JP2013120514A - Dialogue summarization system and dialogue summarization program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a dialogue summarization system capable of generating a summary sentence from text data of a dialogue content for facilitating understanding of the course of the dialogue.SOLUTION: A dialogue summarization system 1 generates summary data by extracting one or more important sentences from a dialogue content, and has an important sentence extraction section 13. On the basis of dialogue structure data 14 including information on each statement in the dialogue content, score information indicating importance of each statement, and block information in a unit of a continuous statement of each speaker, the important sentence extraction section 13 extracts a statement having the highest score from the dialogue structure data 14 as an important sentence until a predetermined summary condition is satisfied, assigns predetermined scores to a first block from which the important sentence is extracted and a second block in the vicinity of the first block, and assigns and adds the predetermined scores to scores of statements included in the first and second blocks according to a predetermined condition.

Description

  The present invention relates to a technology for generating a summary of a document, and more particularly to a technology effective when applied to a dialog summary system and a dialog summary program for generating a summary by extracting an important sentence from the contents of a dialog.

  For example, in a call center operation, an operator or communicator needs to perform a series of operations such as creating a response record summarizing the contents after a telephone response and storing it in a response management system or the like. At this time, the operator or the like relies on his / her memory to create a response record, which takes time to create and increases the work load. Further, quality (content, uniformity, completeness, conciseness, etc.) varies according to the skill of each operator. In this way, the quality of the response records varies or the quality is low, so it is not possible to accurately grasp the customer's voice, or from the accumulation of response records, new technologies such as text mining are used. Problems such as inability to obtain knowledge (for example, finding potential problems and needs, selecting FAQ candidates, etc.) also occur.

  In order to solve such problems, for example, the contents of the dialogue at the reception are recorded and recorded as voice data, the data is converted into text by using voice recognition technology, and further, natural language processing technology is applied. It is also considered to automatically generate a response record by summarizing the contents. As a result, it is possible to improve the usability of the service record by ensuring the uniformity and simplicity while reducing the cost for creating the service record.

  As such a technique, for example, in Non-Patent Document 1, in voice conversation at a call center, voice data is converted into text, and rules are recursively obtained by inductive learning from past conversation contents and business daily reports. It describes a technology that automatically generates daily business reports by determining and extracting important points for unknown dialogue contents using rules and performing style conversion.

  Various techniques for generating a summary from textual document data have also been proposed. For example, in Japanese Patent Laid-Open No. 2002-259371 (Patent Document 1), an input document is subjected to morphological analysis by a word set extraction device, and a word set necessary as a clue for summarization is extracted from the document according to the summary type. The document dividing apparatus divides the document into a plurality of semantic groups, and calculates an important part having a high appearance density of words included in the word set for each semantic group by the important part calculation apparatus, and summarizes from the important parts. A technique is described in which a summary is extracted by a summary sentence extraction device according to a rate, and a high-precision summary is generated according to the summary type based on the importance in consideration of the appearance density of words.

  Further, for example, in Japanese Patent Application Laid-Open No. 2006-59082 (Patent Document 2), for each sentence included in a plurality of documents to be summarized, a general-purpose sentence importance that is a general-purpose sentence importance is calculated. For each sentence, the importance of the question answer sentence, which is the importance of the sentence as a question answer, is calculated, and the general sentence importance and the question answer sentence importance are integrated, and the integrated sentence importance is the integrated sentence importance Based on the integrated sentence importance, extract important sentences from sentences included in multiple documents to be summarized, and generate summary sentences by aligning the extracted important sentences. In general, a technique for generating a summary sentence that contains important information in a balanced manner is described.

