series order or order information order. Therefore, a subscript j (subscript based on time or order information) exists in the DF of the incremental TFIDF at that time. Here, “j” represents the order when records (articles) are arranged in time series order or order information order.

  Therefore, in the document analysis apparatus 10 of the background art, for example, when obtaining TFIDF for an article dj, the total number N of unit documents based on all articles collected finally and the DF based thereon are not used. N (j) (the total number of articles up to the time when article d (j) was sent) considering the time based on the number of articles sent before d (j) was issued, DF ( ti, dj) (the number of appearance documents of the morpheme t (i) up to the time when the article d (j) is transmitted), the TFIDF is sequentially calculated when the article d (j) is transmitted. In the document analysis apparatus 10 of this embodiment, as shown in FIG. 4, a corpus in which the number of unit documents included in the corpus is increased according to the time-series order, and the TFIDF of each morpheme in the corpus is calculated, thereby calculating the temporal Singular words (keywords) and common words in accordance with the order are extracted or detected from text data having the order (order).

Specifically, the normal TFIDF is calculated by the following equation (1), and the incremental TFIDF defined here is calculated by the following equation (2).
[Equation 1]
TFIDF (ti, dj) = TF (ti, dj) * IDF (ti)
IDF (ti) = log ₁₀ (N / DF (ti)) (1)
[Equation 2]
Incremental TFIDF (ti, dj) = TF (ti, dj) * IDF (ti, dj)
IDF (ti _, dj) = log ₁₀ (N (j) / DF (ti, dj)) (2)
Here, t (i) is a morpheme having i as an identifier (ID). That is, it is a keyword candidate that is a target for calculating TFIDF (ti, dj).

  d (j) represents the jth unit document (article). That is, it is a document that includes keyword candidates that are targets for calculating TFIDF (ti, dj) and incremental TFIDF (ti, dj). However, the unit of the document can be arbitrarily set such as a sentence, an article, and a sentence, but in the background art, an article of web news is set as a document unit.

  TFIDF (ti, dj) and incremental TFIDF (ti, dj) are values calculated for each morpheme t (i) of the j-th unit document.

  TF (ti, dj) is a value calculated for each morpheme t (i) of the j-th unit document, and is the number of times that the morpheme t (i) appears in the unit document d (j) (total number).

  DF (ti, dj) is the number of unit documents in which the morpheme t (i) appears in the 1st to jth unit documents.

  N (j) is the number of unit documents that appear when the unit document d (j) is generated. If the IDs of several degrees are assigned to the unit documents in order from 1, The value of N is the same as j.

  For example, as shown in FIG. 5, morphemes t (1), t (2), t (3),... Appearing in each article (unit document) d (1), d (2), d (3),. Suppose that changes. In this case, a table having the number N (j) of unit documents in the field is represented as shown in FIG. Further, a table having the DF (ti, dj) of each unit document in the field is represented as shown in FIG. 7B, and each unit document having the morpheme t (i) as an identifier according to the value of N (j). FIG. 8B shows a table having incremented TFIDF (ti, dj) values in the field. All of these tables are sequentially stored in the text database 16.

  In this way, after the increased TFIDF of all the morphemes is calculated in step S9, in the subsequent step S11, the computer 14 calculates the cumulative value Σincreased TFIDF of the increased TFIDF and the cumulative value ΣTF of the TF. Calculated as actual values up to corpus C (t). Since the increased TFIDF (ti, dj) is as shown in FIG. 8B and DF (ti, dj) is shown in FIG. 7B, TF (ti, dj) is also calculated. As for ΣTF, after calculating TF (ti, dj), it may be calculated as a cumulative value thereof. However, for the incremental TFIDF, the cumulative value may be calculated from the table of FIG.

In the following step S13, the computer 14 sets the cumulative value ΣTF of TF (ti, dj) obtained for the corpus C (t) as X, and sets the cumulative value Σincreased TFIDF of the increasing type TFIDF (ti, dj) as Y. By fitting to the following equation (3), constants a and b are obtained, and a regression curve shown in FIG. 9 is created. This regression curve estimates or predicts the increased TFIDF in the corpus C (t + Δt) for the residual analysis in the next corpus C (t + Δt). That is, when the ΣTF up to the corpus C (t) becomes the horizontal axis, if the increased TFIDF shows the same tendency in the next corpus C (t + Δt), the next corpus C (t + The incremental TFIDF at Δt) will be plotted on this regression curve.
[Equation 3]
Y = aX ^b (3)
In step S15, the computer 14 adds the cumulative value Σincrease TFIDF of the increased TFIDF (ti, dj) in the corpus C (t) at the time j calculated in the previous step S11 and the previous corpus C (t the difference between the estimated value Y -.DELTA.t) for by regression curve Y = aX ^b determined in step S13 obtains the (residual value) (Figure 10). The larger the residual value, the farther away from the Σincrease type TFIDF of the same morpheme t (i) predicted by the immediately preceding corpus C (t−Δt), regardless of positive or negative, That is, it means that it could not be predicted from common sense up to the immediately preceding corpus. Therefore, the residual value is referred to as a value representing the singularity of the morpheme, that is, a singular value (Discriminating Value). On the other hand, a morpheme in which the Σincreased TFIDF shows a positive residual value (singular value) is plotted above the regression curve, meaning that it is specific or characteristic. A morpheme in which the Σincrease type TFIDF shows a negative residual value (singular value) has no singularity and can be said to be a common morpheme having the opposite property.

Referring to FIG. 10, with respect to the regression curve represented by Y = aX ^b, if Σ increasing type TFIDF morpheme t (i) could be plotted above the curve, the morpheme t (i) is a positive Will have a residual value. Having a positive residual value means that the morpheme t (i) does not appear so much by C (t−Δt), but appears abruptly in the elapsed Δt. If the Σincrease type TFIDF of the morpheme t (i) of C (t) is below the regression curve, it indicates that many morphemes have appeared up to C (t−Δt). For each morpheme, a residual analysis is performed between the estimated value or predicted value of the Σincrease TFIDF and the actual measurement value, and the singular value, that is, the residual value of each morpheme is stored in, for example, the text data table 20 (see FIG. It is sequentially stored by adding it as metadata to 2).

