JP5369813B2 - Persuasion support device, persuasion support program, and persuasion support method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a persuasion support device, a persuasion support program and a persuasion support method, for giving support to persuade a person who opposes a case using a document. <P>SOLUTION: The persuasion support device 10 has: a refutation information storage part 33 storing refutation information obtained by associating a reason sentence of an adverse opinion to the case, a ground sentence for the case, and presence/absence of refutation of the reason sentence and the ground sentence; a characteristic word extraction part 35 extracting characteristic words of the reason sentence and the ground sentence from a morpheme analysis result; a clustering part 36 generating cluster information including the ground sentence, rewriting the cluster information such that the reason sentence is arranged in a cluster based on similarity between the characteristic words of the ground sentence and the reason sentence, and rewriting the cluster information such that the cluster allocated with the reason sentence is rearranged based on the refutation information; and an output processing part 37 outputting the ground sentence included in the same cluster as the reason sentence as a ground sentence for re-refutation to the reason sentence based on the cluster information. <P>COPYRIGHT: (C)2011,JPO&amp;INPIT

Description

  The present invention relates to a persuasion support device, a persuasion support program, and a persuasion support method, and more particularly, to a persuasion support device, a persuasion support program, and a persuasion support method that support persuasion in writing.

  Several devices for supporting general discussions and meetings have been disclosed in the past (see, for example, Patent Document 1). These devices are mainly designed to improve the efficiency of communication that takes place in discussions, and to improve the efficiency of the user interface and to structure the agenda data and group the statements by issue.

  For example, in a system that supports general discussions and conferences, participants of the discussion proceed with the discussion while attaching the type (agreement or disagreement) and information to the document. In the system, the state of the discussion is visualized and shown to the participants, and the presenter gathers opinions and draws a conclusion. Also, in order to draw conclusions in the system, the main assertion and the logical structure of the assertion supporting it (the reason for the claim, the basis, etc.) are clarified, and the validity is discussed and scored. The system also calculates the score of the main claim from the score of the lower opinion (s) and the logical structure.

  In addition, there is a methodology that is used in debate, etc., when developing a strategy for discussion in terms of technology to convince the other party. A debate is a competition in which an agenda is divided into an affirmative side and a negative side for the sake of convenience, each argument is demonstrated, and the winner or defeat is judged by the audience or judges. Discussion analysis and strategy development can be done by bringing a typical structure into the content of the argument in the debate.

  The most well-known methodologies used to develop a discussion strategy are those that use Toulmin's discussion model. In the analysis using Toulmin's argument model, first, “three elements of argument” of claim, rationale, and ground are extracted from the content of argument as the skeleton of argument.

  In the analysis using Toulmin's argument model, the argument model composed of three elements and several auxiliary elements is used to reinforce the argument / foundation of one's claim and the argument of the opponent's argument. • Searching for weaknesses in the grounds is often done as part of the discussion strategy formulation.

  However, analysis using Toulmin's argument model is at the level of methodology. In order to perform an analysis using the Toulmin discussion model, the work must be done manually.

  As an example that needs to persuade the other party of the discussion, for example, the act of collecting the opinions of a large number of people for some claims and measures (hereinafter simply referred to as measures) and persuading the opponent in the collected opinions is there. In a situation where many people's opinions are collected on some measure and the opponents are persuaded, the measure implementer may prepare an argument supporting the measure and explain it to the interested parties of the measure. Often done. Stakeholders have the opportunity to comment on the content of the explanation. In a situation where many people's opinions are collected for a certain measure and the opponents are persuaded, discussions between the two parties may start from the opportunity for interested parties to express their approval or disapproval of the explanation.

  Persuasion of stakeholders through explanations and discussions of measures is often required between government agencies and citizens, business organizations and members. Usually, the persuasion of interested parties through explanations and discussions of measures is promoted by human efforts on the part of the implementation as a coordinated act for realizing the measures.

Japanese Patent Application Laid-Open No. 2004-200741

  However, when there are a large number of stakeholders in a measure, it is generally difficult to fully persuade the stakeholders. A typical example is public comment (public opinion solicitation procedure) conducted by government agencies.

  Public comments sponsored by government agencies in governments and local governments ask people for opinions on the content before the government decides on orders and rules. Government agencies are obliged to express their views on the opinions received. However, it is not usually required to answer all opinions.

  For public announcements of public comment results sponsored by government agencies of national and local governments, summarize representative opinions, pick up a few to tens of them, and attach a document that shows the government's “thinking” about them. It is done in the form. In addition, in public comments sponsored by government agencies of national and local governments, it is not essential to refute each objection individually. Here, re-argument refers to an act in which the measure implementation side collects opinions of a large number of people for a certain measure and persuades the opponents.

  By the way, public comments sponsored by government agencies in the country and local governments can receive tens of thousands of opinions on many oppositions. In such a case, publicizing the result of a public comment that summarizes representative opinions, picks up a few to tens of them, and attaches a document indicating the “thinking” of the government office to them is a lot of work. It becomes difficult to do it manually.

  Therefore, it is difficult to treat each objection individually when it is necessary to assemble a rebuttal against many objections, in order to persuade many objections, such as public comments hosted by the state or local government. . Note that in a conventional system for supporting general discussions and meetings, one of the participants in the discussion cannot support the act of persuading the other participant.

  One embodiment of the present invention has been made in view of the above points, and an object of the present invention is to provide a persuasion support device, a persuasion support program, and a persuasion support method that support written persuasion to an opponent to an assertion.

In order to solve the above-mentioned problem, an embodiment of the present invention is a persuasion support device that supports written persuasion to an opponent of an assertion, the reasoning statement of an objection to the claim, the rationale of the claim, From the result of the morphological analysis of the reason sentence and the basis sentence of the rumination information storage section, the repeatability information storage unit that stores the repeatability information associated with the reason sentence and the presence or absence of the rumination of the basis sentence A feature word extraction unit for extracting a feature word included in the reason sentence and the basis sentence, and cluster information including only the basis sentence as an initial arrangement of the cluster is generated, and the feature word and the reason sentence included in the basis sentence are generated. Based on the similarity to the included feature word, clustering is performed to rewrite the cluster information so that the reason sentence is placed in the cluster, and then stored in the repulsive information storage unit A clustering unit that rewrites the cluster information so as to relocate the cluster to which the reason sentence is allocated based on repetitiveness information, and the same as the reason sentence as a ground sentence for re-argument based on the cluster information An output processing unit that outputs the basis sentence included in the cluster.

  In addition, what applied the component, the expression, or arbitrary combinations of the component of one Embodiment of this invention to a method, an apparatus, a system, a computer program, a recording medium, a data structure, etc. is also effective as an aspect of this invention. .

  As described above, according to one embodiment of the present invention, written persuasion to the opponent of the claim can be supported.