JP 2002-259371 A JP 2006-59082 A

Junji Yano, Kenji Araki, “Performance evaluation of daily report generation method using inductive learning for voice conversation in call center”, Information Processing Society of Japan Research Report 2007-NL-178, March 28, 2007, p. 21-28

  When summaries are automatically generated from text-formatted documents, as with the prior art described above, the importance is usually calculated by various methods for each sentence or word in the document, and the high importance Sentences are extracted in order according to a predetermined summarization condition (for example, extraction is performed until a predetermined summarization rate is reached or all importance levels are equal to or higher than a predetermined value), and a summary sentence is generated by arranging the sentences. The method is taken. As a result, it is possible to extract sentences with high importance in a wide and balanced manner.

  However, simple sentence extraction based on importance as described above, especially when the conversation involves multiple topics or topics, such as conversations between operators and customers in a call center. In the summary sentence generation method by, the context and context of the dialogue cannot be taken into account, and the sentences that are highly important are extracted evenly. It may be a thing.

  SUMMARY OF THE INVENTION An object of the present invention is to provide a dialog summarization system and a dialog summarization program that can generate a summary sentence that can easily understand the connection of a story from text data of a dialog content. The above and other objects and novel features of the present invention will be apparent from the description of this specification and the accompanying drawings.

  Of the inventions disclosed in this application, the outline of typical ones will be briefly described as follows.

  A dialog summarizing system according to a typical embodiment of the present invention is a dialog summarizing system that extracts one or more important sentences from dialog contents and generates summary data composed of the important sentences. Based on dialogue structure data having statement information, score information indicating importance for each statement, and block information in units of consecutive statements for each speaker, until a predetermined summary condition is satisfied, The statement having the highest score is extracted from the dialogue structure data as the important sentence, and a predetermined score is assigned to the first block from which the important sentence is extracted and the second block in the vicinity thereof, and the first And the predetermined score with respect to the score of each statement included in the second block according to a predetermined condition. Have a sentence extraction unit for adding allocated Te is characterized in.

  The present invention can also be applied to a dialog summary program that causes a computer to operate as the dialog summary system as described above.

  Among the inventions disclosed in the present application, effects obtained by typical ones will be briefly described as follows. That is, according to a typical embodiment of the present invention, it is possible to generate a summary sentence that makes it easy to grasp the connection of a story from text data of conversation contents.

It is the figure which showed the outline | summary about the structural example of the dialog summary system which is one embodiment of this invention. It is the figure which showed the example about the outline | summary of the dialog structure in one embodiment of this invention. It is the flowchart which showed the outline | summary about the example of the flow of a process which produces | generates the summary data in one embodiment of this invention. It is the figure which showed the example of the dialog structure data about the dialog of the operator and customer in one embodiment of this invention. It is the figure shown about the example of the procedure which extracts an important sentence, performing score propagation from the statement with the highest importance score in one embodiment of this invention. It is the figure shown about the example of the procedure which extracts an important sentence, performing score propagation from the statement with the highest importance score in one embodiment of this invention. It is the figure shown about the example of the procedure which extracts an important sentence, performing score propagation from the statement with the highest importance score in one embodiment of this invention. It is the figure shown about the example of the procedure which extracts an important sentence, performing score propagation from the statement with the highest importance score in one embodiment of this invention.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. Note that components having the same function are denoted by the same reference symbols throughout the drawings for describing the embodiment, and the repetitive description thereof will be omitted.

  The dialogue summarization system according to an embodiment of the present invention calculates a score consisting of importance for each sentence, and extracts important sentences until the summarizing condition is reached based on the conversation data. Thus, a summary sentence is generated. At this time, it is important not only to extract important sentences in order from the highest score, but also to adjust the score of the important sentence by propagating it to other sentences according to the structure of the dialogue, By extracting the sentence, it is possible to generate a summary sentence that makes it easy to grasp the connection and context of the story.

  In addition, “dialogue” is not limited to two-party, but includes one or more speakers such as those between three or more parties, but in the call center business taken as an example in this embodiment It is desirable that the characteristics of the dialogue can be grasped to some extent in a fixed manner, such as a dialogue between the customer and the operator. For example, in a dialogue between a customer (questioner, consultant) and an operator (respondent), the customer cuts out important topics that can be candidates for inclusion in the form of specific questions and consultations. In response to this, the operator answers (that is, the start and switching of topics are mainly led by the customer), and the topics led by the operator are mainly fixed confirmations and notifications (eg identity verification) ) In many cases.

