JP2016139299A - Information processing system, information processing method, and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To improve the accuracy of implication clustering to conversation texts.SOLUTION: A representative candidate extraction part 22 of a clustering system 1 extracts a section having a high possibility that prescribed contents are clearly indicated among sections related to the prescribed contents in each of one or more conversation texts as a representative candidate. A member candidate extraction part 23 extracts a section including a section related to the prescribed contents and being larger than the section related to the prescribed contents as a member candidate in each of one or more conversation texts. A partial text output part 24 outputs the extracted representative candidate and member candidate as a partial text from which an implication relation should be extracted.SELECTED DRAWING: Figure 1

Description

  The present invention relates to an information processing system, an information processing method, and a program.

  Opinions such as dissatisfaction and requests for various products and services are received from customers at the call center. It is important for companies to improve services and utilize them in product development based on customer opinions. When summing up customer opinions, cost increases when listening to voices recorded at the call center. Further, when the operator summarizes and inputs the customer's opinions, there is a possibility that the operator's work increases or the input content is blurred or missing. Therefore, it is desired to extract, summarize, and tabulate opinions from the conversation text generated by voice recognition for the recorded voice.

  As a technique that can be used for tabulating opinions included in text, for example, in Non-Patent Document 1, implication relationships between texts are extracted, and texts with implication relationships are classified into the same group. Technology is disclosed. An implication relationship is a semantic relationship between texts, and when the content of the second text can be read from the content of the first text, the first text is defined to imply the second text. In the implication clustering technique, texts that commonly imply text in a group are extracted as representative sentences. By using such an implication clustering technique, it is possible to extract clearly from the viewpoint of topics included in the text.

Japanese Patent No. 5387870

"NEC develops technology to automatically group large amounts of document data in the same sense", [online], NEC Corporation, [January 8, 2015 search], Internet <URL: http: // jpn. nec.com/press/201411/20141118_02.html>

  The conversation text is not structured as a document, but is a text including a redundant part other than greetings, redundant words, or opinions to be clustered. For this reason, for conversational text, the extraction of implications may not be performed correctly unless such redundant portions are removed.

  In addition, the conversation text is divided by, for example, a silent section by voice recognition, and sentences are generated in units different from grammatical sentences. For this reason, if implication clustering is performed in units of sentences, a group is formed around sentences that are too short, and there is a possibility that meaningful opinions cannot be aggregated.

  Furthermore, in the conversation text, the subject and the predicate are separated in the flow of conversation, and the opinion may not be expressed accurately in one sentence. In this case, for example, if only a part in which an opinion is accurately expressed in a plurality of sentences is extracted as an object of implication clustering, there is a possibility that omission may occur in the aggregation. Conversely, if a large section extending over a plurality of sentences is extracted as an object of implication clustering, the above-described redundant portion is included, and there is a possibility that the extraction of the implication relationship is not performed.

  As described above, when the implication clustering technique is applied to the conversation text, there is a technical problem that the clustering accuracy is lowered.

  An object of the present invention is to provide an information processing system, an information processing method, and a program that can solve the above-described technical problems and improve the accuracy of implication clustering for conversational text.

  As a technical means for solving the above technical problem, the information processing system according to the present invention generates an implication candidate text that is a partial text highly likely to be implied by another partial text from each of one or more texts. An implication candidate extraction means for extracting; an implication candidate extraction means for extracting an implication candidate text that is a partial text that is likely to imply another partial text from each of the one or more texts; Output means for outputting the implication candidate text and the implication candidate text as a plurality of partial texts from which an implication relationship is to be extracted.

  In the information processing method of the present invention, in each of one or more texts, a section that has a high possibility that the predetermined content is clearly specified is extracted as an entailment candidate text, among the sections related to the predetermined content, In each of one or more texts, a section larger than the section related to the predetermined contents including the section related to the predetermined contents is extracted as an implication candidate text, and the extracted implication candidate text and the implication candidate text are extracted. Are output as a partial text from which an implication relationship is to be extracted.

  The program of the present invention extracts, to each of one or more texts, a section having a high possibility that the predetermined contents are clearly specified as an entailment candidate text in the sections related to the predetermined contents. In each of the one or more texts, a section larger than the section related to the predetermined contents including the section related to the predetermined contents is extracted as an implication candidate text, and the extracted implication candidate text and implication candidate A process of outputting the text as a partial text from which an implication relationship is to be extracted is executed.