It is the schematic diagram showing the process of persuasion which a present Example presupposes. It is explanatory drawing of an example showing the attribute of the argument arrangement table. It is explanatory drawing of an example showing the definition of the basis classification. It is a block diagram of an example showing the specific example of the argument arrangement | positioning table | surface. It is a block diagram of an example showing the specific example of the opinion arrangement table. It is a schematic diagram of an example of clustering centering on two ground sentences. It is a block diagram of an example of rumination information. It is a schematic diagram of an example of the clustering using rumination information. It is explanatory drawing showing an example of the ground sentence and the reason sentence. It is a hardware block diagram of an example of a persuasion assistance apparatus. It is a block block diagram of an example of a persuasion assistance apparatus. It is a flowchart which shows an example of the process sequence of a persuasion assistance apparatus. It is a block diagram of an example of feature vector information. It is a block diagram of an example of re-disposal support information. It is a block diagram of the other example of re-disposal support information. It is the flowchart showing the processing procedure of the clustering of the reason sentence which made the ground sentence a nucleus. It is a block diagram of an example showing the initial arrangement of cluster information. Is a block diagram showing an example of a feature vector of the representative point g _k. It is a block diagram of an example showing the final state of cluster information. It is an image figure of an example of a table type interface for rumination information input. It is a flowchart showing the clustering processing procedure using repetitive information in addition to the clustering of reason sentences with the basis sentence as the core. It is a block diagram of an example of rumination information. It is explanatory drawing showing an example of the ground sentence and the reason sentence. It is a block diagram of an example showing the initial arrangement of cluster information. Is a block diagram showing an example of feature vectors of the representative points g _k. It is a block diagram of an example showing the final state of cluster information. It is an image figure of an example of re-disposal support information. It is a comparison figure of an example of the result of clustering which does not use rumination information, and the result of clustering which uses rumination information. It is the schematic diagram which compared the result of the clustering which does not use rumination information, and the result of clustering which uses rumination information.

  Next, modes for carrying out the present invention will be described based on the following embodiments with reference to the drawings.

  When it is necessary to compile a rebuttal against many objections, such as public comments hosted by a national or local government, the objections are organized into groups with similarities, and each objection is reconsidered. That is realistic. For re-argument, the logic of doctrine prepared in advance can be used. However, in general, it is necessary to determine how a combination of a plurality of doctrine logics and an objection group that summarizes dissents with similarities can be persuasive.

  The system that supports such persuasion has the following technical problems. The first technical problem is a method of grouping dissent opinions based on similarity. The second technical issue is how to select a doctrine suitable for re-argument for groups of similar objections.

  As a method of grouping document data such as human opinions by similarity, the (dis) similarity between texts is quantified as the distance between document data, and grouping is performed based on the distance between the document data. The clustering technique to perform is widely known. There are various methods for clustering document data (text data), including a vector space method for defining similarity on a space based on word frequency in a document. It's being used.

  In this embodiment, a clustering technique based on the similarity between documents is used to solve the first and second technical problems. In order to solve the first and second technical problems, it is important to select sentence data to be clustered and to control the calculation process of clustering itself.

  Further, in this embodiment, when supporting persuasion to the opponent, it is important to implement a logical re-argument, construct an efficient re-argument, and comprehensive consideration of the objection. That is, in this embodiment, a typical use scene is a case where it is necessary to carry out the re-argument performed by the measure implementation side with respect to a number of opposing opinions in a logical, efficient and exhaustive manner.

(Premise of this example)
The present embodiment provides a persuasion support system or a persuasion support device that supports the persuasion performed by the policy implementation side (the argument side) to the applied side (the counter side). Hereinafter, a persuasion support device will be described as an example. The persuasion support apparatus is premised on that both pieces of opinion information are given as inputs having a predetermined logical structure.

  There are two types of information in the input given to the persuasion support device, which are called “argument summary table” and “opinion summary table”, respectively. These pieces of information are used in the present embodiment as an idea of exchanging opinions that is structured to some extent.

  FIG. 1 is a schematic diagram showing a persuasion process assumed in this embodiment. This embodiment is premised on the process of discussion, opinion solicitation (refutation), and re-refutation as shown in FIG.

  The persuasion process of FIG. 1 consists of five stages, steps S1 to S5. In step S1, the discussion side creates a discussion point arrangement table 2 based on the document 1, such as the minutes and reports. In step S 2, the discussion side creates a question sentence 3 regarding the basis of the measure based on the argument arrangement table 2. In step S3, the objection side answers the question sentence 3 with the answer sentence 4. In step S4, the discussion side summarizes the answer sentences in the opinion summary table 5. In step S5, the discussion side uses the argument arrangement table 2 and the opinion arrangement table 5 to create the re-rebuttal paper 6 and re-argues.

  Steps S <b> 1 to S <b> 4 are preparatory steps, and are procedures for preparing the argument arrangement table 2 and the opinion arrangement table 5 that are given as inputs to the persuasion support device of the present embodiment. Step S5 is a persuasion support stage. First, the procedures created in steps S1 to S4 and the information created in steps S1 to 5 will be described.

  In step S1, the discussion side creates an argument arrangement table 2 that logically arranges its claims from information such as the minutes and reports of discussions made in the past. The argument arrangement table 2 is information having attributes as shown in FIG. FIG. 2 is an explanatory diagram showing an example of attributes of the discussion point arrangement table. The argument arrangement table 2 in FIG. 2 defines the basis type as shown in FIG. 3 as one of the attributes. FIG. 3 is an explanatory diagram of an example showing the definition of the basis type.

  FIG. 4 is a configuration diagram of an example showing a specific example of the argument arrangement table. Here, we refer to an example of the procedure for public comment conducted by the Ministry of Health, Labor and Welfare, “Recruitment of opinions and information on changes to the basic plan for bovine spongiform encephalopathy”. In general, decisions on major measures made by government agencies are often discussed in a council composed of experts in the field to which the measure belongs.

  In this case, the content of the discussion held by the council provides the basis for the measure. The content of the discussions held by the council is documented as minutes. The results of the discussions held by the council will be documented as a report.

  The Ministry of Health, Labor and Welfare has been building discussions centered on the Food Safety Committee, which is an advisory body, to deal with the BSE (Bovine Spongiform Encephalopathy: also called Mad Cow Disease) problem including this measure. Based on the discussion, the Ministry of Health, Labor and Welfare has organized some logic and the grounds for supporting it, and has accumulated internal logic. The discussion point table 2 in FIG. 4 is obtained by extracting the main claims from the public materials of this case.

  There is usually an assertion behind the measure that it should be done. In the discussion table 2 in FIG. 4, “the BSE test target is limited to dead cattle 24 months of age or more” in the lower left is an example of the assertion. The list of argument points 2 provides multiple grounds and rationales for this claim. When seeking external opinions, not only presenting the measure itself but asking for opinions, but also clarifying the logic on the argument side as shown in the discussion table 2 in FIG. If you use the format, you will have a clearer point of discussion and less discussion.

Therefore, it is assumed here that the questionnaire has been conducted with the following questions.
Question Q1 “The Ministry of Health, Labor and Welfare has been able to detect in about 30 months in the prion administration experiment in the UK. I have decided to limit the subject to dead cattle 24 months of age or older.

  In step S3 of FIG. 1, the side to which the measure is applied answers the question sentence Q1 using the reply sentence 4. Specific questions include the pros and cons of the argument and the reasons. There are no particular restrictions on the method of collecting responses, but considering the automation of the processing from step S4 onward, it is conceivable to collect the responses electronically using a Web server or the like.