<System configuration>
FIG. 1 is a diagram showing an outline of a configuration example of a dialog summary system according to an embodiment of the present invention. The dialogue summarization system 1 is composed of, for example, an information processing device such as a PC (Personal Computer) or a server device. The dialogue summarization system 1 extracts dialogue from the dialogue data 21, which is data obtained by converting the dialogue contents into text, and extracts the important sentences from the dialogue. This is a system for generating and outputting summary data 15. Here, the dialogue data 21 is generated by, for example, a voice recognition engine 2 using a known voice recognition technology from voice data obtained by recording a dialogue content between a customer and an operator.

  The dialogue summarization system 1 includes various units such as a dialogue structure analysis unit 11, a scoring unit 12, and an important sentence extraction unit 13 that are implemented as software programs that operate on middleware such as an OS (Operating System) (not shown). .

  The dialog structure analysis unit 11 analyzes the dialog structure of the input dialog data 21 using a known language processing technique, and the analysis result is expressed using XML (eXtensible Markup Language) or the like. It is stored in a database, a file or the like as dialogue structure data 14 by a data structure. FIG. 2 is a diagram showing an example of the outline of the dialog structure. As shown in the example of FIG. 2, the dialog structure analysis unit 11 analyzes at least the dialog data 21 into each utterance sentence (statement) to determine the speaker, and the speaker (in the example of FIG. 2, the operator And customers)). An identification number (O1, O2,..., And C1, C2,... In the example of FIG. 2) may be assigned to each statement for each speaker.

  Furthermore, for use in scoring described later, so-called morpheme analysis is performed on each statement as necessary, and morphemes (the smallest unit having meaning in the language. In the following, “word” is simply described. May be divided into two columns, and analysis processing such as discrimination of parts of speech may be performed. Information on the determination result is also held in the dialog structure data 14. For the morpheme analysis, a known morpheme analysis engine (for example, MeCab (Japanese cloth candy), ChaSen (tea bowl), etc.) can be used. In addition to this, as an analysis of the dialog structure, for example, processing of determining proper nouns, meaning of each statement, etc. (whether it falls under the category of greetings, companions, questions, answers, etc.) May be performed.

  The scoring unit 12 calculates a score indicating importance for each statement held in the dialogue structure data 14 by a predetermined method. The scoring method is not particularly limited, and various methods can be used. In the present embodiment, for example, a TF-IDF (Term Frequency-Inverse Document Frequency) value, which is an index generally used to indicate the importance of a word in a sentence, is used.

  The TF-IDF value includes, for a word in a sentence (statement), a TF value that indicates the frequency of appearance of the word in the statement (which indicates how much is used in the statement) and all statements ( IDF value that is the reciprocal of the number of statements in which the word appears in the entire dialogue (indicates the degree of concentration being used in one statement (not widely used in multiple statements)) Is the product of A word having a large TF-IDF value well represents the feature of the statement and indicates that the word has a high importance. Therefore, for example, the importance for the statement can be scored by summing the TF-IDF values calculated for each word in the statement.

  The TF value and IDF value may be normalized by the total number of word types in the statement and the total number of statements in the dialog, respectively. In that case, you may normalize by a logarithm. For each word in the statement, for example, based on information such as part of speech determined by morphological analysis in the dialog structure analysis unit 11 or information on the contents of the dialog structure (for example, the meaning of each statement) You may weight with respect to the TF-IDF value of a word. The scoring result is stored in the dialog structure data 14 in association with each statement.

  The important sentence extraction unit 13 extracts an important sentence from each statement held in the dialogue structure data 14 based on a score indicating the importance calculated by the scoring unit 12 according to a predetermined summarization condition. Generate. The summarization condition here can be, for example, the summarization rate that is the ratio of the number of characters or statements between the original dialogue data 21 and the summarization data 15, or the number of extracted important sentences. It is good also as extracting a score more than predetermined value.

