JP2005322239A - Natural sentence processor, natural sentence processing method and natural sentence processing program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To accurately determine whether opinions are affirmative or negative from free description type questionnaire answer sentences or the like. <P>SOLUTION: A natural sentence processor having a determination knowledge data base which stores information about a pair of words in association with whether the information about the pair of words is an affirmative expression or a negative expression, a syntax analysis means for extracting a pair of words in electronic text data from coming electronic text data and an opinion determination part which determines whether words in questionnaire information are affirmative opinions or negative opinions by referring to the determination knowledge data base based on the information about the pair of words extracted by the syntax analysis means is provided. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to a natural sentence processing device, a natural language processing method, and a natural sentence processing program written in a free-description format. For example, when a questionnaire is conducted in a natural language when a questionnaire is conducted during a market research or the like. The present invention relates to a document processing apparatus, a natural language processing method, and a document processing program.

  A questionnaire is conducted as one of means for conducting various surveys. In the questionnaire, there are a free description type questionnaire method in which some or all of the responses are free description type documents, and several options regarding the question are prepared in the questionnaire survey form in advance, and the questionnaire respondent is selected from among them There is a selection method.

  The free description type questionnaire method has a problem that the answer result is a natural sentence, and manual counting and analysis are required to obtain useful information from a large number of respondents, which is very expensive.

  As a means for solving this problem, there is the selection method. However, in the case of this method, there is a problem that information relating to options other than those created in advance cannot be obtained, and important knowledge such as needs that the questionnaire implementer did not expect cannot be obtained.

  For the purpose of reducing the cost at the time of aggregation of this free description questionnaire, a questionnaire aggregation system that performs automatic aggregation by a computer at the time of aggregation has been proposed (for example, see Patent Document 1).

In this questionnaire totalization system, a thesaurus dictionary is provided in the device, the thesaurus dictionary is searched in units of words in the free description sentence, and the results are integrated with the broader terms. For example, it is a device that performs processing such as counting “hard baseball” and “soft baseball” as the broader term “baseball”.
JP 09-190454 A

  However, since this integration is performed in units of words, for example, “hard baseball.” And “soft baseball.” Are answers to the question item “What is lifelong learning?” In the questionnaire given in the description of this device. It is possible to correctly count sentences written only in words as "baseball", but "I thought it was baseball when I was young, but now I think golf is better" There is a problem that accurate tabulation cannot be performed when it is described as a negative object such as.

  There are also questions that make accurate tabulation almost impossible with this method. For example, for questions such as “Please give us your opinion about this product”, the questionnaire writer does not answer only with words, and whether the description is negative or positive From the point of view, it is necessary to distinguish at the time of aggregation, but this cannot be done with this apparatus.

  In other words, in this device, a free description questionnaire in its original meaning, that is, any question sentence is allowed on the questionnaire implementer side, and any description is allowed on the questionnaire description side, is accurate. Cannot be counted.

  In addition to questionnaires, examples of how many people express their opinions include personal computer communications, articles on bulletin boards on web pages, and customer inquiries sent to help desks. Opinions are described in natural sentences, and there is a problem that it takes the same cost as analysis of free description questionnaires to obtain useful information from them.

  Therefore, an object of the present invention is to solve the problems in the free description questionnaire as described above, and to obtain useful information accurately at low cost.

  In the present invention, not only a free description questionnaire, but useful information is obtained at low cost and accurately from an information source in which each person's opinion is described in the form of a natural sentence.

  FIG. 1 shows the principle configuration for achieving the object of the present invention. In FIG. 1, 1 is an input unit, 2 is an answer extraction unit, 3 is an opinion extraction unit, 4 is an opinion aggregation unit, 5 is a result presentation unit, 6 is an opinion determination unit, 7 is a determination knowledge acquisition unit, and 8 is total knowledge. A database, 9 is a judgment knowledge database, and 10 is a tabulation result storage database.

  The said subject of this invention can be achieved by the structure of following (1)-(5).

(1) a judgment knowledge database that stores word pair information in association with whether the word pair information is a positive expression or a negative expression;
A syntax analysis means for extracting word pairs in the electronic text data from the incoming electronic text data;
Based on the word pair information extracted by the syntax analysis means, an opinion determination unit that determines whether the opinion is a positive opinion or a negative opinion with respect to the words in the questionnaire information with reference to the determination knowledge database;
The natural sentence processing apparatus characterized by having.