  In the next step S17, the computer 14 selects a singular word (keyword) and a common word (keyword) according to the singular value (residual value) stored in the database 16 as described above. For example, a morpheme having a positive residual value (singular value) greater than an arbitrary upper number is selected as a singular word representing the corpus. Conversely, a morpheme whose negative residual value (singular value) is less than an arbitrary lower number is selected as a common word. The common language corresponds to a keyword representing the entire constructed text database (language material). By using these singular and common words, text data (language material) of the same theme can be searched efficiently.

  Subsequently, in the last step S19, the computer 14 displays the singular words and common words selected in step S17 on a display (not shown).

  FIG. 11 shows a display example when analyzing using web news issued for the 2004 Niigata Chuetsu Earthquake. In FIG. 11, singular words having a positive residual value are plotted on the upper side of the display screen with time (horizontal axis), and common words having a negative residual value are plotted on the lower side. However, since details cannot be drawn in FIG. 11, only two “death” and “dispatch” are specified as singular words, and only two “earthquake” and “Chuetsu” are specified as common words. Note that morphemes (words) constituting the graph are displayed in each graph portion. According to the display example as shown in FIG. 11, the singular words and the common words are displayed separately above and below, so that there is an advantage that they can be listed.

  As a display example, a tabular display shown in FIG. 12 is also conceivable. In the table of FIG. 12, the horizontal axis shows the passage of time, and the vertical axis displays the appropriate number of unique words for each time segment (rank: Rank).

  However, it is needless to say that other arbitrary display forms are possible, and the present invention is not limited to the display examples of FIGS.

  As described above, if index information indicating the degree of characteristics is added to a keyword at a certain point in time, a more specific keyword can be identified based on the evaluation result of the index. At a certain point in time, if a certain matter is sent centrally on web news, many words representing the meaning of the certain matter may appear. However, there are two types of frequently used keywords: any news article, a keyword that is frequently used in composing a document, and a keyword that appears frequently in some news articles. Is imagined. The keyword that characterizes the news article is the latter.

  As an index that gives a high weight to the latter keyword, there is the TFIDF described above. Here, as described above, TF (ti, dj) indicates the number of times the keyword t (i) appears in the article d (j), and DF (ti) indicates the number of documents in which the keyword t (i) appears. IDF (ti) is the reciprocal of the ratio of the number of documents in which the keyword t (i) appears to the total number of documents. That is, in this embodiment, a low weight is given to a morpheme that appears in any article, and a high weight is given to a morpheme that does not appear much in other articles. Incremental TFIDF, which is the product of this and TF, is an index that indicates how many appear in the article and how it does not appear in other articles, and is an index that evaluates the degree of the feature of the keyword It can be said.

  And, in the experiment in the background art of the inventors etc., when obtaining the increased TFIDF for an article d (j), N and DF based on all 2623 articles collected in the end are not used. N (j) (the total number of articles up to the time when article d (j) was sent) considering the time based on the number of articles sent before d (j) was issued, DF ( ti, dj) (the number of documents in which the morpheme ti appears until the article d (j) is sent) is used to calculate the TFIDF sequentially when the article d (j) is sent. This is called incremental TFIDF. That is, in normal TFIDF, N and DF are constant, and do not correspond to weighting for morphemes extracted from increasing language material. For this reason, in the background art, the total number of documents and the number of documents in which an arbitrary morpheme appears are used as parameters that change based on the order reference, and an increased TFIDF in which TFIDF is modified is used.

  However, it is difficult to evaluate whether or not a keyword is characteristic only by the value of incremental TFIDF. In a pattern in which the value of the increased TFIDF up to a certain point is highly evaluated, the IDF is high (DF is low) even if the value of the TF is low. In some cases, even if the DF is high, the incremental TFIDF is calculated to be high because the TF is remarkably large. The fact that the TF is remarkably large is likely to be a word that must be used many times to describe an article because of the high generality of the word. It is not possible to simply evaluate whether the morpheme is characteristic by the value of incremental TFIDF.

  The fact that the information at a certain point in time is characteristic can be understood by comparing the keyword group spoken up to the previous point in time with the keyword group spoken at a certain point in time. If there is a difference between the two, it may mean that there was a large quality difference before and after the arbitrary time point. In other words, by comparing the corpus at a certain point in time with a corpus that has passed a certain amount of time from a certain point in time, it is possible to grasp the change in the quality of information and identify the keyword that caused the change. It is done. Therefore, in the background art of the present invention, as described above, the residual analysis (step S15) is performed to compare the corpus characteristics at a certain time point and the next time point.

When the inventors evaluated the relationship between the cumulative value of TF and the cumulative value of increasing TFIDF for each morpheme at different times from the occurrence of actual news regarding a disaster, the cumulative value of TF and the cumulative value of increasing TFIDF There was a strong relationship between the values expressed by the previous equation (3). In a period when the number of samples (number of keywords) is small, R ² is 0.90 to 0.99 as a power function, except for the relationship between the cumulative value of TF and the cumulative value of incremental TFIDF (TF, and TF and incremental TFIDF It has been clarified that there is a systematic relationship between the cumulative values of the power functions of the TF, such that the keyword in the vicinity of the approximate curve is the cumulative value of the TF and the cumulative value of the increasing TFIDF. This means that the value relationship has the same tendency as the average corpus relationship, and keywords with such a tendency are considered to have an average appearance pattern. When the actual cumulative value of TFIDF is lower than the estimated value based on the approximate curve, the cumulative value of incremental TFIDF is low, that is, the degree of features is too high as viewed from the average image of the corpus. Conversely, if the actual measured value exceeds the estimated value, it can be said that the increase type TFIDF is high and is a characteristic keyword. This is made possible by calculating the difference (residual) between the accumulated value and the estimated value based on the approximate curve, applying the above relationship to evaluate the degree of characteristic of the keyword at an arbitrary point in time using the model shown in FIG. To do.