  When extracting important sentences, as described later, the score of the extracted important sentence is propagated to other statements according to the structure of the dialogue and adjusted, and more important sentences are extracted based on the results. The summary data 15 consisting of the extracted important sentences is made easy to understand the connection and context of the story.

  In the present embodiment, the dialog data 21 is input, the dialog structure is analyzed and scored to generate the dialog structure data 14, and an important sentence is extracted based on this to generate the summary data 15. However, the system configuration is not limited to this. For example, the configuration may be such that the speech recognition engine 2 is included in the dialog summary system 1. Conversely, when the data corresponding to the dialog structure data 14 can be obtained by another system or the like, this is used. As an input, it is possible to adopt a configuration having only a function of extracting the important sentence by the important sentence extracting unit 13 and generating the summary data 15.

<Process flow>
FIG. 3 is a flowchart showing an outline of an example of the flow of processing for generating summary data 15 in the dialog summary system 1. When the dialog summary system 1 starts the process, first, the dialog data 21 converted into text is input, and the dialog structure analysis unit 11 analyzes the dialog structure (S01). Here, as described above, the statements in the dialogue data 21 are grouped into blocks, and each statement is subjected to morphological analysis and divided into words, and other analysis processing is performed as necessary. Next, the scoring unit 12 calculates a score indicating importance for each statement (S02). Here, as described above, for example, the score for each statement is calculated by the sum of the TF-IDF values calculated for each word in the statement.

  Next, the important sentence extraction unit 13 extracts a statement having a maximum score indicating importance as an important sentence (S03). Next, it is determined whether or not a predetermined summarization condition is satisfied (S04). As described above, the summarization condition may be, for example, a summarization rate, the number of extracted important sentences, or a condition in which a score having a predetermined value or more is extracted. If the summary condition is not satisfied, the score is propagated from the statement extracted as the important sentence to other statements according to the structure of the dialogue (S05). As a result, the score of the statement near the statement extracted as the important sentence is raised, and the sentence is easily extracted as the important sentence.

  Specifically, when the score of a statement extracted as an important sentence is S, for example, when an important sentence is extracted from an operator's utterance block in step S03, S / 2 is added to the block, immediately before the block, and A score of S / 4 is assigned to each block of the customer's utterance immediately after. When the important sentence is extracted from the block of the utterance of the customer, the score of S / 2 is assigned to the block, and the score of S / 2 is assigned to the block of the utterance of the operator immediately after the block. The score assigned to each block is allocated and added to each statement in the block (excluding statements already extracted as important sentences) according to the respective scores.

  In this way, a statement (block) that propagates the score differs depending on whether the statement extracted as an important sentence is an operator utterance block or a customer utterance block, for example, an important topic The customer and operator in the call center business, such as the customer cutting out mainly in the form of specific contents of questions and consultations, and the operator answering this (starting and switching of topics is led by the customer) The importance can be adjusted in consideration of the characteristics of the dialogue structure between the two.

  In a state where the score is propagated to the statements in the vicinity of the statement extracted as the important sentence, the process returns to step S03, and the process of extracting the statement having the maximum score as the next important sentence is repeated. If the predetermined summary condition is satisfied in step S04, the important sentences extracted so far are arranged in time series and output as summary data 15 (S06), and the process is terminated.

  At this time, the user may use the output summary data 15 as a summary as it is, or may edit the summary data 15 as a base and create a final summary. At that time, for example, the dialogue summarization system 1 matches, as reference information, a statement that is not extracted as an important sentence but has a high score (for example, a sentence having a score higher than a predetermined value or higher than a predetermined order). May be output and presented.