(2) having a storage means for storing the counting results;
The syntax analysis means extracts free description parts of a plurality of questionnaire information as electronic text data,
The syntax analysis means performs syntax analysis of each questionnaire information extracted as electronic text data by the extraction unit, extracts word pairs,
Furthermore, the opinion totaling unit that aggregates the determination results of the plurality of questionnaire information determined by the opinion determination unit and accumulates the totalization results in the storage unit;
The natural sentence processing apparatus according to (1), characterized in that:

  (3) The natural sentence processing apparatus according to (1) or (2), wherein the syntax analysis unit extracts word pairs based on information obtained by converting electronic text data into a syntax tree.

(4) A computer that can access a judgment knowledge database that stores word pair information in association with whether the word pair information is a positive expression or a negative expression,
A syntax analysis means for extracting word pairs in the electronic text data from the incoming electronic text data;
Based on the word pair information extracted by the syntax analysis means, an opinion determination unit that determines whether the opinion is a positive opinion or a negative opinion with respect to the words in the questionnaire information with reference to the determination knowledge database;
Natural language processing program that makes it work.

(5) A computer that can access a judgment knowledge database that stores word pair information in association with whether the word pair information is a positive expression or a negative expression.
A parsing step for extracting word pairs in the electronic text data from the incoming electronic text data;
Based on the word pair information extracted by the syntax analysis step, an opinion determination step for determining whether the opinion in the questionnaire information is a positive opinion or a negative opinion with reference to the determination knowledge database;
Natural language processing method characterized by executing

  Thereby, the following effects can be obtained.

  (1) Since word pairs such as the opinion provider's assertions, opinions, and impressions can be automatically extracted from the response text of the free description parts submitted by multiple opinion providers, many free description parts It is possible to extract opinions accurately and in a short time from the answer sentence.

  (2) By searching a judgment knowledge database that stores knowledge for judging whether the extracted opinion gives a positive evaluation or negative evaluation regarding the object , Accurate and automatic determination.

  (3) When both a positive opinion and a negative opinion are obtained from a single answer sentence, the strength of the evaluation can be compared to automatically obtain a comprehensive and detailed judgment result in the answer sentence. .

  (4) Since the presentation information can be sequentially displayed in a plurality of graphs in which the entire set is changed, the analysis result can be presented in an easy-to-understand manner.

  (5) Since the knowledge for judging the extracted opinion is automatically acquired from the answer result in the opinion questionnaire of a specific format, it is possible to make an opinion judgment without preparing the knowledge for judgment in advance. it can. Moreover, accurate results can be obtained as learning is performed.

  (6) Since knowledge for judging the extracted opinion can be automatically acquired without using a specific form opinion survey form, it is also possible to prepare knowledge for judgment in advance. , Opinion judgment can be performed. Moreover, accurate results can be obtained as learning is performed.

  (7) An answer extracting means for extracting the electronic data of the free description part of the opinion provider from the provided opinions as an answer sentence for providing the opinion to the computer, an opinion extracting means for extracting a word pair from the extracted answer sentence, By installing a program that functions as an opinion aggregation means that aggregates the extracted opinion contents and a result providing means that provides the aggregation results for each opinion contents, a normal computer was provided by multiple opinion providers. A natural sentence processing function for analyzing opinions can be performed.

  An embodiment of the present invention will be described with reference to an analysis of questionnaire responses based on FIGS. 1 is an embodiment of the present invention, FIG. 2 is an explanatory view of opinion extraction, FIG. 3 is an explanatory view of an opinion summary result showing an example of an opinion summary result in the present invention, and FIG. 4 is an aggregate showing an example of an aggregate result in the present invention. FIG. 5 is an explanatory diagram of results, FIG. 5 is a graph state explanatory diagram of the present invention, and FIG. 6 is an example of an answer sentence analysis of the present invention.

  In the figure, 1 is an input unit, 2 is an answer extraction unit, 3 is an opinion extraction unit, 4 is an opinion aggregation unit, 5 is a result presentation unit, 6 is an opinion determination unit, 7 is a determination knowledge acquisition unit, and is a first determination. Comprising knowledge acquisition unit 7-1 and second determination knowledge acquisition unit 7-2, 8 is a total knowledge database, 9 is a determination knowledge database, 10 is a total result storage database, 20 is a processor, 21 is a display unit, A Is a questionnaire respondent and B is a questionnaire analyst.