  The left side of FIG. 13 schematically shows a change in the corpus when a unit time width Δt elapses from a certain t−Δt. Such a relationship can be expressed by the following equation (4).

  As shown in FIG. 13A, when C (Δt) contains many keywords that have appeared so far, or there are only morphemes that do not appear very frequently, As shown in the upper right side of FIG. 13, the relationship between the cumulative value of TF and the cumulative value of incremental TFIDF is the case where it is composed of a corpus (C (t-Δt)) at time t-Δt and the time point t A large difference does not occur in the case of a corpus (C (Δt)). On the other hand, as shown in FIG. 13B, when a keyword that does not appear until t−Δt appears in Δt or there is a morpheme that appears frequently, t At this point, the corpus (C (t)) changes greatly, and as shown in the lower right side of FIG. 13, the shape of the curve representing the relationship between the accumulated value of TF and the accumulated value of increased TFIDF also changes greatly.

  In other words, the residual between the cumulative value of the incremental TFIDF at a certain time t and the estimated value based on the relational expression composed of the corpus at the time t−Δt represents the change in the corpus during this Δt. A morpheme having a large residual is considered to be a keyword (single word, common language) representing the content of language material generated during Δt.

  As described above, in the embodiment, as an index for evaluating the feature amount of the keyword representing the qualitative change in the information content at time t, the cumulative value of the TF composed of the corpus of the arbitrary time t-Δt and the incremental TFIDF The difference (residual) between the estimated value of the cumulative value of the incremental TFIDF based on the relational expression based on the measured value of the cumulative value of the incremental TFIDF at the time t is adopted. Here, a keyword having a significantly high residual is called a feature word or singular word (residual value or singular value: positive), and a keyword having a very low residual is called a general word or common word (residual value or singular value: negative) ).

  According to the document analysis apparatus 10 of the background art shown in FIG. 1, according to the following procedure shown in the flowchart shown in FIG. 3, the computer 14 does not use human subjective judgment, but increases TFIDF indices and residual values. Because it consists of a series of quantitative indicators, and consists of a series of processes, if the tools and what to be referred to are properly prepared, records of past events can be used as inputs, and automated through a series of processes. It is possible to objectively detect a keyword that is a final product.

  Thus, in the document analysis apparatus 10 of the embodiment shown in FIG. 1, the computer 14 basically executes the following steps.

  1) Build a database of text data (in this case, web news) that increases over time.

  2) Divide the text into morphemes and add part-of-speech information.

  3) Extract nouns, verbs, adverbs, and adjectives other than independence and suffix based on part of speech information.

  4) For the morpheme, obtain an incremental TFIDF based on TF and time information for each document (here, web news article).

  5) In order to extract a keyword representing a characteristic text between time t-Δt and t, a relational expression between the cumulative value of TF and the cumulative value of incremental TFIDF in the corpus up to t-Δt is obtained. The difference between the estimated value of the cumulative value of the incremental TFIDF at the time point t and the actual measurement value is obtained. This residual value is defined as a feature amount of a keyword that appears at a certain Δt, that is, a singular value.

  6) Select keywords (singular terms) from the largest residual value (singular value) to any number of upper ranks, and use the singular terms as the metadata of the language material in the articles where the singular terms are detected.

  As described above, the proposed document analysis method generates a time-series data set by numerically evaluating the singularity of an arbitrary time section for words in a corpus that increases in time series. is there. If the trend variation of this time series data can be accurately grasped, there is a possibility that development of an arbitrary event can be predicted.

  In the background art described above, the incremental TFIDF is defined so that it can be applied to a corpus that considers only the generation time of a unit document. However, even if a document does not exist in real time, such as a free answer description, each free answer description is arranged on a pseudo time axis if it is generated according to a certain order criterion. Thus, it is possible to weight words by increasing TFIDF. In this manner, by arranging the unit documents of the free answer description according to a certain order reference, the incremental TFIDF in the background art can be applied in a pseudo manner.

  Therefore, the document analysis apparatus 10 according to the embodiment of the present invention shown in FIG. 14 analyzes the corpus having the free answer text data as a unit document by using the background art shown in FIGS. 1, the document analysis apparatus 10 includes a computer 14 having an operation means 15A and a display means (monitor) 15B. The computer 14 is provided with the text database 16 and the analysis database 18 described above. And the network 12 is coupled.

  Further, in the embodiment of FIG. 14, an image scanner 30 attached to the computer 14 is installed, and a text recognition technique is applied by reading a survey form 32 including a free answer description by the image scanner 30 and applying a character recognition technique. You may make it convert into data. At this time, face items (to be described later) normally set in the free answer description may be restored from the read data of the image scanner. Then, the contents of the survey form including the free answer description can be automatically taken into the text database 16 as the text data 20.

  However, the text database 16 may be manually created from such a survey form. That is, the collected survey form is coded and a free answer description is input as text. In the coding process, as is well known, actual answer results (numbers, etc.) are input into a matrix of case (respondent) × attribute (question item). However, if the number itself is the answer, such as “age”, the number is input as it is. If a matrix is created by such a coding process, survey forms can be arranged in ascending or descending order according to a predetermined order criterion such as age.

  Even when the image scanner 30 automatically reads the survey form 32, the matrix can be automatically created according to the restored face item. In this case, the survey forms are arranged in ascending or descending order according to a predetermined order standard. be able to.

  FIG. 15 shows an example of a survey form including a free answer description that can be analyzed by the document analysis apparatus 10 of the embodiment of FIG. The survey form 32 is entered on, for example, A4 size paper, and a face item entry area 34 is set at the top of the page. Further, the lower part of the survey form 32 is set as a free answer description area 36. In this embodiment, the survey form 32 is shown as only one for the purpose of illustration, but it may be a set of a plurality of survey forms, and furthermore, only the face item entry area 34 and the free answer description area 36 are included. Instead, a question answer area (not shown) in a precode answer format (a format in which options are checked) may be set.