  The technical effect of the present invention is that the accuracy of implication clustering for conversational text can be improved.

It is a block diagram which shows the basic composition of the 1st Embodiment of this invention. It is a block diagram which shows the structure of the clustering system 1 in the 1st Embodiment of this invention. It is a block diagram which shows the structure of the clustering system 1 implement | achieved by the computer in the 1st Embodiment of this invention. It is a flowchart which shows operation | movement of the clustering system 1 in the 1st Embodiment of this invention. It is a figure which shows the example of the conversation text 81 in the 1st Embodiment of this invention. It is a figure which shows the example of partial text extraction in the 1st Embodiment of this invention. It is a figure which shows the combination of the partial text which performs the implication relation determination process, and the extraction result in the 1st Embodiment of this invention. It is a figure which shows the production | generation result of the group 84 in the 1st Embodiment of this invention. It is a figure which shows the group of the partial text which performs the implication relation determination process, and the extraction result in the 2nd Embodiment of this invention. It is a figure which shows the other example of the group of the partial text which performs the implication relation determination process, and the extraction result in the 2nd Embodiment of this invention.

(First embodiment)
A first embodiment of the present invention will be described.

  In the first embodiment of the present invention, a case will be described as an example in which implication clustering related to a defect occurring in a product is performed based on conversation text 81 in a call center.

  In the first embodiment of the present invention, the implication relationship is defined as follows, as in Patent Document 1. That is, when the content of the second text can be read from the content of the first text, it is defined that the first text implies the second text. Further, if the content of the first text is true, the first text may be defined to imply the second text if the content of the second text is true.

  First, the configuration of the first exemplary embodiment of the present invention will be described.

  FIG. 2 is a block diagram showing a configuration of the clustering system 1 in the first exemplary embodiment of the present invention.

  Referring to FIG. 2, the clustering system 1 according to the first embodiment of the present invention includes a conversation text storage unit 10, a partial text extraction unit 20, a partial text storage unit 30, an implication relationship extraction unit 40, and a group generation unit. 50 is included. The clustering system 1 is an embodiment of the information processing system of the present invention.

  The conversation text storage unit 10 stores one or more conversation texts 81 (or simply text).

  The partial text extraction unit 20 extracts a plurality of partial texts that are targets of clustering (extraction of implications and generation of groups) from the conversation text 81. The partial text extraction unit 20 extracts representative candidates 82 (or implication candidate texts) and member candidates 83 (or implication candidate texts), which will be described later, as such partial texts.

  The partial text storage unit 30 stores the partial text (representative candidate 82 and member candidate 83) extracted by the partial text extraction unit 20.

  The implication relationship extraction unit 40 extracts an implication relationship between partial texts stored in the partial text storage unit 30.

  Based on the implication relationship between the partial texts extracted by the implication relationship extraction unit 40, the group generation unit 50 uses the partial text in the partial text as the representative text, and other partial texts that imply the partial text as members. Generate a group. The representative text is a text that represents the group (can grasp an overview of the group).

  The partial text extraction unit 20 includes an utterance section extraction unit 21, a representative candidate extraction unit 22 (or an implication candidate extraction unit), a member candidate extraction unit 23 (or an implication candidate extraction unit), and a partial text output unit. 24 (or simply an output unit).

  The utterance section extraction unit 21 divides each conversation text 81 into a plurality of utterance sections 91 (or simply sections), and extracts a target section 92 from the plurality of utterance sections 91. The target section 92 is a section (a section related to the predetermined contents) including at least a part of the topics and contents (predetermined contents) to be clustered among the plurality of utterance sections 91. As the predetermined content, when clustering call center conversation text 81 related to a product, for example, a phenomenon such as a defect that has occurred in the product, its cause, countermeasure, or an opinion such as a request, dissatisfaction, evaluation, etc. Used.

  The representative candidate extraction unit 22 represents the target section 92 that has a high possibility that the content of the clustering target is clearly indicated (exactly and concisely) among the target sections 92 extracted by the utterance section extraction unit 21. Extract as candidate 82.

  The member candidate extraction unit 23 extracts a section larger than the target section 92 including the target section 92 extracted by the speech section extraction section 21 as the member candidate 83.