  The input unit 1 performs a process in which the questionnaire respondent A inputs each answer on a questionnaire sheet and collects this answer in the apparatus. The process of collecting this response includes, for example, distributing a questionnaire on paper media directly to the survey target person, or distributing it by fax etc. This can be achieved by distributing survey forms, filling in responses and collecting them using a web bulletin board.

  In the case of collecting responses to questionnaires or collecting responses by fax, the response entry field is fixed so that it can be detected by the coordinate position of the paper. In the process of collecting such answers, conventional means for collecting questionnaire results can be used as they are.

  Note that the input unit 1 is not limited to the one that handles character information as described above, and it is also possible to collect the subject's answers in the form of voice data by telephone response, etc. Data on the web using the Internet may be used, and any information transmission means can be used as a response medium. If a translation means is provided, it can be entered in a foreign language.

  The answer extraction unit 2 converts the free description part of each collected questionnaire answer result into an electronic text and extracts it as an answer sentence. This process includes, for example, a method in which an operator on the questionnaire operator side creates a digitized text by means of digitized data input means such as a keyboard, and an automatic digitizing means suitable for the answer medium, For example, it is realized by a method of creating digitized text without using a character recognition device or a speech recognition device.

  Depending on the means adopted for the questionnaire input unit 1, the collected responses may already be digitized. In such a case, the digitized free description portion is sent to the opinion extraction unit 3 as it is. can do.

  The opinion extraction unit 3 analyzes the digitized response sentence created by the answer extraction unit 2, and performs processing to extract information that is so-called in the answer sentence as an opinion.

  The opinions extracted here include, for example, a product questionnaire on a notebook computer, which is included in an answer sentence “the body is small but the screen is large”, “the body is small”, “the screen is large” This is the information that the surveyor wants to know when collecting the questionnaires.

  In this opinion extraction process, as shown in FIG. 2, the answer sentence is converted into a syntax tree using the syntax analysis means 3-2 to extract the subject and the predicate, and the binary relations “body-small”, “screen” This is achieved by setting “large” as the total processing unit.

  Note that the information extracted as an opinion is not limited to the above-mentioned “subject-predicate” pair, but for example, “screen-easy to see” is extracted from “easy-to-view screen” from the “modified subject-modifier” relationship, etc. Various things are possible.

  In addition to the binary relations as described above, it is also possible to use the relation of three or more terms in the syntax tree, for example, "subject-object-predicate" as an opinion, and further interpret the relation by the extracted notation. It is also possible to convert it into a semantic structure by means 3-1 and make it an opinion.

  For example, “# screen → # beautiful”. Here, # is an operator representing the meaning of the word behind it, not only the screen that appears in the text, but for example, the whole synonym of the screen in the field, monitor, screen, screen, The entire liquid crystal is shown. In addition, “→” is an operator representing an inference result, and in this example, indicates that “the screen is beautiful”. These processes are all performed using a computer and can be realized without manpower.

  The opinion totaling unit 4 performs processing for totalizing the appearance frequency for each opinion extracted by the opinion extracting unit 3, for example, binary relations such as “main body-small” and “screen-large”. Thereby, for example, as shown in FIG. 3, the appearance frequency for each binary relation can be totaled.

  As an aggregation method, for example, when the opinions are expressed in the binary relation as described above, in addition to counting the number for each opinion, the aggregation is based only on the first item (or “body” if “body-small”). By performing the aggregation only by the second term ("small" in the above example) and sending the aggregation results to the result presentation unit 5, it is possible to present information that matches the analysis purpose of the analyst B. .

  In addition, a tabulation method with a function of tabulating the result of absorbing the fluctuation of the expression of the opinion can be used. For example, regarding the opinion extracted from a negative sentence, “main body—not big”, the opinion “main body—not big” is counted as “main body—small” as the same opinion.

  For example, when the aggregate knowledge database 8 including an antonym database indicating an antonym expression of an antonym is connected to the opinion tabulation unit 4 and the second term of the opinion includes a denial, for example, an affirmative expression “not large” is “large”. This antonym database is searched, and the antonym “small” is acquired and combined with the first term to be used for aggregation as “main body-small”.