  In step S4, it is assumed that the collected opinions are collected in an opinion arrangement table 5 as shown in FIG. FIG. 5 is an example configuration diagram showing a specific example of the opinion arrangement table. In Opinion Arrangement Table 5, the pros and cons of each question, the claimant, the reason, etc. are arranged.

  By using the above-mentioned framework for collecting opinions, the argument side can accumulate the concrete argument of the argument side behind the measure and the objection against it. The claim is accompanied by a rationale and rationale. The reason is attached to the objection. The present embodiment is intended to solve the described problem on the premise that the information of the argument arrangement table 2 and the opinion arrangement table 5 accumulated in such a format is input to the persuasion support device.

  The persuasion process and the data collection process described here are merely examples, and are not requirements of the present embodiment. In the present embodiment, there is no problem as long as text data corresponding to the argument arrangement table 2 and the opinion arrangement table 5 exists as usable data.

(Principle of this embodiment)
Next, a description will be given of information created in the persuasion support stage procedure in step S5 and the persuasion support stage procedure in step S5. A clustering technique is used in the persuasion support stage of step S5.

  One method of grouping conflicting claims such as arguments, objections, and re-arguments that occur in the discussion by similarity in content is implied by the degree to which the words that characterize each argument are common. It is a simple combination and external separation, and pulls out some subsets from the theory set.

  Data analysis methods that give quantitative indicators to the interrelationships of sets, and consider division and classification into subsets based on those indicators are clustering or cluster analysis (BSEveritt, “Cluster Analysis”, Edward Arnold, 4th edition) (2001)) and widely used in data mining.

  In this embodiment, a text data clustering technique is used to determine that discussions at each stage belong to similar topic areas. As a result, the present embodiment realizes matching between a group of reason statements of rebuttals similar to each other and a rationale statement of arguments of the measure implementation side. An argument rationale matched with a group of reasons for rebuttal is recommended as appropriate for the construction of re-argument.

  For clustering, various methods are already known that differ depending on the algorithm, the method of giving the distance function, and the like. In order to realize the object of the present embodiment, it is important to determine which part of the discussion text data is to be subjected to clustering, how to control and advance the process, and the like.

  Therefore, in the persuasion support stage of step S5, clustering with the basis sentence as the core is performed as described later. As a scene where this embodiment is useful, there may be a case where a great number of objections are received against the dozens of grounds prepared by the measure implementation side.

  In such a case, for example, by simply merging the ground sentence and the reason sentence (of disagreement) and performing clustering, a large number of sentence groups (clusters) that do not contain any ground sentence may occur. There is sex. For a group of reason sentences included in a cluster that does not include such a ground sentence at all, a ground sentence for re-argument cannot be recommended. In order to avoid this, clustering may be performed in which a cluster including only each ground sentence is generated as an initial arrangement of the cluster, and a reason sentence is assigned to this cluster.

  Such clustering can be realized by a known clustering technique using the k-means method or the like by expressing a sentence with frequency information of feature words included in the sentence. Since each cluster contains only one rationale, it is possible to recommend this rationale as a basis for re-arguing against the objection based on the reasoning within the cluster.

FIG. 6 is a schematic diagram of an example of clustering with two ground sentences as cores. For the reasoning statement (×) of the objection that is included in the cluster (C ₁ , C ₂ ) in FIG. 6, re-argument using the ground statement (○) that is the core of the same cluster is promising. To be judged.

  In addition, in the persuasion support stage of step S5, clustering using rumination information is performed in addition to clustering with the basis sentence as the core as described later. Clustering with the basis sentence as the core uses only word level similarity included in the sentence. Clustering with the basis sentence as the core can expect to some extent that a basis sentence suitable for re-argument can be selected.

  However, clustering with the basis sentence as the core does not use the semantic level information of the sentence, so if the reason sentence included in a cluster can be re-refuted by the basis sentence that is the core of the cluster, Is not limited. Conversely, a reason sentence that is not included in a certain cluster cannot always be re-refuted by a ground sentence that is the core of the cluster.

  In addition, when the degree of similarity is 0 for any cluster, a sentence that is not placed in the cluster may occur. Therefore, in this example, in order to improve the quality of the clustering result, the user of the persuasion support device judged the effectiveness of the objection using a certain reasoning sentence for the reasoning sentence of some objections. Think about giving information.

  FIG. 7 is a configuration diagram of an example of rumination information. In the rumination information in FIG. 7, the reason sentence ID, the ground sentence ID, and the rumination are associated. The basis sentence ID included in the rumination information in FIG. 7 is the basis sentence ID of the basis sentence that determines the rumination for the reason sentence of a certain reason sentence ID. The rumination included in the rumination information in FIG. 7 is a result of determining the validity of the objection using the basis sentence of the basis sentence ID. The rumination included in the rumination information in FIG. 7 is “+” for what is considered to be highly ruminant (a rebuttal using the ground sentence of the ground sentence ID is effective), and the rumination is low (of the ground sentence ID) An index of “-” is given for those who think that the objection using the rationale is not valid.

  When actual clustering is performed, the repulsive information as shown in FIG. 7 is used as follows. The reason sentence of the reason sentence ID included in the record having the rumination “+” is fixed to the current cluster when it belongs to the same cluster as the ground sentence that is the source of the rumination. If the reason sentence with the reason sentence ID included in the record with the rumination “+” does not belong to the same cluster as the ground sentence that is the source of the rumination, a cluster having itself as a nucleus is generated. The reason sentence of the reason sentence ID included in the record having the rumination “−” is not assigned to the same cluster as the ground sentence that is the source of the rumination.

FIG. 8 is a schematic diagram of an example of clustering using rumination information. In FIG. 8, a part of the sentence that is not arranged in the cluster (C ₁ , C ₂ ) in the clustering result of FIG. 6 is arranged in the enclave cluster (C ₁ ″) of the cluster (C ₁ ). Further, in FIG. 8, in the clustering result of FIG. 6, a part of the sentence arranged in the cluster (C ₂ ) is an enclave cluster (C ₁ ′) of the cluster (C ₁ ) having a more appropriate ground for rumination. ).

A typical case where an enclave cluster (C ₁ ″) is generated is a topic area in which the ground sentence and the reason sentence are similar in content, but synonyms are used. This is a case where it does not increase. For example, it is assumed that there are a basis sentence of the cluster (C ₁ ) as shown in FIG. 9 and two reason sentences A and B. FIG. 9 is an explanatory diagram showing an example of a ground sentence and a reason sentence.

In FIG. 9, the ground sentence of the cluster (C ₁ ) and the reason sentence A share four feature words, and the similarity is high. For this reason, the reason sentence A is highly likely to be assigned to the cluster (C ₁ ). On the other hand, the reason sentence B is semantically almost the same as the reason sentence A, but the similarity is 0 because there is no feature word in common with the ground sentence of the cluster (C ₁ ). For this reason, the reason sentence B is not assigned to the cluster (C ₁ ).

In the case shown in FIG. 9, if a “+” rumination is given between the ground sentence of the cluster (C ₁ ) and the reason sentence B, the cluster (C ₁ ′) around the reason sentence B in this embodiment. ') Can be expected.