  The partial text output unit 24 outputs the extracted representative candidate 82 and member candidate 83 as a partial text, which is a unit for clustering.

  Here, as described above, the representative candidate 82 is the target section 92 in which the content of the clustering target is highly likely to be clearly indicated. On the other hand, the member candidate 83 includes a target section 92 including at least a part of the contents to be clustered and is a section larger than the target section 92. Therefore, the representative candidate 82 is likely to accurately and concisely represent the contents of the clustering target with a smaller number of words than the member candidate 83. Moreover, although the member candidate 83 includes a redundant portion that is not the content of the clustering target, it is highly likely that the member candidate 83 represents the content of the clustering target. Therefore, if implication clustering is performed using the representative candidate 82 and the member candidate 83 as a unit for clustering, a group is generated in which the representative candidate 82 is set as the representative text and the member candidate 83 that implies the representative candidate 82 is set as the member. There is a high possibility. That is, in the implication clustering, an appropriate group can be generated that includes representative text in which the contents to be clustered are specified and members that imply the representative text.

  The clustering system 1 may be a computer that includes a CPU (Central Processing Unit) and a storage medium that stores a program, and that operates by control based on the program.

  FIG. 3 is a block diagram showing a configuration of the clustering system 1 realized by a computer according to the first embodiment of the present invention.

  The clustering system 1 includes a CPU 2, a storage device (storage medium) 3 such as a hard disk and a memory, a communication device 4 that communicates with other devices, an input device 5 such as a mouse and a keyboard, and an output device 6 such as a display. Including.

  The CPU 2 executes a computer program for realizing the functions of the partial text extraction unit 20, the implication relationship extraction unit 40, and the group generation unit 50. The storage device 3 stores data of the conversation text storage unit 10 and the partial text storage unit 30. The input device 5 receives input of the conversation text 81 from a user or the like. The output device 6 outputs the extracted partial text, implication relationship, and generated group to the user or the like. Further, the communication device 4 may receive the conversation text 81 from another device and transmit the partial text, the implication relationship, and the group to the other device.

  Further, the clustering system 1 may be configured by disposing each component shown in FIG. 2 in a plurality of physical devices connected by wire or wirelessly.

  Next, the operation of the first exemplary embodiment of the present invention will be described.

  FIG. 5 is a diagram showing an example of the conversation text 81 in the first embodiment of the present invention. The conversation text 81 is generated by, for example, voice recognition for voice data recorded at a call center. In FIG. 5, “CU:” and “OP:” given to the conversation texts 81a, 81b,... Indicate that they are utterances of the customer and the operator, respectively.

  Here, it is assumed that the conversation text 81 as shown in FIG. 5 is stored in the conversation text storage unit 10. Note that the conversation text storage unit 10 may store the conversation text 81 of each conversation in association with the voice data of the conversation.

  FIG. 4 is a flowchart showing the operation of the clustering system 1 in the first embodiment of the present invention.

  First, the speech segment extraction unit 21 of the partial text extraction unit 20 divides each conversation text 81 stored in the conversation text storage unit 10 into a plurality of speech segments 91 (step S101).

  Here, for example, the utterance section extraction unit 21 divides the conversation text 81 according to the change of the speaker. In this case, the utterance section extraction unit 21 may detect the change of the speaker based on the sound detection status of the microphone prepared for each speaker. Further, the utterance section extraction unit 21 may detect the change of the speaker by recognizing the speaker based on the voice data.

  Further, the utterance section extraction unit 21 may divide the conversation text 81 into silence sections having a predetermined time length or more.

  Further, the utterance section extraction unit 21 may divide the conversation text 81 in units of grammatical sentences corresponding to the language.

  Further, the utterance section extraction unit 21 may divide the conversation text 81 based on a preset division rule. In this case, for example, an expression or a word that appears at the beginning or end of a statement is set in the division rule. And the utterance area extraction part 21 divides | segments the conversation text 81 by those expressions and words. In this case, the division rule may be a division rule learned by machine learning.

  Note that the utterance section extraction unit 21 may exclude the utterance section 91 including only a greeting or a redundant word from the extraction result of the utterance section 91.

  FIG. 6 is a diagram showing an example of partial text extraction processing in the first exemplary embodiment of the present invention.