  In the same manner, the total knowledge database 8 can be provided with a synonym database, and synonyms can be unified. In this process, for example, opinions such as “screen-beautiful” and “screen-beautiful” can be tabulated as “screen-beautiful”.

  Other thesauruses, real world knowledge, and the like can also be provided in the aggregate knowledge database 8. Real world knowledge is, for example, the opinion that “the previous one was good” is judged as “the current one is bad”, and the knowledge that should be extracted from the reply sentence that “the operability was good before” The operability is bad ".

  The result of performing the above processing is expressed as a table shown in FIG. 3, for example. FIG. 3 relates to a notebook personal computer. As opinions, “main body-small”, “screen-large”, “screen-small”... Are extracted, and the total number is 200, 50, 150. Yes, “main body-small” and “screen-large” indicate a favorable determination result, and “screen-small” indicates a negative determination result.

  As synonymous opinions of “main body-small”, “main body-compact”, “body-small”..., And as synonymous opinion of “screen-large”, “screen-wide”. As a synonymous opinion of “small”, “screen-narrow”, “screen-not big”,... Exist.

  The data format sent to the result presentation unit 5 may be converted into a format handled by various tabular software and sent in the form of a table as shown in FIG. 3, or may be sent in another format. Note that all the processes related to the opinion totaling unit can be realized by a computer.

  The result presentation unit 5 performs processing for presenting the total result created by the opinion totaling unit 4 to the questionnaire analyst B. As a method of presenting the aggregated results, it may be presented to the questionnaire analyst B in the form of a table as shown in FIG. 4 or, in addition, the analysis purpose of the questionnaire analyst B using various spreadsheet software functions You may create and present a table or graph that matches

  According to FIG. 4, it can also be seen that the respondent is interested in the target product being larger or smaller than the clean or dirty main body.

  As the presenting means, for example, it is necessary to be able to present a chart for monitoring to the questionnaire analyst on the display unit 21 shown in FIG. 1, but in the present invention, it is not always necessary to limit the chart information to the presenting means. As a result reading means, it may be converted into voice information using a speaker or the like and presented to a questionnaire analyst.

  When presenting as a graph, as shown in FIG. 5, the first chart is represented by a bar graph in which the length is the total number of opinions totaled in the first term. For example, as shown in FIG. 4, since the number of favorable opinions is 500 and the number of negative opinions is 20, as shown in FIG. 4, a bar graph indicating the total number of 520 is created, and positive evaluation opinion (500) and negative evaluation opinion ( 20) each of the ratios is presented in different colors. For example, if the questionnaire analyst designates the positive evaluation part with a pointing device such as a mouse, each ratio of the second item in the positive opinion is presented in a pie chart, and When an item in the graph is instructed by a pointing device, an answer sentence in which the opinion is extracted is displayed.

  In FIG. 5, when a positive evaluation portion of the bar graph relating to the main body is pointed with the pointing device, a pie chart divided into three is displayed, and when the area (a) is pointed, an answer sentence “small” is displayed. Further, when the area (b) is pointed, “Convenient” is displayed, and when the area (c) is pointed, “Light” is displayed.

  Furthermore, when both a positive opinion and a negative opinion are obtained from one answer sentence, the strengths of the evaluations are compared, and a comprehensive determination result in the answer sentence is estimated. For this reason, when these two opinions are given, each is given a score in advance and weighted.

  In this way, not only the result of the opinion unit, but also the evaluation of the response sentence unit can be aggregated, so it is color-coded by the proportion of the graph and presented to the questionnaire analyst first, for example the proportion of the positive opinion If a part is designated by a pointing device, the ratio of the opinions in the answer sentences giving a comprehensive evaluation can also be presented graphically.

  The opinion determination unit 6 performs a process of determining whether the opinion is a positive opinion or a negative opinion using the opinion determination knowledge stored in the determination knowledge database 9 for each opinion information extracted by the opinion extraction unit 3. Is.

  The opinion judgment knowledge stored in the judgment knowledge database 9 can be described by a ternary relationship such as “body-small-positive evaluation” when the opinion can be expressed by the binary relationship given as an example. Information. Further, considering that the object dependency in the judgment knowledge database 9 is eliminated, and the questionnaire regarding other objects is also used in the same database, information on the object is further added to “main body-small-positive evaluation, object: It may be described in terms of four terms such as “information equipment”.