  In general, in text data clustering, synonym dictionaries are often constructed to correctly handle the characteristics of words having the same meaning. However, as in the case of public comments, when a variety of topic discussions must be processed in a short period of time, it is not practical to correctly handle the characteristics of words having the same meaning by constructing a synonym dictionary.

(Persuasion support device)
FIG. 10 is a hardware configuration diagram of an example of a persuasion support apparatus. As shown in FIG. 10, the hardware of the persuasion support device 10 is realized by a computer system, for example. 10 includes an input device 11, an output device 12, a drive device 13, an auxiliary storage device 14, a main storage device 15, an arithmetic processing device 16, and an interface device 17 that are connected to each other via a bus B. .

  The input device 11 is a keyboard or a mouse. The input device 11 is used for inputting various signals. The output device 12 is a display device or the like. The output device 12 is used for displaying various windows and data. The interface device 17 is a modem, a LAN card, or the like. The interface device 17 is used for connecting to a network.

  The persuasion support program of this embodiment is at least a part of various programs that control the persuasion support device 10. The persuasion support program is provided, for example, by distributing the recording medium 18 or downloading from the network. The recording medium 18 on which the persuasion support program is recorded is information such as a CD-ROM, a flexible disk, a magneto-optical disk, etc., a recording medium for recording information optically, electrically or magnetically, a ROM, a flash memory, etc. Various types of recording media such as a semiconductor memory for electrical recording can be used.

  When the recording medium 18 on which the persuasion support program is recorded is set in the drive device 13, the persuasion support program is installed from the recording medium 18 to the auxiliary storage device 14 via the drive device 13. The persuasion support program downloaded from the network is installed in the auxiliary storage device 14 via the interface device 17.

  The auxiliary storage device 14 stores the installed persuasion support program and also stores necessary files, data, and the like. The main storage device 15 reads and stores the persuasion support program from the auxiliary storage device 14 when the persuasion support program is started. The arithmetic processing unit 16 implements various processes as described later in accordance with a persuasion support program stored in the main storage device 15.

  FIG. 11 is a block diagram of an example of a persuasion support apparatus. 11 includes a document information storage unit 21, a data input unit 22, a processing unit 23, a display unit 24, a feature vector information storage unit 25, a cluster information storage unit 26, and a representative point information storage unit 27.

  In addition, the document information storage unit 21 includes an argument arrangement table storage unit 31, an opinion arrangement table storage unit 32, and a rumination information storage unit 33. The processing unit 23 includes a morphological analysis unit 34, a feature word extraction unit 35, a clustering unit 36, and an output processing unit 37.

  The document information storage unit 21 corresponds to, for example, the auxiliary storage device 14 in FIG. The data input unit 22 corresponds to the input device 11 of FIG. The processing unit 23 corresponds to the arithmetic processing device 16 in FIG. 10 that has executed the persuasion support program. The display unit 24 corresponds to the output device 12 of FIG. The feature vector information storage unit 25, the cluster information storage unit 26, and the representative point information storage unit 27 correspond to the auxiliary storage device 14 or the main storage device 15 of FIG.

  The issue point arrangement table storage unit 31 in FIG. 11 stores, for example, the issue point arrangement table 2 shown in FIG. The opinion arrangement table storage unit 32 stores, for example, the opinion arrangement table 5 shown in FIG. The rumination information storage unit 33 stores, for example, the rumination information shown in FIG.

  FIG. 12 is a flowchart illustrating an example of a processing procedure of the persuasion support apparatus. In step S11, the processing unit 23 reads the text data of the rational sentence from the argument arrangement table 2 stored in the argument arrangement table storage unit 31. Further, the processing unit 23 reads the text data of the reason sentence from the opinion organizing table 5 stored in the opinion organizing table storage unit 32.

  In step S12, the morphological analysis unit 34 of the processing unit 23 performs morphological analysis of the text data of the ground sentence and the reason sentence read in step S11. The morpheme analysis unit 34 decomposes the text data of the ground sentence and the reason sentence into words, and determines the part of speech. In addition, since the process of step S12 is an existing technique, detailed description is abbreviate | omitted.

  In step S13, the feature word extraction unit 35 extracts the feature words of each sentence from the result of the morphological analysis in step S12. The feature word extraction unit 35 extracts nouns of each sentence, and extracts those that well indicate the features of the sentence as feature words. In addition, since the process of step S13 is an existing technique, detailed description is abbreviate | omitted.

  In step S14, the feature word extraction unit 35 writes the feature words of each sentence extracted in step S13 in the feature vector information storage unit 25 as feature vector information. FIG. 13 is a configuration diagram of an example of feature vector information.

The feature vector information of FIG. 13 arranges the document IDs of two ground sentences G ₁ and G _{2 on} the vertical axis and the document IDs of ten reason sentences R _{1 to} R ₁₀ on the vertical axis, and extracts from each sentence on the horizontal axis. The feature words are arranged. The feature vector information in FIG. 13 includes 23 feature words (t _{1 to} t ₂₃ ), and a feature vector indicating whether or not a feature word is included in each sentence is represented by 0/1. In the feature vector information of FIG. 13, presence of feature words is represented by a feature vector “1”, and absence of feature words is represented by a feature vector “0”.

  In the feature vector information in FIG. 13, the blank represents “0”. Note that the feature vector information in FIG. 13 shows an example of a feature vector indicating whether or not a feature word is included in each sentence, but it may be a feature vector indicating the frequency of inclusion of a feature word in each sentence.

  In step S15, the clustering unit 36 writes an initial arrangement of cluster information described later in the cluster information storage unit 26. In step S16, the clustering unit 36 performs clustering as described later. Further, the output processing unit 37 can select a rational sentence suitable for re-argument with respect to each objection by referring to the execution result of the clustering in step S16 written in the cluster information storage unit 26.

  In step S17, the output processing unit 37 executes the clustering execution result in step S16 written in the cluster information storage unit 26, the argument arrangement table 2 read from the argument arrangement table storage unit 31 and the opinion arrangement table storage unit 32, and Based on the opinion arrangement table 5, the re-argument support information of FIG. 14 is assembled and output to the display unit 24 as described later.

FIG. 14 is a configuration diagram of an example of re-argument support information. In the re-refusal support information in FIG. 14, a reason sentence ID, a reason sentence, and a re-refutation recommendation basis sentence are associated with each other. It should be noted that the statement of the reason R ₈ is has become opponents of factual errors content.

  As shown in the argument arrangement table 2 illustrated in FIG. 4, when the measure implementation side arranges the rationale sentence in addition to the rationale sentence, it is also possible to present the template of the repetitive article as the reexamination support information. For example, with regard to the two rebuttal recommendation grounds shown in FIG. 14, “there is no problem to relax scientifically insignificant inspection conditions”, “if the SRM removal equivalent to Japan has been done, BSE If it is given as a rationale, there is no problem even if the examination conditions are relaxed, and they can be presented as a document.

A document that combines a rationale sentence and a rationale sentence is considered to be a “reasonable sentence” because, for example, “a rationale sentence”. ]
It should be presented according to such a pattern. FIG. 15 is a configuration diagram of another example of re-argument support information. In the rebuttal support information in FIG. 15, the reason sentence ID, the reason sentence, and the template of the rebuttal paper are associated with each other. The reason sentence R8 is misidentified by the opponent.