  For example, as shown in FIG. 6, the utterance section extraction unit 21 divides the conversation text 81a of FIG. 5 into utterance sections 91a1, a2,... According to the speaker (operator, customer). Similarly, the speech segment extraction unit 21 divides the conversation text 81b into speech segments 91b1, b2,.

  The utterance section extraction unit 21 extracts the target section 92 from the utterance section 91 (step S102).

  Here, the utterance section extraction unit 21 may extract the target section 92 based on, for example, a preset target extraction rule. In this case, for example, an expression or a word used when expressing predetermined contents that are clustering targets is set in the target extraction rule. Then, the utterance section extraction unit 21 extracts the utterance section 91 including those expressions and words as the target section 92. In this case, the target extraction rule may be a target extraction rule learned by machine learning.

  For example, it is assumed that the content of the clustering target is a defect in the product, and “does not move”, “freeze” or the like is set in the target extraction rule as a word indicating the defect. In this case, as shown in FIG. 6, the utterance section extraction unit 21 extracts a utterance section 91a5 including the word “does not move” as the target section 92a1 from the conversation text 81a. Similarly, the speech segment extraction unit 21 extracts speech segments 91b4 and 91b6 including the word “freeze” as the target segments 92b1 and 92b2 from the conversation text 81b.

  The representative candidate extraction unit 22 extracts a representative candidate 82 from the target section 92 extracted by the utterance section extraction unit 21 (step S103).

  Here, the representative candidate extraction unit 22 extracts, as the representative candidate 82, a target section 92 that is highly likely to have specified content that is a clustering target.

  The representative candidate extraction unit 22 extracts, as the representative candidate 82, for example, the target section 92 including the subject and the predicate as the target section 92 that is highly likely to have the predetermined content clearly specified. In this case, the representative candidate extraction unit 22 may exclude the representative candidate 82 including the instruction word in the subject or predicate from the extraction result of the representative candidate 82. Further, the representative candidate extraction unit 22 may extract a target section 92 including a predetermined number or more of body words and predicates as representative candidates 82.

  Further, the representative candidate extraction unit 22 may extract the representative candidate 82 based on a preset representative candidate extraction rule. In this case, the representative candidate extraction rule is set with, for example, a sentence or expression that clearly expresses predetermined contents that are clustering targets. Then, the representative candidate extraction unit 22 extracts a target section 92 including those sentences and expressions as a representative candidate 82. In this case, the representative candidate extraction rule may be a representative candidate extraction rule learned by machine learning.

  For example, when the target section 92 including the subject and the predicate is extracted as the representative candidate 82, the target section 92a1 of the conversation text 81a does not include the subject as shown in FIG. Therefore, the representative candidate extraction unit 22 does not extract the representative candidate 82 for the conversation text 81a. On the other hand, the target section 92b1 of the conversation text 81b includes the subject “PC” and the predicate “freeze”. Therefore, the representative candidate extraction unit 22 extracts the target section 92b1 as the representative candidate 82_1. Similarly, the representative candidate extraction unit 22 extracts the target section 92b2 as a representative candidate 82_2.

  The member candidate extraction unit 23 extracts member candidates 83 from the utterance section 91 extracted by the utterance section extraction unit 21 (step S104).

  Here, the member candidate extraction unit 23 extracts, for example, a plurality of utterance sections 91 including the target section 92 as member candidates 83. In this case, the member candidate extraction unit 23 may extract, as member candidates 83, the utterance section 91 that is a predetermined number before the target section 92 to the utterance section 91 that is a predetermined number after. Further, the member candidate extraction unit 23 may extract the utterance section 91 from a predetermined time before the target section 92 to the utterance section 91 after a predetermined time as member candidates 83. When two different member candidates 83 are overlapped or continuous, the member candidate extraction unit 23 may merge these two member candidates 83.

  Moreover, the member candidate extraction part 23 may extract the member candidate 83 based on the preset member candidate extraction rule. In this case, a sentence, expression, or word that appears at the beginning or end of a topic in a conversation is set in the member candidate extraction rule. Then, the member candidate extraction unit 23 extracts a series of utterance sections 91 including the target section 92 and divided by those sentences, expressions, and words as member candidates 83. In this case, the member candidate extraction rule may be a member candidate extraction rule learned by machine learning.