  A judgment knowledge database storing such judgment knowledge is provided, and the judgment knowledge database is searched for each extracted opinion, for example, “main body-small”, and, for example, “main body-small-positive evaluation” is obtained. This process can be realized.

  Also, for example, by giving a score (weight) of +1 to a positive opinion and -1 to a negative opinion, the sum of the scores of a plurality of opinions obtained in one answer sentence can be comprehensively evaluated for one answer sentence. By considering it, it is possible to request the evaluation of the object in response sentence units.

  Furthermore, the determination result is not a binary value such as a positive evaluation / negative evaluation, but takes a multi-value, for example, by adding “strong positive” and “strong negative” to the evaluation to give a four-value to the strength of the evaluation. A more accurate determination can also be achieved by giving a corresponding score. It is also possible to express each evaluation value by an evaluation function and calculate a comprehensive evaluation by performing synthesis based on bundle value logic, so-called “fuzzy inference”.

  The determination knowledge acquisition unit 7 automatically acquires the determination knowledge without using human hands. The judgment knowledge used by the opinion judgment unit 6 is object-dependent knowledge, and its maintenance is costly. In order to solve this problem, it is only necessary to automatically obtain judgment knowledge from a questionnaire response sentence.

  As a first method, the knowledge for judging the extracted opinion is automatically acquired and accumulated from the questionnaire response result in the questionnaire questionnaire in the judgment format.

  In the initial stage of operation, sufficient judgment knowledge is not stored in the judgment knowledge database 9, so for each question in the questionnaire, for example, “Please fill in the points that you think are good about this product” The question is such that the evaluation of the answer result is clarified, the opinion is extracted from the answer sentence described therein, and the expected evaluation of the answer is combined and stored in the judgment knowledge database as judgment knowledge.

  In addition, when the attribute to be evaluated for the object is known in advance, the question is “Please select what you feel about the following items of this product from the following, and enter the reason why you feel it” Next, arrange the attributes to be evaluated such as “screen”, “keyboard”, etc., and enter the option entry item “very good-good-normal-bad-very bad” and the reason in the answer part. Two free description items are provided, and the results of digitizing the questionnaire results indicate that items such as “screen” and the terminology from the free description part, for example, “large and easy to see”, “large” and “easy to see” Extraction is made, an opinion is created based on the set, and judgment knowledge is created by combining the evaluation value of the selected portion, for example, good. For example, “screen-large-positive evaluation” and “screen-easy to read-positive evaluation” are created. By storing the knowledge acquired by such processing in the judgment knowledge database 9, it can be used in the subsequent processing. Each process is performed by the first determination knowledge acquisition unit 7-1 shown in FIG.

  As described above, as described above, the determination knowledge is acquired without setting a special question at the initial stage of operation as described above, and the second determination knowledge acquisition unit 7-2 shown in FIG. Is to do.

  This is because, for example, two opinions “screen-easy to see” and “screen-good” can be extracted from an answer sentence “screen is easy to see”, and one of the second terms represents an evaluation given in advance. In this example, since “good” represents an evaluation given in advance, this knowledge is expressed by a binary relation of “good-positive evaluation”. Then, by combining these two, it is possible to acquire the determination knowledge information “screen−easy to see−positive evaluation”.

  This method can not only extract judgment knowledge but also obtain words representing evaluations that are not given to the apparatus in the initial state of operation. For example, it is assumed that knowledge of “good—positive evaluation” is given to the apparatus in advance, and determination knowledge “screen—easy to see—positive evaluation” is acquired from the above example sentence.

  Next, two opinions such as “screen—easy to see” and “screen—great” are extracted from the answer sentence “screen is easy to see and great”. Knowing that there is new knowledge such as “Excellent-Positive Evaluation”. The knowledge learning function can be realized by storing in the judgment knowledge database 9 that “easy to see” and “great” are positive evaluations in this way and using them in subsequent processing.

  All of the processes described above can be processed by a computer without requiring manual operation. Note that the determination knowledge acquisition methods of the first method and the second method are not rejected, and a combination of both can be configured.

  The tabulated knowledge database 8 stores data necessary for tabulating answer sentences. For example, an antonym database, a synonym database, a thesaurus, real world knowledge, and the like are stored.