  In the face item entry area 34, for example, entry fields such as gender, age, occupation, number of years of service (year of service or number of years of service), number of households, household income, etc. are set. It is set as a question. Order criteria when the free answer description in the free answer description area 36 of the survey form 32 is arranged in ascending or descending order as the unit document, for example, age, years or household income entered or selected in the face item entry area 34 Can be used as In the free answer description area 36, literal answers and opinions are described literally.

  In order to analyze the free answer description, the computer 14 in FIG. 14 enables the setting by the user by displaying, for example, the GUI 40 shown in FIG. 16 on the monitor (display means) 15B. In the GUI 40, analysis item selection buttons 42, 44 and 46 for analyzing a free answer description are set. Further, a corpus selection area 48 and a target setting area 50 are set on the right side of the GUI 40. In the corpus selection area 48, the contents of the corpus (language material) that can be selected at that time are described as the corpus name, and the user can obtain the evaluation item selected by the buttons 42-46 displayed above. When the corpus to be selected is selected, the corpus name may be operated (clicked). Further, in the target setting area 50, the upper rank of the cumulative singular value ΣD of each morpheme (word) that is a reference for evaluation in this embodiment is input. That is, an arbitrary number is input as n in the window 52 set in the area 50. For example, when “50” is set in the window 52, it means that the user has set to target the top 50 morphemes of the cumulative singular value ΣD.

  The representative keyword value selection button 42 is clicked by the user, for example, with a mouse (not shown) when selecting a representative keyword value as an evaluation item. As explained above, a numerical value obtained by adding a singular value (Discriminating Value) from the beginning to an arbitrary time (order reference) section is called a cumulative singular value, and is used as an index for identifying an important singular term. Words with positive singular values are words that characterize each time (order reference) cross section (singular words), and words with negative values are ubiquitous words in the corpus (common language). If singular values are simply added together, negative values are added together when an arbitrary word changes from a singular word to a common word. These terms are important because they will reduce the accumulated singular value despite being a singular term associated with an event that is important for a certain period of time, i.e. a range of order criteria. There is a possibility that it cannot be obtained as a singular term. Therefore, here, the sum of only positive singular values is taken as the cumulative singular value (ΣD). The cumulative singular value ΣD is an index indicating how much weight the morpheme (word) has.

  However, in the background art of FIG. 1, in principle, a high weight is given to words (morphemes) included in a document arranged rearward in the order reference. Therefore, if the free answer description is analyzed by the method as it is, it is expected that a keyword that strongly reflects the characteristics of the order criterion is extracted. For example, ascending cumulative singular values ΣD (i, ord, asc) when free answer descriptions are arranged in ascending order on an order criterion such as age, and descending cumulative singular values ΣD (i, ord) when arranging free answer descriptions in descending order on an order criterion , dsc), the former may extract keywords representing the characteristics of the elderly, and the latter may extract keywords representing the characteristics of the younger generation. Therefore, in this embodiment, the arithmetic average of the cumulative singular value ΣD (i, ord, asc) in ascending order and the cumulative singular value ΣD (i, ord, dsc) in descending order obtained by Equation 5, that is, the average cumulative singular value aveΣD The idea is to obtain a more objective (or representative) analysis result that eliminates the characteristics of the order criteria that we focused on. This average cumulative singular value aveΣD therefore indicates that the morpheme (keyword) is prominent, that is, how representative the morpheme (keyword) is in both the young and the elderly, for example when the order criterion is age. Become. Therefore, the average cumulative singular value aveΣD can also be referred to as a representative keyword value. This representative keyword value aveΣD is one index for evaluating the description (unit document) in the free answer format.

Where D (i, j, ord, asc / dsc): a singular value of word i in unit document j obtained in ascending order (asc) or descending order (dsc) of order reference (ord).

  The second index is an index based on the sum (cumulative) of the above-mentioned ascending / descending order cumulative singular values ΣD. The cumulative singular value ΣD is an index indicating how important the word (morpheme) is. Therefore, the idea that an article having a larger cumulative singular value ΣD and containing more morphemes (words) having a higher weight is an important free description (article) holds. Therefore, in the embodiment shown in FIG. 14, ΣΣD (cumulative singular value summation) indicating how important morphemes are included in one article is adopted as the second index. However, if ΣΣD is adopted as it is, there is a problem that the numerical value is influenced by the length of the description sentence (number of words). Therefore, the cumulative singular value sum ΣΣD obtained by accumulating the cumulative singular values ΣD of the morphemes that are higher than a certain level is used. This accumulated singular value sum ΣΣD is given by Equation 6.

Where rank is the number (rank) of upper words of the cumulative singular value summation ΣΣD (j, ord, asc / dsc) to be considered for weighting the free answer description, and rank (j) is the ΣD (of the word (morpheme) i i, ord, asc / dsc).

  On the other hand, as described above, the cumulative singular value ΣD is affected by whether the documents included in the corpus are arranged in ascending order or descending order of the order reference. Therefore, it can be easily predicted that the cumulative singular value sum ΣΣD, which is the sum of the cumulative singular values ΣD of each morpheme in each document (article), is also affected by the ascending / descending order.

  Therefore, in the embodiment, after adopting the specific higher order ascending cumulative singular value sum ΣΣD (j, ord, asc, rank) and the descending order cumulative singular value sum ΣΣD (j, ord, dsc, rank), Calculate the arithmetic mean of the ascending cumulative singular value sum ΣΣD (j, ord, asc, rank) and the descending cumulative singular value sum ΣΣD (j, ord, dsc, rank) to calculate the average cumulative singular value sum aveΣΣD (j, ord, rank). This average cumulative singular value sum aveΣΣD (j, ord, rank) is called a representative document value or a representative case value.