<Important sentence extraction by importance score>
In the following, an example of a specific process in a method for extracting an important sentence by propagating an importance score from the conversation data 21 will be described with an example of a conversation between an operator and a customer in a call center operation. FIG. 4 is a diagram showing an example of the dialog structure data 14 regarding the dialog between the operator and the customer. Here, the dialogue data 21 is analyzed by the dialogue structure analysis unit 11 and the importance scoring by the scoring unit 12 is arranged in a time series in units of statements for each speaker. Yes.

  The left column (column) of each speaker indicates an identification number (O1, O2,... For an operator, C1, C2,... For a customer), and the right column (column). ) Indicates the importance score of the target statement. Although the statement of the statement itself is omitted, the content of the actual topic across one or more blocks (in the example of FIG. 4 is mainly questions and answers about insurance contracts) and the flow is on the left side of the figure it's shown.

  Here, for example, when five important sentences are extracted as summary conditions, as in the case of the prior art, when the above-described score propagation is not performed, for example, as shown in FIG. The five statements of C8, C4, C9, O14, and C14, which have the top five scores, are extracted as important sentences (when output as summary data 15, C4, C8, C9, O14, and C14 are arranged in this order). At this time, there are three conversation topics: “contract content inquiry”, “dividend inquiry”, and “contract content change consultation”.

  In the present embodiment, an important sentence is extracted while score propagation is performed on the dialog structure data 14 similar to FIG. 5 to 8 are diagrams illustrating an example of a procedure for extracting an important sentence while performing score propagation from a statement having the highest importance score. 5 shows a state where the statement (C8) having the highest importance score among all the statements in the initial state shown in FIG. 4 is selected and extracted as the first important sentence. 5 shows a state in which the score of the statement (C8) extracted in the left diagram is propagated to a nearby statement.

  Here, the statement (C8) extracted as the important sentence is extracted from the block of the customer's utterance. Therefore, as described above, for the score S = 50 of the statement (C8), a score of S / 2 = 25 is assigned to the block, and a score of S / 2 = 25 is assigned to the block of the operator's speech immediately after the block.

  The score assigned to each block is allocated and added to each statement in the block (excluding statements already extracted as important sentences) according to the respective scores. Therefore, in the block where the statement (C8) extracted as the important sentence is extracted, for example, the score of the statement of C7 is 25 * 5 / (5 + 30) = 4 (rounded off to the nearest decimal place) to the original 5. It becomes nine. Similarly, the score of the statement of C9 is 51 by adding 25 * 30 (5 + 30) = 21 to the original 30. It is assumed that it is not assigned to the already extracted C8 statement.

  Similarly, in the block of the operator's utterance immediately after the block where the C8 statement is extracted, the score of the O10 statement is 25 * 5 / (5 + 20 + 10 + 5) = 3 (rounded off to the nearest decimal place). Are added to 8. Similarly, the scores of the statements O11 to O13 are 32, 16, and 8, respectively. In the state of the right diagram where the score has been propagated, the statement (C9) having the highest importance score is selected and extracted as the second important sentence, excluding the extracted statements.

  The left diagram of FIG. 6 shows a state in which the statement C9 is selected and extracted as the second important sentence. 6 shows a state in which the score of the statement (C9) extracted in the left diagram is propagated to a nearby statement. Also here, since the statement (C9) extracted as the important sentence is extracted from the block of the customer's utterance, the score S = 51 of the statement (C9) is set to S / 2 = 26 in the block as described above. A score of S / 2 = 26 is assigned to the block of the operator's speech immediately after the block.

  As a result, the score of the statement C7 is 35, and the scores of the statements O10, O12, and O13 are 11, 22, and 11, respectively. In the state of the right diagram where the score has been propagated, the statement (O11) having the highest importance score is selected and extracted as the third important sentence, excluding the extracted statements.