  The judgment knowledge database 9 stores data necessary for judging whether the answer sentence is a positive (favorable) opinion or a negative opinion. For example, part of speech, opinion judgment knowledge, “first item” “+ 2nd term + evaluation”, “1st term + second term + evaluation + target field”, points (weighting) for each opinion when a positive opinion and a negative opinion are mixed are stored.

  The totaling result storage database 10 temporarily stores the totaling results obtained by the present invention, and stores the totaling data shown in FIGS. 3 and 4, the display data shown in FIG.

  The processor 20 constitutes the natural processing apparatus of the present invention, and includes, for example, an answer extraction unit 2, an opinion extraction unit 3, an opinion totaling unit 4, a result presentation unit 5, an opinion determination unit 6, a determination knowledge acquisition unit 7, and the like. It has.

  The display unit 21 is for the questionnaire analyst B to display the processing result of the present invention and view the processing result, and is configured by a display, a printer, or the like.

  The operation of the embodiment of the present invention shown in FIG. 1 will be described according to the flowchart shown in FIG.

  S1. An answer to the questionnaire from the questionnaire respondent A is input to the input unit 1 by facsimile or electronic information, for example. The input unit 1 sends this answer result to the answer extraction unit 2.

  S2. The answer extracting unit 2 converts the answer result, particularly a free description part, into an electronic text and extracts it as an answer sentence. Then, this electronic response sentence is sent to the opinion extraction unit 3.

  S3. The opinion extraction unit 3 performs a morphological analysis on the digitized answer sentence, converts it into a syntax tree using the syntax analysis unit 3-2, creates a binary relation or a ternary relation, and the like. Perform extraction. Then, this is sent to the opinion totaling unit 4.

  S4. The opinion totaling unit 4 totals the appearance frequencies of the opinions extracted by the opinion extracting unit 3 for each binary relationship, for example. At this time, the opinion determination unit 6 determines whether the opinion extracted by the opinion extraction unit 3 is a positive opinion or a negative opinion using the opinion determination knowledge stored in the determination knowledge database.

  S5. The opinion totaling unit 4 totals the appearance frequency of the opinions extracted by the opinion extracting unit 3 in this way, determines whether the opinion extracted by the opinion extracting unit 3 via the opinion determining unit 6 is positive or negative, Further, the synonym database of the total knowledge database 8 is accessed to recognize synonymous opinions. Then, an opinion summary table is created as shown in FIG.

  S6. When the opinion total created in this way is sent to the result presenting unit 5, the result presenting unit 5 processes the processed result and presents the questionnaire result in an easy-to-analyze manner for the questionnaire analyst B. For example, as shown in FIG. 4, it may be presented in a table format, or a table or graph that matches the analysis purpose of a questionnaire analyst may be created and presented. Further, as shown in FIG. 5, when a pointing device is designated, the portion may be presented to be displayed in more detail.

  By the way, when the answer sentence shown in FIG. 6 is input, in the present invention, the opinion extraction unit 3 and the opinion determination unit 6 perform morphological analysis on the answer sentence, and the subject-predicate of the first half sentence is “baseball-I think. "," Recognizing that "ga" is a negative conjunction and recognizing "baseball-thinking-negative".

  Then, since the subject-predicate of the latter half sentence is “golf-good” and the evaluation of good is affirmative, “gokufu-good-affirmation” is recognized. In this way, the reply sentence in FIG. 6 obtains “not golf but baseball—golf”.

  In the above description, the present invention has been described with respect to an example of a questionnaire response sentence processing. However, the present invention is not limited to this.

  The answer extraction unit 2, the opinion extraction unit 3, the opinion totaling unit 4, the result presentation unit 5, the opinion determination unit 6, the determination knowledge acquisition unit 7, and the like can be configured and processed by a program, for example.

  The form of this invention is described in an appendix.

  (Supplementary note 1) In a natural sentence processing device that analyzes opinions provided by a plurality of opinion providers, the answer extraction is to extract the electronic data of the free description portion of the opinion provider from the provided opinions as a reply sentence for providing the opinions. Means, opinion extraction means for extracting word pairs from the extracted answer sentences, opinion totaling means for totaling each extracted opinion content, and result providing means for providing a total result for each opinion content Natural sentence processing device characterized by.