(Definition example of document feature vector and similarity between documents)
In this embodiment, the degree of similarity between documents is used to determine whether or not the re-argument performed by the measure implementation side and the objection are well meshed. The degree of similarity is a measure of the degree of similarity between the document of the ground sentence in the argument arrangement table 2 on the argument side and the document of the reason sentence in the opinion arrangement table 5 on the objection side. If the similarity between the ground sentence and the reason sentence is high, both are considered to belong to the same topic area. If they belong to the same topic area, re-argument is likely to be effective for the opposite reason.

Various methods for document similarity are already known. Here, a general tfidf method is exemplified. In the tfidf method, a document group: D _i (i = 1, 2, 3,..., N) and feature words t _j (j = 1, 2, 3,..., n) included therein are included. think of. The document is a document that specifically describes the rationale on the argument side (may include the rationale) and a document that describes the reason on the answer side.

Here, when tf _ij : the frequency of occurrence of the word t _j in the document D _i and df _j : the number of documents including the word t _j , vectors (w _i1 , w _i2 , w) representing the characteristics of the document: D _i are used. _i3, the _{···, w in)} and notation. n is the total number of feature words.

Further, the amount obtained by dividing the total document number N by df _j and taking the logarithm is defined as the reverse appearance frequency idf _j as shown in the following equation (1).

idf _j = log (N / df _j ) (1)
Next, as shown in the following formulas (2) and (3), each element of the feature vector is given by the following formula and further normalized.
w _ij = tf _ij × idf _j (2)
a _ij = w _ij × | w _ij | (3)
| W _ij | is the magnitude of the vector. By using the amounts as described above, in the tfidf method, the similarity S between the two documents D _s and D _r is given by, for example, the following equation (4).

_{S (D s, D r)} = a s · a r ··· (4)
“·” Represents an inner product of vectors. Using the similarity S in equation (4), it is possible to better refute the reason sentence by calculating the similarity of all the ground sentences for one reason sentence: R _i and ordering by the magnitude of the value. It is possible to prioritize highly rational texts.

As a simpler definition example of the similarity between documents, there is a method in which only the presence / absence (for example, 0 or 1) is vectorized instead of the word frequency. Specifically, for the document group: D _i and the feature word t _j included therein, w _ij = 1 when the word t _j is included in the document D _i , and w _ij = 0 when the word t _j is not included _Is defined as the feature vector w _i of the document D _i . Further, the similarity S between the two documents D _s and D _r is given by, for example, the following equation (5). “·” Represents an inner product of vectors.

S (D _s , D _r ) = w _s · w _r (5)
(Clustering based on the rationale)
Here, as an example of the processing in steps S15 and S16 in FIG. 11, an example will be described in which clustering of reason sentences with the basis sentence as a core is performed by the k-means method. FIG. 16 is a flowchart showing a processing procedure for clustering reason sentences with the basis sentence as a core. The k-means method is, for example, “J. McQueen” Some methods for classification and analysis of
multivariate observations "In Proceedings of the Fifth Berkeley Symposium
on Mathematical Statistics and Probability, pp.281-297 (1967) ”.

In step S21, the clustering unit 36 sets a cluster (C _k ) including each rational sentence as an initial state of the cluster. The process proceeds to step S22, the clustering section 36 calculates the representative point _{g k} clusters. There are various definitions of the representative points. In the k-means method, the representative point g _k of the cluster is often defined by the centroid of the feature vector of the document included in the cluster. The clustering unit 36 writes the calculated cluster representative point g _k into the representative point information storage unit 27.

Specifically, the reason sentence belonging to the cluster (C _k ) is _defined as R _ik, and its median value g _jk is defined as the centroid of the feature vector a _ijk of the reason sentence R _ik by Equation (6). Further, the representative point g _k of the cluster, which is an n-dimensional vector quantity, is expressed as in Expression (7).

  The clustering unit 36 writes the relationship between each rational sentence and the cluster in the cluster information storage unit 26 as the initial arrangement of the cluster information. FIG. 17 is a block diagram showing an example of the initial arrangement of cluster information.

The cluster information in FIG. 17 includes the representative point of cluster (C _k ): g _k , the similarity of each reason sentence (numerical part in FIG. 17), and the name of the cluster to which the reason sentence belongs at that time: C _k Is a table. The cluster information in FIG. 17 is calculated based on the feature vector information in FIG.

To the cluster containing the rationale statements in the initial state of the cluster in step S21, the clustering section 36 obtains the representative point g _k in step S22. However, in the initial state, since only the ground sentence is included in the cluster, the representative point g _k becomes the feature vector itself of the ground sentence.

  In step S 23, the clustering unit 36 creates an all-reason sentence list composed of feature vectors of all the reason sentences from the feature vector information as shown in FIG. 13 written in the feature vector information storage unit 25.

In step S24, the clustering unit 36 extracts one reason sentence vector R _i from the all reason sentence list created in step S23. In step S25, the clustering unit 36 calculates the similarity S _ik between the reason sentence vector R _i extracted in step S24 and the representative point g _{k of} each cluster.

In the case of the cluster information of FIG. 17, the clustering unit 36 calculates the similarity S _ik between the reason sentence vector R _i extracted in step S24 and the cluster representative points g ₁ and g ₂ . The process proceeds to step S26, the clustering section 36 the maximum degree of similarity _{S ik} and _{(S max),} the similarity _{S ik} seek a cluster _{C max} when the maximum value _{(S max).}

For example, in the case of cluster information of FIG. 17, the clustering section 36 the maximum degree of similarity _{S ik} in step S26 after the removal of the reason sentence vector _{R 1} at step S24 and (4), the similarity _{S ik} is the maximum value (4 ) seek and cluster _{C 1} at the time of the.

In step S27, the clustering unit 36 determines whether the maximum value (S _max ) of the similarity S _ik obtained in step S26 is greater than zero. If the maximum value (S _max ) of the similarity S _ik obtained in step S26 is not greater than 0, in other words, the maximum value (S _max ) of the similarity S _ik is 0 for the representative points g _k of all clusters. If so, the clustering unit 36 returns to step S24. In other words, the clustering unit 36 does not _assign the reason sentence R _i to any cluster if the maximum value (S _max ) of the similarity S _ik is 0 for the representative points g _k of all clusters.

If the maximum value of the similarity _{S ik} obtained in step S26 _{(S max)} is greater than 0, the clustering unit 36 proceeds to step S28, a statement of the reason _{R i} reason sentence vector _{R i} extracted in step S24, step S26 Is assigned to the cluster C _max obtained in (1). The process of step S24 to S28, the reason statement _{R i} is based on the similarity _{S ik} the representative point _{g k} of clusters, the representative point _{g k} are arranged in the closest cluster _{C max.}

  Proceeding to step S29, the clustering unit 36 returns to step S24 if a reason sentence that has not been extracted remains in the all reason sentence list. The clustering unit 36 proceeds to step S30 when there are no reason sentences that have not been extracted from the all reason sentence list.