  For example, when extracting the utterance section 91 immediately before and after the target section 92 as member candidates 83, the member candidate extraction unit 23 starts from the utterance section 91a4 including the target section 92a1 as shown in FIG. Up to 91a6 are extracted as member candidates 83_1. Further, the member candidate extraction unit 23 extracts the speech sections 91b3 to 91b7 including the target sections 92b1 and 92b2 as member candidates 83_2.

  The partial text output unit 24 outputs the extracted representative candidate 82 and member candidate 83 as a partial text, which is a unit for clustering, and stores it in the partial text storage unit 30 (step S105).

  For example, the partial text output unit 24 stores the representative candidates 82_1 and 82_2 and the member candidates 83_1 and 83_2 extracted from the conversation texts 81a and 81b as the partial text in the partial text storage unit 30.

  Next, the implication relationship extraction unit 40 extracts an implication relationship between partial texts stored in the partial text storage unit 30 (step S106). Here, the implication relationship extraction unit 40 extracts the implication relationship between the partial texts by performing, for example, the same determination process as in Patent Document 1. In other words, the implication relationship extraction unit 40 compares the content words included in the partial texts and calculates the coverage rate, thereby determining the presence or absence of the implication relationship. The implication relationship extraction unit 40 has a direction in which one partial text implies the other partial text, and the other partial text is one of the two sets of partial texts stored in the partial text storage unit 30. Judgment processing is performed for directions that imply partial text. Note that the implication relationship extraction unit 40 may determine the implication relationship between the partial texts by a determination process different from that of Patent Document 1 as long as the implication relationship between the partial texts can be extracted.

  FIG. 7 is a diagram showing a set of partial texts for performing implication relationship determination processing and extraction results in the first embodiment of the present invention. In FIG. 7, arrows (thick lines and thin lines) indicate partial text sets and directions in which the determination process is performed. Here, a determination process is performed for a direction in which the original partial text of the arrow implies the partial text of the arrow destination. A bold line indicates that it is determined that there is an implication relationship as a result of the determination process. The thin line indicates that it is determined that there is no implication relationship as a result of the determination process.

  For example, as illustrated in FIG. 7, the implication relationship extraction unit 40 sets the direction in which the member candidate 83_1 implies the representative candidate 82_1 and the representative candidate 82_1 sets the member candidate 83_1 for the set of the representative candidate 82_1 and the member candidate 83_1. The process of determining the direction to imply is performed. Then, the implication relationship extraction unit 40 determines that there is an implication relationship in a direction in which the member candidate 83_1 implies the representative candidate 82_1. Similarly, other groups (representative candidate 82_1 and member candidate 83_2, representative candidate 82_2 and member candidate 83_1, representative candidate 82_2 and member candidate 83_2, representative candidate 82_1 and representative candidate 82_2, member candidate 83_1 and member candidate 83_2) are determined in the same manner. Processing is performed. As a result, the implication relationship extraction unit 40 extracts the implication relationship as shown in FIG.

  Based on the implication relationship between the partial texts extracted by the implication relationship extraction unit 40, the group generation unit 50 creates a group 84 having a partial text as a representative text and another partial text implying the partial text as a member. Generate (step S107).

  FIG. 8 is a diagram illustrating a generation result of the group 84 according to the first embodiment of this invention. For example, based on the implication relationship of FIG. 7, the group generation unit 50 sets the representative candidate 82_1 as representative text, the representative candidate 82_2 implying the representative candidate 82_1, and the member candidates 83_1 and 83_2 as members, as shown in FIG. Group 84_1 to be generated is generated. Similarly, the group generation unit 50 generates a group 84_2 having the representative candidate 82_2 as the representative text and the member candidate 83_2 implying the representative candidate 82_2 as a member.

  The group generation unit 50 may further integrate the two groups into one group based on the degree of duplication of members between two different groups.

  Thus, the operation of the first exemplary embodiment of the present invention is completed.

  In the first embodiment of the present invention, the text to be clustered is the conversation text 81 generated based on the speech data about the conversations of a plurality of speakers, and the content of the clustering target is a defect that has occurred in the product. The case has been described as an example.