  As explained earlier, TFIDF (i), due to its nature, may add high weight even to words that do not have content meaning themselves or words that have a high degree of comprehensive abstraction. Therefore, there is a possibility that appropriate weighting is not made for the free answer description aimed by the present invention. On the other hand, the ascending cumulative singular value ΣD (j, ord, asc, rank) according to the order criterion and the descending cumulative singular value ΣD (j, ord, dsc, rank) according to the order criterion are more specific as described above. A meaningful morpheme (word) shows a high value. Therefore, if these words and their weights are used as they are, the ascending cumulative singular value sum ΣΣD (j, ord, asc, rank) and the descending cumulative singular value sum ΣΣD (j, ord, dsc, rank) One of the characteristics of the order criterion is strongly shown. On the other hand, a homogeneous representative value that does not require the effect of the order criterion, such as TFIDF (j, rank), is also required. Therefore, in order to offset the effect of the order criterion, consider taking the average value of both indicators. This is called the average cumulative singular value sum aveΣΣD (j, ord, rank), and is adopted as an index indicating how representative the article (free answer description) is.

  The free answer description showing the high values of the ascending cumulative singular value sum ΣΣD (j, ord, asc, rank) and the descending cumulative singular value sum ΣΣD (j, ord, dsc, rank) Is expected to be. There is a possibility that there is a free answer description having both characteristics, and the effect of the order criterion cannot be expressed appropriately. When the effect of the order criterion is to be emphasized more, the difference between the ascending cumulative singular value sum ΣΣD (j, ord, asc, rank) and the descending cumulative singular value sum ΣΣD (j, ord, dsc, rank) is obtained. Therefore, it is considered that weighting that reflects the ascending / descending order characteristics of the order reference becomes possible with the absolute value. This is called the difference accumulated singular value sum diffΣΣD (j, ord, rak), and is expressed by the following equation (8).

  For example, if the difference accumulated singular value sum diffΣΣD (j, ord, rak) shows a positive value, the free answer description indicating the ascending nature of the order reference ord (in the case of age, the nature of the elderly) When a high weight was shown and a negative value was shown, it was expected that the free answer description indicating the descending order (in the case of age, the nature of the younger generation) would show a higher weight. That is, the index of the number 7 is a numerical value indicating how specific the article (free answer description) shows, that is, a specific document value or a specific case value.

  As described above, in the document analysis apparatus 10 of the embodiment of FIG. 14, three indexes that are considered to be important in analyzing the free answer description, the average cumulative singular value aveΣD (i, ord, rank), and the average cumulative singular value summation. aveΣΣD (j, ord, rak) and difference summation singular value sum diffΣΣD (j, ord, rak) were adopted.

  First, referring to FIG. 17, the user clicks the representative keyword value selection button 42 in the GUI 40 of FIG. 16, and the corpus of the free answer description is expressed by the evaluation item of the average cumulative singular value aveΣD (i, ord, rank). The operation when the analysis is selected will be described.

  In the background art, since the document is increased according to the time reference, the timing for executing the operation according to the flowchart of FIG. 3 is determined based on whether the set time has elapsed. However, in this embodiment, since the document is directed to a corpus in which the documents are arranged according to an arbitrary ordering criterion, the computer 14 performs the ordering criterion every time one or more documents (free answer descriptions) increase, that is, within the corpus. The operation shown in FIG. 17 is executed whenever one or more documents are added. That is, in order to execute the operation of FIG. 17, it is necessary to increase the number of documents according to the order reference.

In step S31 of FIG. 17, all documents (corpus text) including the document i increased by one at that time included in the corpus selected by the user operating the corpus selection area 48 of the GUI 40 are ordered, for example, age. According to the ascending order. Thereafter, the residual value of each morpheme, that is, the singular value D (i) is obtained by executing steps S3-S19 of the background art of FIG. However, at that time, the cumulative value ΣTF of TF (ti, dj) obtained in the previous corpus including the previous document i-1 in the order reference is set to X, and the cumulative value Σincrease type of the increasing type TFIDF (ti, dj). By applying TFIDF to Y and fitting to the above-mentioned equation 3 to obtain constant a and constant b, step S13 (FIG. 3) creates the regression curve of FIG. In step S15 (FIG. 3), the cumulative value Σincrease TFIDF of the increased TFIDF (ti, dj) in the current corpus including up to the document i and the regression curve Y obtained from the previous corpus including up to the document i-1. = difference between the estimated value Y by aX ^b (residual value) ie obtaining the singular values. Accordingly, in step S33 of FIG. 17, all singular values of the morphemes in the current corpus in which the documents i are arranged in ascending order according to the order criterion are calculated.

  After that, in step S35, the ascending cumulative singular value ΣD (i, ord, asc) is calculated according to the above equation 5.

  Next, Steps S37 to S41 are executed in the same manner as Steps S31 to S35, and all the singular values of each morpheme in the current corpus in which the documents i are arranged in descending order according to the order criterion are calculated. Is used to calculate the descending cumulative singular value ΣD (i, ord, dsc).

  In the subsequent step S43, an arithmetic average of the ascending cumulative singular value ΣD (i, ord, asc) and the descending cumulative singular value ΣD (i, ord, dsc) is calculated to obtain a representative keyword value.

  In step S45, the monitor 15B causes the average cumulative singular value aveΣD, the ascending cumulative singular value ΣD (i, ord, asc) and the descending cumulative singular value ΣD (i, ord, dsc), Displays a graph (described later) expressing the forward cumulative singular value ΣD (i, ord, asc) and the descending cumulative singular value ΣD (i, ord, dsc) on the vertical axis and the horizontal axis.

  In an experiment conducted by the inventors, a site established by the Nuclear Safety Technology Center (IINET (Incident Information Network System) system: http://www.n-linet.ne.jp) ) Was used in the “Accident / Fault Information Database” (http://www.n-linet.ne.jp/default.htm). The results are shown in FIGS.