  In step S30, the clustering unit 36 determines whether or not the final condition for clustering is satisfied. If the final condition for clustering is not satisfied, the process returns to step S23. If the final condition for clustering is satisfied, the clustering unit 36 terminates the processing shown in FIG. Note that the cluster information in FIG. 17 is an example when the clustering unit 36 first visits the process of step S30.

Thus, in the clustering by the k-means method, cluster _C calculation representative point _{g k} of _k, the computation of Ruido _{S ik} the representative point _{g k} and the reason statement _{R i,} cluster C which belongs reason statement _{R i It} continues until the loop (S23-S30) which consists of repetition of the update of _k satisfy | fills the final condition of clustering.

When the cluster information is as shown in FIG. 17, the clustering unit 36 calculates a feature vector of the representative point g _k as shown in FIG. Figure 18 is a block diagram showing an example of a feature vector of the representative point g _k.

As the final condition of the clustering in step S30, clustering convergence (no updating of the cluster C _k to which the reason sentence R _i belongs), or the number of occurrences of cluster updating being smaller than a threshold value are usually used. The final state of clustering when clustering converges with cluster information as shown in FIG. 17 is as shown in FIG. FIG. 19 is a block diagram showing an example of the final state of cluster information.

(Providing rumination information)
The user of the persuasion support device 10 according to the present embodiment can provide repulsive information as shown in FIG. 7 in which the validity of the objection using a certain ground sentence is determined for the reason sentences of some objections. . It is arbitrary whether to provide the rumination information for which combination of reason sentence and ground sentence. However, it can be expected that the accuracy of clustering results will be improved if more rumination information is given.

In particular, it is desirable to give rumination information is a reason sentence that does not belong to any cluster. In the example of FIG. 19, it is desirable to provide rumination information to the reason sentences R _{4 to} R ₆ . Under the current situation, it is not possible to recommend a ground sentence for any objection as re-argument support information for the reason sentences R _{4 to} R ₆ .

  However, if a positive “+” rumination is given to a reason sentence, it is possible to provide re-refutation support information for the reason sentence and other reason sentences similar to the reason sentence. It becomes. Therefore, when prompting the user to add rumination information, it is desirable to give priority to reason sentences that do not belong to any cluster.

FIG. 20 is an image diagram of an example of a table type interface for inputting rumination information. FIG. 20 shows an example in which a rationale sentence is taken on the vertical axis and a reason sentence is taken on the horizontal axis, and the user arbitrarily inputs rumination in the intersection column. In FIG. 20, there is a positive “+” rumination for the ground sentence G ₁ and the reason sentences R ₄ and R ₇ , and a negative “−” rumination for the ground sentence G ₂ and the reason sentence R _7. Is given.

  In FIG. 20, “0” is entered in the field where there is no repulsive input. The last line in FIG. 20 represents the cluster to which each reason sentence currently belongs. As a table-type interface for inputting repulsive information, it is not always necessary to display a cluster to which each reason sentence in the last line belongs.

  However, as a table-type interface for entering rumination information, the reason statement (indicated by φ) that does not currently belong to any cluster can be clearly told to the user, or it can currently belong to any cluster. It is possible to raise the priority of input by expressing the reason sentence that does not exist in an easy-to-understand manner (in the example of FIG. 20, it is moved to the left side). In FIG. 20, there are parts where the reason sentence is indicated by a symbol for the sake of simplicity, but it is desirable that the reason sentence be expanded with a complete reason sentence.

  Also, in this embodiment, when a reason sentence having a positive rumination belongs to the same cluster as the ground sentence that is the source of the rumination, the reason sentence is fixed to the current cluster. Therefore, in this embodiment, one reason sentence cannot belong to two different clusters. For this reason, it is forbidden to give positive rumination to two ground sentences at the same time.

  The table-type interface for inputting rumination information as shown in FIG. 20 is in a form that makes it easy for the user to check forbidden matters that positive rumination cannot be given to two ground sentences at the same time. Yes. It is also possible for the persuasion support device 10 to check for violations of prohibition and to urge the user to correct the violation.

(Clustering using rumination information)
Here, as an example of the processing of steps S15 and S16 in FIG. 11, an example of performing clustering using rumination information in addition to clustering of reason sentences with the basis sentence as a core will be described. FIG. 21 is a flowchart showing a processing procedure of clustering using repetitive information in addition to clustering of reason sentences with the basis sentence as a core.

  In the flowchart of FIG. 21, one turn of preliminary clustering that does not use rumination information is performed at the beginning, and the initial arrangement of clusters is made. This preliminary clustering is based on the normal k-means method, and is the same processing as the flowchart of FIG. After the preliminary clustering, the clustering using the repulsive information is executed in the flowchart of FIG.

In the following, an example is shown in which the flowchart of FIG. 21 is executed using the same feature vector as that described in the flowchart of FIG. 16 and the repulsive information shown in FIG. FIG. 22 is a configuration diagram of an example of rumination information. The rumination information in FIG. 22 shows an example in which a positive “+” rumination is given to the reason sentences R ₄ and R ₇ .

  More specifically, it is assumed that a part of the ground sentence and the reason sentence handled in the description of the flowchart of FIG. 21 are as shown in FIG. FIG. 23 is an explanatory diagram showing an example of a ground sentence and a reason sentence.

  Proceeding to step S41 in FIG. 21, the clustering unit 36 performs one turn of preliminary clustering without using the repulsive information. This preliminary clustering is clustering as shown in FIG. Therefore, the cluster information immediately after step S1 is equal to that in FIG.

  In the flowchart of FIG. 21, step S49 is clustering using the repulsive information. Steps S42 to S48 between step S41 and step S49 are parts for preparing clustering using the repulsive information of step S49.

The process proceeds to step S42, the clustering section 36 fetches one reason statement R _i from the total reason statement list created in step S41. Note that the clustering unit 36 reads the repulsive information as shown in FIG. 22 from the repulsive information storage unit 33. In step S43, the clustering unit 36 determines whether there is rumination information for the reason sentence R _i extracted in step S42.

If refute property information for why statements R _i extracted in step S42, the clustering unit 36 proceeds to step S44, with reference to the refutation of information for reason statement R _i, refutation of positive "+" in the statement of the reason R _i Is determined. If a positive “+” rumination is given to the reason sentence R _i , the clustering unit 36 proceeds to step S45, and confirms the ground sentence that is the source of the rebuttal of the reason sentence R _i from the rumination information.

The clustering unit 36 refers to the cluster information written in the cluster information storage unit 26, and determines whether or not the reason sentence R _i extracted in step S42 belongs to the same cluster as the ground sentence that is the source of the rumination. judge.

If the reason sentence R _i extracted in step S42 does not belong to the same cluster as the ground sentence that is the source of rumination, the clustering unit 36 proceeds to step S46, and uses the reason sentence R _i extracted in step S42 as a nucleus. After performing an operation for generating a new cluster to be performed, the process proceeds to step S47.

On the other hand, if the reason sentence R _i extracted in step S42 belongs to the same cluster as the ground sentence that is the source of rumination, the clustering unit 36 proceeds to step S47. In step S42, the clustering unit 36 fixes the reason sentence R _i extracted in step S42 to the current cluster.