  (Additional remark 2) In the natural sentence processing apparatus of the said additional remark 1, the opinion determination means which determines whether the extracted opinion has given positive evaluation or negative evaluation, and for determining an opinion A natural sentence processing apparatus comprising a judgment knowledge database for storing knowledge.

  (Supplementary note 3) In the natural sentence processing device described in the supplementary note 2, when two affirmative opinions and negative opinions are obtained in one reply sentence, the strength is determined, and a comprehensive confirmation of the reply sentence is made. A natural sentence processing apparatus comprising an opinion determining means for determining negative.

  (Additional remark 4) The natural sentence processing apparatus of Additional remark 1 and Additional remark 2 WHEREIN: The presentation means which displays presentation information by the some graph which changed the whole set was provided, The natural sentence processing apparatus characterized by the above-mentioned.

  (Additional remark 5) In the natural sentence processing apparatus according to the above additional remarks 2 and 3, determination knowledge for automatically acquiring and accumulating knowledge for determining an opinion extracted in particular from an answer result in a specific form opinion survey form A natural sentence processing apparatus comprising an acquisition means.

  (Supplementary note 6) In the natural sentence processing apparatus according to supplementary notes 2 and 3, particularly, knowledge for judging the extracted opinion is acquired automatically from a specific expression pattern in the answer sentence and acquired. A natural sentence processing apparatus characterized by comprising means.

  (Supplementary note 7) In order to operate the computer as a natural sentence processing means for analyzing opinions provided by a plurality of opinion providers, the electronic data of the free description part of the opinion provider is provided from the provided opinions. Answer extracting means for extracting as answer sentences, opinion extracting means for extracting word pairs from the extracted answer sentences, opinion counting means for counting for each extracted opinion content, and result providing means for providing a total result for each opinion content Natural sentence processing program to function as

It is an embodiment of the present invention. It is opinion extraction explanatory drawing in this invention. It is an opinion total result explanatory drawing in the present invention. It is an aggregation result explanatory view in the present invention. It is a graph state explanatory view in the present invention. It is an example of an answer sentence analysis in the present invention. It is an operation state explanatory view of the present invention.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Input part 2 Answer extraction part 3 Opinion extraction part 4 Opinion extraction part 5 Result presentation part 6 Opinion judgment part 7 Determination knowledge acquisition part 8 Total knowledge database 9 Determination knowledge database 10 Total result storage database 20 Processor 21 Display part

Claims (5)

A judgment knowledge database that stores word pair information in association with whether the word pair information is a positive expression or a negative expression;
A syntax analysis means for extracting word pairs in the electronic text data from the incoming electronic text data;
Based on the word pair information extracted by the syntax analysis means, an opinion determination unit that determines whether the opinion is a positive opinion or a negative opinion with respect to the words in the questionnaire information with reference to the determination knowledge database;
The natural sentence processing apparatus characterized by having. It has a storage means to store the total results,
The syntax analysis means extracts free description parts of a plurality of questionnaire information as electronic text data,
The syntax analysis means performs syntax analysis of each questionnaire information extracted as electronic text data by the extraction unit, extracts word pairs,
Furthermore, the opinion totaling unit that aggregates the determination results of the plurality of questionnaire information determined by the opinion determination unit and accumulates the totalization results in the storage unit;
The natural sentence processing apparatus according to claim 1, comprising:   The natural sentence processing apparatus according to claim 1, wherein the syntax analysis unit extracts a word pair based on information obtained by converting electronic text data into a syntax tree. A computer capable of accessing a judgment knowledge database that stores word pair information in association with whether the word pair information is a positive expression or a negative expression,
A syntax analysis means for extracting word pairs in the electronic text data from the incoming electronic text data;
Based on the word pair information extracted by the syntax analysis means, an opinion determination unit that determines whether the opinion is a positive opinion or a negative opinion with respect to the words in the questionnaire information with reference to the determination knowledge database;
Natural language processing program that makes it work. A computer that can access a judgment knowledge database that stores word pair information in association with whether the word pair information is a positive expression or a negative expression,
A parsing step for extracting word pairs in the electronic text data from the incoming electronic text data;
Based on the word pair information extracted by the parsing step, an opinion determination step for determining whether the opinion in the questionnaire information is a positive opinion or a negative opinion with reference to the determination knowledge database;
Natural language processing method characterized by executing

Images