  In the experiment, each record was obtained from the database and a corpus was created. Then, the date of occurrence is adopted as the order reference, the corpus texts are arranged in ascending / descending order, the unit document order is changed according to the order reference, and the process shown in FIG. 17 is executed.

  FIG. 18 is a graph showing the average cumulative singular values ave ΣD arranged in ascending order by averaging ΣD in ascending order and ΣD in descending order. In this graph, however, the ascending cumulative singular value ΣD (i, ord, asc, rank) and the descending cumulative singular value ΣD (i, ord, dsc, rank) calculated using only the top 50 morphemes (words) in the GUI 40 ) Is the average cumulative singular value aveΣD (i, ord, rank). FIG. 19 is a graph showing the ascending order singular values ΣD (i, ord, asc, rank), and FIG. 20 is a graph showing the descending order singular values ΣD (i, ord, dsc, rank).

  Since FIG. 19 shows ascending cumulative singular values ΣD (i, ord, asc, rank), the morphemes that appear later in the generation date as the order reference tend to have higher weights. , “Facility”, “management”, “system”, “nucleus”,... Since FIG. 20 shows the descending order cumulative singular values ΣD (i, ord, dsc, rank), the morphemes in which the date of occurrence as the order reference appears earlier tend to have a higher weight. Here, it can be seen that the cumulative singular value increases in the order of “PU (plutonium)”, “leakage”, “apparatus”, “allowable”, “exposure”,. In other words, in nuclear incident reports, there were many failures and accidents caused by equipment (hardware) in the early days, but recently, such incidents have not occurred. It can be seen that many failures have occurred.

  In the average cumulative singular value aveΣD (i, ord, rank) shown in FIG. 18, a morpheme having a large cumulative singular value in ascending order and a morpheme having a large cumulative singular value in descending order have relatively large cumulative singular values. calculated. That is, it can be seen that a morpheme (word) having a large average cumulative singular value aveΣD (i, ord, rank) can be a representative keyword representing the corpus.

  Fig. 21 The vertical axis shows the ascending cumulative singular value ΣD (i, ord, asc, rank) and the horizontal axis shows the descending cumulative singular value ΣD (i, ord, dsc, rank). It is a graph which shows whether it is done. When FIG. 21 is seen, as described above, morphemes of “discovery”, “facility”, “management”, “system”, “nucleus”,... Morphological elements such as “Pu (plutonium)”, “leakage”, “apparatus”, “acceptance”, “exposure”,. Therefore, by looking at the graph of FIG. 21, it is possible to easily understand what keywords are representative of the corpus of this nuclear power-related failure accident report.

  FIG. 22 is a graph in which each morpheme is plotted with the average cumulative singular value aveΣD (i, ord, rank) on the vertical axis and the increased TFIDF (i) on the horizontal axis, and was evaluated in this example. It can be seen that a morpheme having a large average cumulative singular value aveΣD (i, ord, rank) is plotted at a high level. That is, it shows that the average cumulative singular value aveΣD (i, ord, rank) is reliable to some extent. For example, the keywords “facility”, “acceptable”, “Pu”, or “leakage” show almost the same results as the conventional weighting index TFIDF (i) that does not consider time information, while Japan like “Yes” Words that are used very frequently due to the characteristics of words, or words that have no meaning in themselves, or words that are naturally used frequently due to the nature of the corpus (article), such as "contamination" or "work", have a high weight. Since it is not shown, reliability is guaranteed.

  When the selection button 38 or 40 is operated on the GUI 40 shown in FIG. 16, the processing shown in FIG. 23 is executed by the computer 14 of the embodiment shown in FIG.

  Steps S51 to S61 in FIG. 23 are the same as steps S31 to S41 shown in FIG. 17 except for the following points, and redundant description is omitted here.

  That is, the ascending order cumulative singular value ΣD (i, ord, asc, rank) is calculated in step S35 of FIG. 17, whereas the ascending order cumulative singular value sum ΣD according to the above-described equation 6 in step S55 of this embodiment. Calculate (j, ord, asc, rank). While the ascending order cumulative singular value ΣD (i, ord, dsc, rank) is calculated in step S41 in FIG. 17, the ascending order cumulative singular value sum ΣD (j, ord, asc, rank) is calculated in step S61 of this embodiment. Calculate

  Thereafter, in step S63 of FIG. 23, the average cumulative singular value sum aveΣΣD (j, ord, rank) is calculated according to Equation 7. This evaluation value alleviates the influence of the characteristics of the order criterion in the ascending cumulative singular value sum ΣΣD (j, ord, asc, rank) and the descending cumulative singular value sum ΣΣD (j, ord, dsc, rank). This is an index indicating how representative an article (free answer description) is. That is, an article having a large average cumulative singular value sum aveΣΣD (j, ord, rank) is a representative article representing the corpus at this time.

  On the other hand, in step S65, the difference accumulated singular value sum diffΣΣD (j, ord, rank) is calculated. As described above, an article showing a high value in ascending cumulative singular value sum ΣΣD (j, ord, asc, rank) and an article showing a high value in descending order cumulative singular value sum ΣΣD (j, ord, dsc, rank) Are greatly affected by the ascending and descending characteristics. However, there is a possibility that an article having both characteristics may exist in the corpus, and the effect of the order criterion may not be expressed appropriately. Therefore, the difference accumulated singular value sum diffΣΣD (j, ord, rank) is adopted as an index indicating how much the article has specificity in the corpus by further emphasizing the effect of the order criterion.