Proceeds to step S48, the clustering unit 36 and there are still a reason statement R _i that is not taken out the entire reason statement list created in step S41, the flow returns to step S42. On the other hand, if there is no reason sentence R _i remaining in the all reason sentence list, the clustering unit 36 proceeds to step S49.

Through the processing up to step S48, two clusters C ₁ ′ and C ₁ ″ are generated in the present embodiment. The cluster C ₁ ′ includes a reason sentence R ₇ first. Cluster C ₁ ″ initially includes reason sentence R ₄ .

In step S49, the clustering unit 36 performs clustering on the reason sentence R _i that is not fixed. The cluster information first calculated by the clustering in step S49 is, for example, as shown in FIG. FIG. 24 is a block diagram showing an example of the initial arrangement of cluster information.

In FIG. 24, “1” is set as a fixed flag in the flag column of the reason sentences R ₄ and R ₇ based on the rumination information of FIG. If negative “−” rumination is given, “−1” is set in the flag field.

The clustering in step S49 is the same as the clustering in FIG. 16 except for handling negative “−” rumination. In the clustering in step S49, when the reason sentence R _i has a negative “−” repulsion for a certain cluster, it is prohibited to assign the reason sentence R _i to the cluster.

Note that in the case of the refractory information shown in FIG. 22 to which no negative “−” repellency is given, the behavior in step S49 is the same as the processing in the flowchart of FIG. Among the clustering step S49, the feature vector of the representative point g _k of each cluster is as shown in FIG. 25 for example to be calculated first. Figure 25 is a block diagram showing an example of a feature vector of the representative point g _k.

  Further, the final result of the clustering in step S49 is as shown in FIG. 26, for example. FIG. 26 is a block diagram showing an example of the final state of the cluster information. The output processing unit 37 reads cluster information as shown in FIG. 26 as a result of clustering, and outputs re-argument support information as shown in FIG. 27 to the display unit 24.

FIG. 27 is an image diagram of an example of re-argument support information. The re-argument support information in FIG. 27 is associated with a reason sentence ID, a reason sentence, and a re-argument recommendation basis sentence. It should be noted that the statement of the reason R ₈ is has become opponents of factual errors content. By referring to the re-refutation support information in FIG. 27, the user can easily know a valid rationale for the judgment opinion, and it becomes easy to create a re-refutation paper 6 used for re-refutation. .

(Summary)
In the following, the result of clustering not using the repulsive information shown in the flowchart of FIG. 16 is compared with the result of clustering using the repulsive information shown in the flowchart of FIG.

FIG. 28 is a comparison diagram of an example of a result of clustering that does not use rumination information and a result of clustering that uses rumination information. As can be seen from FIG. 28, as a result of clustering that does not use rumination information, reason sentences R _{4 to} R _{6 in} which the similarity cannot be placed in any cluster are generated. The reason why the reason sentences R _{4 to} R ₆ that cannot be placed in any cluster occurs is because the similarity is zero for any reason sentence in which the reason sentences R _{4 to} R ₆ are placed in the cluster.

As a result of clustering that does not use rumination information, the reason sentence R ₇ “Because the etiology of BSE has not been determined, it cannot be said that safety is ensured by removing SRM (specific risk sites)” in cluster C ₂ Has been placed. However, refuted is difficult to grounds sentence G ₂ grounds against reason statement R _7. This is because the statement of the reason R ₇ is in rebuttal from the front to the grounds sentence G ₂ of the cluster C _2.

On the other hand, the result of clustering using repulsive information, in other words, using repulsive information as a guide, is improved as follows. The reason sentences R _{4 to} R _{6 that} are not arranged in any cluster as a result of the clustering not using the repulsive information are arranged in the cluster C ₁ (the enclave cluster C ₁ ″).

In addition, the reason sentences R ₇ and R _{10 which} are arranged in the cluster C _{2 as} a result of the clustering not using the rumination information are arranged in the cluster C ₁ (the enclave cluster C ₁ ′) having a more appropriate basis for the rumination. Is done.

  FIG. 29 shows an easy-to-understand illustration of the improvement of the result of clustering using the repulsive information from the result of clustering not using the repulsive information. FIG. 29 is a schematic diagram comparing the result of clustering that does not use rumination information and the result of clustering that uses rumination information.

In FIG. 29, the result of clustering using the repulsive information has the effect of increasing the possibility of recommending a rational sentence suitable for re-refutation by allowing the enclaves C ₁ ′ and C ₁ ″ for the cluster C ₁ . .

The present invention may have the following configurations as described below.
(Appendix 1)
A persuasion support device that supports written persuasion to the opponent of the claim,
A rebuttal information storage unit that stores reciprocity information in which a reason sentence of a disagreement with respect to the claim, a ground sentence of the claim, and the reason sentence and the presence or absence of rumination of the ground sentence are associated;
A feature word extraction unit that extracts a feature word included in the reason sentence and the basis sentence from a result of morphological analysis of the reason sentence and the basis sentence in the rumination information storage unit;
Generate cluster information representing the cluster of the initial placement including the basis sentence, and place the reason sentence in the cluster based on the similarity between the feature word included in the basis sentence and the feature word included in the reason sentence After performing clustering to rewrite the cluster information, the cluster information is rewritten so as to relocate the cluster to which the reason sentence is allocated based on the repetitive information stored in the repetitive information storage unit. A clustering unit;
A persuasion support apparatus comprising: an output processing unit that outputs the basis sentence included in the same cluster as the reason sentence as a basis sentence for re-refutation on the reason sentence based on the cluster information.
(Appendix 2)
The clustering unit, based on the rumination information, when the reason sentence having the rumination and the basis sentence are included in the same cluster, the reason sentence is fixed to the currently arranged cluster. After rewriting the cluster information, the cluster information is rewritten so that the reason sentence is rearranged in the cluster based on the similarity between the feature word contained in the basis sentence and the feature word contained in the reason sentence. The persuasion support device according to supplementary note 1, wherein clustering is performed.
(Appendix 3)
The clustering unit generates an enclave cluster of the basis sentence including the reason sentence when the reason sentence having the rumination and the basis sentence are not included in the same cluster based on the rumination information. After rewriting the cluster information, the cluster information is further relocated to the cluster based on the similarity between the feature word included in the basis sentence and the feature word included in the reason sentence. The persuasion support apparatus according to Supplementary Note 1 or 2, which performs clustering to be rewritten.
(Appendix 4)
The clustering unit, based on the rumination information, prohibits placing the reason sentence in the currently arranged cluster when the reason sentence having no rumination and the basis sentence are included in the same cluster. The cluster information is rewritten so that, based on the similarity between the feature word included in the basis sentence and the feature word included in the reason sentence, the reason sentence is rearranged in the cluster. The persuasion support apparatus according to any one of appendices 1 to 3, which performs clustering for rewriting cluster information.
(Appendix 5)
The output processing unit selects, based on the cluster information, the basis sentence included in the same cluster as the reason sentence as a basis sentence for re-argument to the reason sentence, and the reason sentence and the selected basis sentence are The persuasion support apparatus according to any one of supplementary notes 1 to 4, wherein a model of a re-reversion paper for the reason sentence is generated and output.
(Appendix 6)
A computer that supports written persuasion to the opponent of the claim,
A rebuttal information storage unit that stores reciprocity information in which a reason sentence of a disagreement with respect to the claim, a ground sentence of the claim, and the reason sentence and the presence or absence of rumination of the ground sentence are associated;
A feature word extraction unit that extracts a feature word included in the reason sentence and the basis sentence from a result of morphological analysis of the reason sentence and the basis sentence in the rumination information storage unit;
Generate cluster information representing the cluster of the initial placement including the basis sentence, and place the reason sentence in the cluster based on the similarity between the feature word included in the basis sentence and the feature word included in the reason sentence After performing clustering to rewrite the cluster information, the cluster information is rewritten so as to relocate the cluster to which the reason sentence is allocated based on the repetitive information stored in the repetitive information storage unit. A clustering unit;
A persuasion support program for functioning as an output processing unit that outputs the ground sentence included in the same cluster as the reason sentence as a ground sentence for re-refutation on the reason sentence based on the cluster information.
(Appendix 7)
A persuasion support method implemented by a computer to support written persuasion of opponents against a claim,
The computer is
The reason sentence of the rebuttal information storage unit that stores the reason sentence of the objection to the claim, the ground sentence of the claim, and the rebuttal information in which the reason sentence and the presence or absence of the rumination of the ground sentence are associated A feature word extraction step of extracting a feature word included in the reason sentence and the basis sentence from a result of morphological analysis of the sentence;
Generate cluster information representing the cluster of the initial placement including the basis sentence, and place the reason sentence in the cluster based on the similarity between the feature word included in the basis sentence and the feature word included in the reason sentence After performing clustering to rewrite the cluster information, the cluster information is rewritten so as to relocate the cluster to which the reason sentence is allocated based on the repetitive information stored in the repetitive information storage unit. A clustering step;
A persuasion support method for executing, based on the cluster information, an output processing step of outputting the ground sentence included in the same cluster as the reason sentence as a ground sentence for re-refutation with respect to the reason sentence.
The present invention is not limited to the specifically disclosed embodiments, and various modifications and changes can be made without departing from the scope of the claims.