  However, the present invention is not limited to this, and text generated based on text-format message data such as chat, e-mail, and electronic bulletin board may be used as the text to be clustered. In addition, text generated for a speech by a single speaker may be used as the text to be clustered. Further, as the contents (topics) to be clustered, various phenomena and events other than defects, their causes, countermeasures, and the like may be used. Further, as contents (topics) to be clustered, phenomena and events in various categories such as weather, disaster, economy, society, etc. may be used. Further, as the contents (topics) to be clustered, speaker opinions such as request, dissatisfaction, and evaluation of speakers in various categories may be used.

  Next, the basic configuration of the first exemplary embodiment of the present invention will be described.

  FIG. 1 is a block diagram showing a basic configuration of the first embodiment of the present invention. Referring to FIG. 1, the clustering system 1 (information processing system) includes a representative candidate extraction unit 22 (implication candidate extraction unit), a member candidate extraction unit 23 (entailment candidate extraction unit), and a partial text output unit 24 (output). Part).

  In each of the one or more conversation texts 81 (text), the representative candidate extraction unit 22 selects a section that has a high possibility that the predetermined content is clearly indicated among the sections related to the predetermined content. Extracted as implication candidate text). In each of the one or more conversation texts 81, the member candidate extraction unit 23 extracts a section that includes a section related to the predetermined content and is larger than the section related to the predetermined content as the member candidate 83 (entailment candidate text). . The partial text output unit 24 outputs the extracted representative candidate 82 and member candidate 83 as a partial text whose implication relationship is to be extracted.

  Next, effects of the first exemplary embodiment of the present invention will be described.

  According to the first embodiment of the present invention, the accuracy of implication clustering for conversational text can be improved. The reason is that in each of the conversation texts 81, among the sections related to the contents of the clustering target, a section where the content of the clustering target is highly likely to be clearly included includes the representative candidate 82 and the section related to the contents of the clustering target. This is because a larger section is extracted as the member candidate 83.

(Second Embodiment)
Next, a second embodiment of the present invention will be described.

  In the first embodiment of the present invention, the implication relation extraction unit 40 has a direction in which one partial text implies the other partial text and one of the other partial texts for all two sets of partial texts. Judgment processing was performed for the direction that implied the partial text. However, when the number of partial texts is large, there is a problem that the number of sets for which the determination process is performed becomes enormous, and the processing time for extracting the implication relationship becomes long.

  Here, it is considered that it is unlikely that an implication relationship exists between the member candidates 83 in the two sets of partial texts. Similarly, the possibility that there is an implication relationship in a direction in which the member candidate 83 implies the representative candidate 82 is considered to be low.

  Therefore, in the second embodiment of the present invention, the implication relation extraction unit 40 excludes such a pair that is unlikely to exist and a direction implication relation among the two sets of partial texts. The determination process is performed only for the implication relationship in the direction in which the member candidate 83 implies the representative candidate 82.

  FIG. 9 is a diagram illustrating a set of partial texts for performing implication relation determination processing and extraction results in the second embodiment of the present invention.

  For example, as shown in FIG. 9, the implication relationship extraction unit 40 implies that the member candidate 83 implies the representative candidate 82 for the set of the representative candidate 82_1 and the member candidate 83_1 and the set of the representative candidate 82_1 and the member candidate 83_2. The determination process is performed for the direction to be performed. Further, the implication relationship extraction unit 40 performs a determination process on the direction in which the member candidate 83 implies the representative candidate 82 for the pair of the representative candidate 82_2 and the member candidate 83_1 and the pair of the representative candidate 82_2 and the member candidate 83_2. . As a result, the meaning relationship extraction unit 40 extracts an implication relationship as shown in FIG.

  Furthermore, the implication relationship extraction unit 40 may perform a determination process in a direction in which the member candidate 83 implies the representative candidate 82 after performing the determination process between the representative candidates 82 first. In this case, when the implication relation extraction unit 40 determines that the member candidate 83 implies the representative candidate 82, and already determines that the representative candidate 82 implies another representative candidate 82, the member candidate 83 is The process of determining the direction implying the other representative candidate 82 is omitted. Then, the implication relationship extraction unit 40 determines that the member candidate 83 implies the other representative candidate 82 without performing the determination process.

  FIG. 10 is a diagram showing another example of a partial text set and extraction result for performing implication relationship determination processing in the second exemplary embodiment of the present invention.