  Finally, in step S67, the magnitude of the average cumulative singular value sum aveΣΣD (j, ord, rank) is shown on the monitor 15B on the vertical axis, and the horizontal axis is the difference cumulative singular value sum diffΣΣD (j, ord, rank). The graph as illustrated in FIG. 24 showing the size is displayed. In this graph as well, as in the previous example, accident / failure report articles related to nuclear power are selected as corpora. Since the horizontal axis is the difference accumulated singular value sum diffΣΣD (j, ord, rank), characteristic articles are plotted on the right half and the left half of the graph. In other words, articles with a large cumulative singular value sum in ascending order of the order criteria are plotted on the right from “0” on the horizontal axis, and articles with a large cumulative singular value sum in descending order are plotted on the left from “0” on the horizontal axis. The The vertical axis represents the average cumulative singular value sum aveΣΣD (j, ord, rank), and an article representing this corpus is plotted at the top. However, less important articles are plotted at the bottom. As can be seen particularly in FIG. 25 where the height of the bar graph indicates the number of articles, the upper half of the vertical axis (corresponding to the lower half in FIG. 24) is packed with less important articles. . So if you have time constraints, you can see that you don't have to read the articles plotted down the vertical axis.

  Here, let us look at the results of experiments conducted by the inventors and the like to see what articles were specifically found in the three zones bcd, ab, and de shown in FIG.

  Zone bcd is a zone in which the average cumulative singular value sum aveΣΣD (j, ord, rank) indicates the presence of articles in the top 5%. That is, important articles representing the corpus are plotted in the zone bcd. Examples 1, 2, and 3 of articles corresponding to the zone bcd are shown in Table 1 below. However, in each table, proper nouns (place names and personal names) in the article are written in lowercase so that unexpected effects do not spread.

  Article 1 is an article describing an event that occurred in the United Kingdom on December 5, 1978, and aveΣΣD was the maximum value “2083” (diffΣΣD = 193). The important keywords in this article were "facility", "management", "acceptable", "safety", "exceed", "Pu". Article 2 describes an event that occurred in the United Kingdom on September 14, 1987, and aveΣΣD was “1719” (diffΣΣD = −730). The important keywords in this article were "acceptable", "suspect", "exceed", "result" and so on. Article 3 is an article describing an event that occurred in England on August 1, 1978, and aveΣΣD was “1696” (diffΣΣD = −695). The important keywords in this article were “acceptable”, “suspect”, “exceed”, “result”, “monitoring” and so on.

  Zone ab is a zone in which the positive difference cumulative singular value sum diffΣΣD (j, ord, rank) indicates the presence of articles in the top 5%. That is, important articles are plotted relatively recently in the zones ab. Examples 4, 5, and 6 of articles corresponding to this zone ab are shown in Table 2 below.

  Article 4 describes an event that occurred in the United States on March 11, 2008, and diffΣΣD was the maximum value “1947” (aveΣΣD = 974). The key keywords in this article were “discovery”, “nuclear”, “safety”, and “violation”. Article 5 is an article describing an event that occurred in Japan on March 11, 2007, and diffΣΣD was “1731” (aveΣΣD = 971). The key keywords in this article were “discovery”, “management”, “nuclear”, and “fuel”. Article 6 is another article describing the event that occurred in Japan on October 4, 2007, which is described in article 2 above, and diffΣΣD was “1731” (aveΣΣD = 971). The key keywords in this article were “discovery”, “management”, “nuclear”, and “fuel”.

  The zone de is a zone indicating articles in which the negative difference cumulative singular value sum diffΣΣD (j, ord, rank) is present in the upper 5%. That is, articles that are relatively important in the past are plotted in the zone de. Examples 7, 8 and 9 of articles corresponding to this zone de are shown in Table 3 below.

  Article 7 describes an event that occurred in France on December 1, 1978, and diffΣΣD was “−1984” (ave ΣΣD = 992). Important keywords in this article were "Pu", "Leakage", "Damage", and "Exposure". Article 8 describes an event that occurred in the United Kingdom on June 13, 1989, and diffΣΣD was “−1879” (ave ΣΣD = 940). Important keywords in this article were "Pu", "Leakage", "Piping", and "Tank". Article 9 describes an event that occurred in the United Kingdom on June 22, 1977, and diffΣΣD was “−1872” (aveΣΣD = 1052). Important keywords in this article were "Pu", "Leakage", "Rise", and "Abnormal".

  As shown in FIGS. 24 and 25, a graph showing the average cumulative singular value sum aveΣΣD (j, ord, rank) on one of the two axes and the difference cumulative singular value sum diffΣΣD (j, ord, rank) on the other axis is shown. By displaying, it is possible to easily find a representative article or case that represents the entire corpus (the average cumulative singular value sum aveΣΣD is large). Further, the feature in the order reference in which articles in the corpus are arranged can be grasped more clearly by the magnitude of the difference accumulated singular value sum diffΣΣD (j, ord, rank). In the above example, it is possible to easily find articles characteristic in the old age of occurrence of nuclear-related incidents, articles characteristic in the new age of occurrence.

  Further, in the above-described embodiment of the present invention, a survey form or questionnaire including a description in a free answer format is assumed as a corpus. The present invention can be applied to arbitrary corpora such as news, web news, blogs, newspapers, magazines, interview records, statements, questionnaires, and novels.

DESCRIPTION OF SYMBOLS 10 ... Document analysis apparatus 12 ... Network 14 ... Computer 16 ... Text database 18 ... Analysis database 30 ... Image scanner 32 ... Survey form 40 ... GUI

Claims (10)