DESCRIPTION OF SYMBOLS 1 Material 2 Discussion point arrangement table 3 Question sentence 4 Answer sentence 5 Opinion arrangement table 6 Reexamination paper 10 Persuasion support device 11 Input device 12 Output device 13 Drive device 14 Auxiliary storage device 15 Main storage device 16 Arithmetic processing device 17 Interface device 18 Record Medium 21 Document information storage unit 22 Data input unit 23 Processing unit 24 Display unit 25 Feature vector information storage unit 26 Cluster information storage unit 27 Representative point information storage unit 31 Discussion point organization table storage unit 32 Opinion organization table storage unit 33 Repetitive information storage Unit 34 morphological analysis unit 35 feature word extraction unit 36 clustering unit 37 output processing unit B bus

Claims (6)

A persuasion support device that supports written persuasion to the opponent of the claim,
A rebuttal information storage unit that stores reciprocity information in which a reason sentence of a disagreement with respect to the claim, a ground sentence of the claim, and the reason sentence and the presence or absence of rumination of the ground sentence are associated;
A feature word extraction unit that extracts a feature word included in the reason sentence and the basis sentence from a result of morphological analysis of the reason sentence and the basis sentence in the rumination information storage unit;
Cluster information including only the basis sentence as an initial arrangement of clusters is generated, and the reason sentence is arranged in the cluster based on the similarity between the feature word included in the basis sentence and the feature word included in the reason sentence. Clustering for rewriting the cluster information so as to rearrange the clusters to which the reason sentence is allocated based on the repetitive information stored in the repetitive information storage unit after performing clustering to rewrite the cluster information as described above And
A persuasion support apparatus comprising: an output processing unit that outputs the basis sentence included in the same cluster as the reason sentence as a basis sentence for re-refutation on the reason sentence based on the cluster information.   The clustering unit, based on the rumination information, when the reason sentence having the rumination and the basis sentence are included in the same cluster, the reason sentence is fixed to the currently arranged cluster. After rewriting the cluster information, the cluster information is rewritten so that the reason sentence is rearranged in the cluster based on the similarity between the feature word contained in the basis sentence and the feature word contained in the reason sentence. The persuasion support apparatus according to claim 1, wherein clustering is performed.   The clustering unit generates an enclave cluster of the basis sentence including the reason sentence when the reason sentence having the rumination and the basis sentence are not included in the same cluster based on the rumination information. After rewriting the cluster information, the cluster information is further relocated to the cluster based on the similarity between the feature word included in the basis sentence and the feature word included in the reason sentence. The persuasion support apparatus according to claim 1, wherein rewriting clustering is performed.   The clustering unit, based on the rumination information, prohibits placing the reason sentence in the currently arranged cluster when the reason sentence having no rumination and the basis sentence are included in the same cluster. The cluster information is rewritten so that, based on the similarity between the feature word included in the basis sentence and the feature word included in the reason sentence, the reason sentence is rearranged in the cluster. The persuasion support apparatus according to any one of claims 1 to 3, wherein clustering for rewriting cluster information is performed. A computer that supports written persuasion to the opponent of the claim,
A rebuttal information storage unit that stores reciprocity information in which a reason sentence of a disagreement with respect to the claim, a ground sentence of the claim, and the reason sentence and the presence or absence of rumination of the ground sentence are associated;
A feature word extraction unit that extracts a feature word included in the reason sentence and the basis sentence from a result of morphological analysis of the reason sentence and the basis sentence in the rumination information storage unit;
Cluster information including only the basis sentence as an initial arrangement of clusters is generated, and the reason sentence is arranged in the cluster based on the similarity between the feature word included in the basis sentence and the feature word included in the reason sentence. Clustering for rewriting the cluster information so as to rearrange the clusters to which the reason sentence is allocated based on the repetitive information stored in the repetitive information storage unit after performing clustering to rewrite the cluster information as described above And
A persuasion support program for functioning as an output processing unit that outputs the ground sentence included in the same cluster as the reason sentence as a ground sentence for re-refutation on the reason sentence based on the cluster information. A persuasion support method implemented by a computer to support written persuasion of opponents against a claim,
The computer is
The reason sentence of the rebuttal information storage unit that stores the reason sentence of the objection to the claim, the ground sentence of the claim, and the rebuttal information in which the reason sentence and the presence or absence of the rumination of the ground sentence are associated A feature word extraction step of extracting a feature word included in the reason sentence and the basis sentence from a result of morphological analysis of the sentence;
Cluster information including only the basis sentence as an initial arrangement of clusters is generated, and the reason sentence is arranged in the cluster based on the similarity between the feature word included in the basis sentence and the feature word included in the reason sentence. Clustering for rewriting the cluster information so as to rearrange the clusters to which the reason sentence is allocated based on the repetitive information stored in the repetitive information storage unit after performing clustering to rewrite the cluster information as described above Steps,
A persuasion support method for executing, based on the cluster information, an output processing step of outputting the ground sentence included in the same cluster as the reason sentence as a ground sentence for re-refutation with respect to the reason sentence.

Images