  The diagram on the left side of FIG. 7 shows a state where the statement of O11 is selected and extracted as the third important sentence. Further, the right diagram in FIG. 7 shows a state where the score of the statement (O11) extracted in the left diagram is propagated to a nearby statement. Here, the statement (O11) extracted as the important sentence is extracted from the block of the operator's utterance. Therefore, as described above, for the score S = 45 of the statement (O11), S / 2 = 23 for the block, and S / 4 = 11 for the customer utterance block immediately before and after the block, respectively. assign.

  As a result, the scores of the statements of O10, O12, and O13 are 17, 34, and 17, respectively. The score of the statement C7 is 46, and the scores of the statements C10 to C12 are 7, 7, and 28, respectively. In the state of the right diagram where the score has been propagated, the statement (C7) having the highest importance score is selected and extracted as the fourth important sentence, excluding the extracted statements.

  The diagram on the left side of FIG. 8 shows a state where the C7 statement is selected and extracted as the fourth important sentence. Further, the diagram on the right side of FIG. 8 shows a state in which the score of the extracted statement (C7) is propagated to a nearby statement. Here, since the statement (C7) extracted as the important sentence is extracted from the block of the utterance of the customer, the score S = 46 of the statement (C7) is set to S / 2 = 23 in the block as described above. The score of S / 2 = 23 is assigned to the block of the operator's speech immediately after the block.

  As a result, the score of each statement of O10 and O13 is 23. In the block including the C7 statement, since all the statements have already been extracted as important sentences, no score is assigned. In the state of the right diagram where the score has been propagated, the statement (O12) having the highest importance score is selected and extracted as the fifth important sentence, excluding the extracted statements.

  Through the above processing, five statements C7 to C9, O11, and O12 are extracted as important sentences. At this time, the topic of dialogue has been narrowed down to “dividend inquiry” that seems to be the most important in the dialogue, and summary data 15 consisting of these statements mainly indicates the contents related to “dividend inquiry”. It becomes easy to grasp the connection of the story.

  There are various methods for propagating a score from a statement having the highest importance score to a nearby statement. In the present embodiment, as described above, when an important sentence is extracted from an operator's utterance block, S / 2 is input to the block, and S / 4 is input to the customer's utterance block immediately before and after the block. When the important sentence is extracted from the block of the utterance of the customer, the score of S / 2 is assigned to the block, and the score of S / 2 is assigned to the block of the utterance of the operator immediately after the block. However, it is not limited to this.

  For example, the allocation ratio of scores to each block is not limited to the above, and may be other ratios, or a predetermined value may be allocated regardless of the value of the score S. . In addition, not only the block immediately before and immediately after the block from which the statement with the highest importance score is extracted, but also a block that is assigned two, two, two, or even before and after, is assigned a certain ratio. May be.

  In addition, when assigning the score assigned to each block to each statement in the block, as in the present embodiment, it is not limited to the distribution according to the importance score of each statement, but equally. You may make it assign, and you may make it assign a predetermined value irrespective of a score. In this embodiment, a score is not assigned to a statement that has already been extracted as an important sentence. However, the present invention is not limited to this, and a score including a statement that has already been extracted is allocated and assigned. Thus, an excessive score may not be assigned to other statements in the block.

  As described above, the technique for propagating the score can be appropriately determined based on an empirical rule or the like according to the type and characteristics of the target conversation. In addition, for example, a statement whose score is less than a predetermined value is excluded from the processing target of important sentence extraction in advance and cut off, depending on the required accuracy of the summary, processing speed, etc. Adjustments may be made as appropriate.

  Further, in the present embodiment, from the first important sentence, the sentence having the highest importance score is automatically selected and extracted from all the statements. The specific statement determined by the designation etc. is made the first important sentence, and the important sentence is extracted sequentially by propagating the score to other statements in the vicinity using the same procedure as above. Also good. Thereby, the summary data 15 related to the specific statement or centered on the specific statement can be generated.

  Further, as described above, the content of the dialogue in the dialogue data 21 is not limited to the dialogue between the customer and the operator in the call center business as exemplified in the present embodiment. Further, the dialog data 21 is not limited to the text data of the actual dialog content converted into text data by the voice recognition engine 2, and may be text data created directly. Therefore, for example, a dialogue in a minutes created after the fact or a fictional dialogue in a novel or the like can be targeted.