  For example, as illustrated in FIG. 10, the implication relationship extraction unit 40 performs a determination process on a pair of the representative candidate 82_1 and the representative candidate 82_2, and determines that the representative candidate 82_2 implies the representative candidate 82_1. Then, when the implication relation extraction unit 40 determines that the member candidate 83_2 implies the representative candidate 82_2, the member candidate 83_2 does not perform the determination process on the pair of the member candidate 83_2 and the representative candidate 82_1, and the member candidate 83_2 does not represent the representative candidate 82_1. To imply.

  Next, effects of the second exemplary embodiment of the present invention will be described.

  According to the second embodiment of the present invention, it is possible to suppress an increase in the processing time for implication relation extraction even when the number of partial texts is large. The reason for this is that the implication relation extraction unit 40 determines implication relations except for the pair of partial texts and the direction of the implication relations in the pair, which are unlikely to exist and the implication relation of the direction This is for processing.

  While the present invention has been described with reference to the embodiments, the present invention is not limited to the above embodiments. Various changes that can be understood by those skilled in the art can be made to the configuration and details of the present invention within the scope of the present invention.

  The present invention can be applied to a system for overviewing contents included in conversation text related to conversation in a call center or the like, and conversation text in a chat, electronic mail, blog, electronic bulletin board, or the like.

1 Clustering system 2 CPU
DESCRIPTION OF SYMBOLS 3 Memory | storage device 4 Communication device 5 Input device 6 Output device 10 Conversation text memory | storage part 20 Partial text extraction part 21 Speech area extraction part 22 Representative candidate extraction part 23 Member candidate extraction part 24 Partial text output part 30 Partial text storage part 40 Implication relationship Extraction unit 50 Group generation unit 81 Conversation text 82 Representative candidate 83 Member candidate 84 Group 91 Speaking section 92 Target section

Claims (10)

In each of the one or more texts, an implication candidate extraction unit that extracts, as an implication candidate text, an interval that has a high possibility that the predetermined content is clearly specified in an interval related to the predetermined content;
In each of the one or more texts, an implication candidate extracting unit that extracts a section larger than the section related to the predetermined content, including the section related to the predetermined content, as an implication candidate text;
An output means for outputting the extracted implication candidate text and the implication candidate text as a partial text from which an implication relationship is to be extracted;
Information processing system with The implication candidate extraction means extracts, as the implication candidate text, an interval including a subject and a predicate in an interval related to the predetermined content.
The information processing system according to claim 1. The implication candidate extraction means extracts, as the implication candidate text, a section including a predetermined sentence or expression in the section related to the predetermined content.
The information processing system according to claim 1. The implication candidate extraction means extracts a section composed of a plurality of continuous sections including the section related to the predetermined content as the implication candidate text.
The information processing system according to claim 1. Furthermore, an implication relationship extracting means for extracting an implication relationship between the partial texts is provided.
The information processing system according to any one of claims 1 to 4. The implication relationship extraction means is an implication relationship excluding an implication relationship between the implication candidate texts and an implication relationship in which the implication candidate text implies the implication candidate text among the implication relationships between the partial texts. Determine the presence or absence,
The information processing system according to claim 5. When the implication relation extraction unit determines that the implication candidate text implies the other implication candidate text when it is determined that the implication candidate text implies the implication candidate text, the implication candidate text is It is determined to entail the other implication candidate text,
The information processing system according to claim 6. Furthermore, based on the extracted implication relationship, a group generation means for generating a group having a partial text that implies one partial text of the partial text as a member,
The information processing system according to claim 1. In each of the one or more texts, a section having a high possibility that the predetermined content is clearly specified is extracted as an entailment candidate text among the sections related to the predetermined content.
In each of the one or more texts, a section larger than the section related to the predetermined content including the section related to the predetermined content is extracted as an implication candidate text.
The extracted implication candidate text and the implication candidate text are output as a partial text from which an implication relationship is to be extracted.
Information processing method. On the computer,
In each of the one or more texts, a section having a high possibility that the predetermined content is clearly specified is extracted as an entailment candidate text among the sections related to the predetermined content.
In each of the one or more texts, a section larger than the section related to the predetermined content including the section related to the predetermined content is extracted as an implication candidate text.
The extracted implication candidate text and the implication candidate text are output as a partial text from which an implication relationship is to be extracted.
A program that executes processing.

Images