A document analysis apparatus for analyzing a language material that increases in accordance with an order standard based on an increased TFIDF, between an estimated value based on a cumulative value of the increased TFIDF up to the previous corpus and a cumulative value of the increased TFIDF in the current corpus In a document analysis apparatus that obtains a singular value for each morpheme by performing residual analysis in
An ascending cumulative singular value calculating means for calculating an ascending cumulative singular value for each morpheme when the language material is arranged in ascending order in the order reference;
A descending-order cumulative singular value calculating means for calculating a descending-order cumulative singular value for each morpheme when the language material is arranged in descending order in the order criterion; and an average cumulative singular value by averaging the ascending-order cumulative singular value and the descending-order cumulative singular value A document analysis apparatus comprising an average cumulative singular value calculation means for calculating a value.   The document analysis apparatus according to claim 1, further comprising: a cumulative singular value graph display unit that displays a cumulative singular value graph in which one of the ascending cumulative singular value and the descending cumulative singular value is one of two axes and the other is the other. An ascending cumulative singular value sum calculating means for calculating a sum of ascending cumulative singular values for a morpheme having a specific upper cumulative singular value;
A descending cumulative singular value sum calculating means for calculating a sum of descending cumulative singular values for a morpheme having a specific upper cumulative singular value, and an average cumulative singular value sum by averaging the ascending cumulative singular value sum and the descending cumulative singular value sum The document analysis apparatus according to claim 1, further comprising an average cumulative singular value summation calculating means for calculating.   The document analysis apparatus according to claim 3, further comprising cumulative singular value sum difference calculation means for calculating a difference between the ascending cumulative singular value sum and the descending cumulative singular value sum.   The document analysis apparatus according to claim 4, further comprising a cumulative singular value sum graph display means for displaying a cumulative singular value sum graph in which the average cumulative singular value sum and the cumulative singular value sum difference are one and the other of two axes. In a document analysis apparatus that analyzes a language material that increases in accordance with an ordering standard based on an increased TFIDF, there is a residual between an estimated value based on the total value of the increased TFIDF up to the previous corpus and a total value of the increased TFIDF in the current corpus. a Rubun manual analysis method obtains the singular values of each morpheme by performing a differential analysis, computer of the document analysis device,
An ascending cumulative singular value calculation step of calculating an ascending cumulative singular value for each morpheme when the language material is arranged in ascending order in the order criterion;
A descending order cumulative singular value calculation step for calculating a descending order cumulative singular value for each morpheme when the linguistic materials are arranged in descending order in the order criterion; and an arithmetic average of the ascending order cumulative singular value and the descending order cumulative singular value A document analysis method for executing an average cumulative singular value calculation step of calculating a cumulative singular value. A document analysis apparatus for analyzing a language material that increases in accordance with an order criterion based on an increased TFIDF, between an estimated value based on the total value of the increased TFIDF up to the previous corpus and a total value of the increased TFIDF in the current corpus In the computer of the document analysis device that calculates the singular value for each morpheme by performing residual analysis in
An ascending cumulative singular value calculation step of calculating an ascending cumulative singular value for each morpheme when the language material is arranged in ascending order in the order criterion;
A descending order cumulative singular value calculation step for calculating a descending order cumulative singular value for each morpheme when the linguistic materials are arranged in descending order in the order criterion; and an arithmetic average of the ascending order cumulative singular value and the descending order cumulative singular value A document analysis program for executing an average cumulative singular value calculation step of calculating a cumulative singular value. A document analysis apparatus for analyzing a language material that increases in accordance with an order standard based on an increased TFIDF, between an estimated value based on a cumulative value of the increased TFIDF up to the previous corpus and a cumulative value of the increased TFIDF in the current corpus In a document analysis apparatus that obtains a singular value for each morpheme by performing residual analysis in
An ascending cumulative singular value calculating means for calculating an ascending cumulative singular value for each morpheme when the language material is arranged in ascending order in the order reference;
A descending order cumulative singular value calculating means for calculating a descending order cumulative singular value for each morpheme when the language material is arranged in descending order in the order reference;
An ascending cumulative singular value sum calculating means for calculating a sum of ascending cumulative singular values for a morpheme having a specific upper cumulative singular value;
A descending cumulative singular value sum calculating means for calculating a sum of descending cumulative singular values for a morpheme having a specific upper cumulative singular value, and an average cumulative singular value sum by averaging the ascending cumulative singular value sum and the descending cumulative singular value sum A document analysis apparatus comprising an average cumulative singular value summation calculating means for calculating Remaining between the document analysis device for analyzing based corpus that increasing the order criteria increasing type TFIDF, the cumulative value of the increase-type TFIDF in the estimate and the current corpus based on the cumulative value of the increase-type TFIDF and before Corpus A document analysis method for obtaining a singular value for each morpheme by executing a difference analysis , wherein the computer of the document analysis device comprises:
An ascending cumulative singular value calculation step of calculating an ascending cumulative singular value for each morpheme when the language material is arranged in ascending order in the order criterion;
A descending order cumulative singular value calculation step of calculating a descending order cumulative singular value for each morpheme when the language material is arranged in descending order in the order reference;
An ascending cumulative singular value sum calculating step for calculating the sum of ascending cumulative singular values for morphemes having a specific upper cumulative singular value;
A descending cumulative singular value sum calculating step for calculating the sum of descending cumulative singular values for a morpheme having a specific upper cumulative singular value; and averaging cumulative singular value sums by averaging the ascending cumulative singular value sum and the descending cumulative singular value sum A document analysis method for executing an average cumulative singular value summation calculating step for calculating. A document analysis apparatus for analyzing a language material that increases in accordance with an order criterion based on an increased TFIDF, between an estimated value based on the total value of the increased TFIDF up to the previous corpus and a total value of the increased TFIDF in the current corpus In the computer of the document analysis device that calculates the singular value for each morpheme by performing residual analysis in
An ascending cumulative singular value calculation step of calculating an ascending cumulative singular value for each morpheme when the language material is arranged in ascending order in the order criterion;
A descending order cumulative singular value calculation step of calculating a descending order cumulative singular value for each morpheme when the language material is arranged in descending order in the order standard;
An ascending cumulative singular value sum calculating step for calculating the sum of ascending cumulative singular values for morphemes having a specific upper cumulative singular value;
A descending cumulative singular value sum calculating step for calculating the sum of descending cumulative singular values for a morpheme having a specific upper cumulative singular value; and averaging cumulative singular value sums by averaging the ascending cumulative singular value sum and the descending cumulative singular value sum A document analysis program characterized by causing an average cumulative singular value summation calculating step to calculate.