  As mentioned above, the invention made by the present inventor has been specifically described based on the embodiment. However, the present invention is not limited to the embodiment, and various modifications can be made without departing from the scope of the invention. Needless to say.

  INDUSTRIAL APPLICABILITY The present invention can be used for a dialog summarization system and a dialog summarization program that extract an important sentence from the contents of a dialog and generate a summary.

1 ... Dialogue summary system, 2 ... Speech recognition engine,
DESCRIPTION OF SYMBOLS 11 ... Dialog structure analysis part, 12 ... Scoring part, 13 ... Important sentence extraction part, 14 ... Dialog structure data, 15 ... Summary data, 21 ... Dialog data.

Claims (11)

A dialog summarization system that extracts one or more important sentences from conversation contents and generates summary data composed of the important sentences,
Based on dialogue structure data having information on each statement in the dialogue content, score information indicating importance for each statement, and block information in units of continuous statements for each speaker,
Until the predetermined summary condition is satisfied, the statement having the highest score is extracted as the important sentence from the dialog structure data, and the first block from which the important sentence is extracted and the second block in the vicinity thereof are extracted. An interactive summary comprising: an important sentence extracting unit that assigns a score, and further assigns and adds the predetermined score according to a predetermined condition to the score of each statement included in the first and second blocks system. The dialog summary system according to claim 1,
The important sentence extraction unit
The dialog summary system, wherein the predetermined score assigned to the first and second blocks is a value of a predetermined ratio with respect to the score of the important sentence. In the dialog summary system according to claim 1 or 2,
The important sentence extraction unit
The dialog summarizing system, wherein the second block is made different according to the speaker information of the first block. The dialog summary system according to claim 3,
When the dialogue content is a dialogue between a questioner or a consultant and a respondent,
The important sentence extraction unit
When the speaker of the first block is the respondent, the block where the questioner or the consultant is the speaker immediately before and after the first block is the second block, respectively.
When the speaker of the first block is the questioner or the consultant, the block immediately after the first block and the respondent is the speaker is the second block. Interactive summarization system. The dialog summary system according to claim 4,
The important sentence extraction unit
If the speaker of the first block is the respondent, ½ of the score of the important sentence is assigned to the first block, and the score of the important sentence is assigned to each of the second blocks. 1/4 of
When the speaker of the first block is the questioner or the consultant, 1/2 of the score of the important sentence is assigned to the first block, and the important sentence is assigned to the second block. An interactive summarization system characterized by assigning 1/2 of a score. In the dialog summary system according to any one of claims 1 to 5,
The important sentence extraction unit
When assigning and adding the predetermined score to the score of each statement included in the first and second blocks, the predetermined score is apportioned according to the score value of each statement. A dialog summarization system characterized by that. In the dialog summary system according to any one of claims 1 to 6,
The important sentence extraction unit
When assigning and adding the predetermined score to the score of each statement included in the first and second blocks, it is not assigned to a statement that has already been extracted as the important sentence. Dialog summarization system. In the dialog summary system according to any one of claims 1 to 7,
A dialogue summarizing system, wherein the first important sentence extracted from the dialogue structure data is selected based on designation from a user. In the dialogue summary system according to any one of claims 1 to 8,
Furthermore, dialogue data in which the dialogue content is converted into text is input, the dialogue data is divided into statements, a speaker is discriminated, and the dialogue structure data is stored in a block in units of continuous statements for each speaker. An interactive structure analysis unit
A dialog summarizing system comprising: a scoring unit that calculates the score for each statement held in the dialog structure data by a predetermined method and stores the score in the dialog structure data. In the dialog summary system according to claim 9,
The dialog structure analysis unit performs a morphological analysis on each statement held in the dialog structure data and divides it into words,
The scoring unit calculates a TF-IDF value in the dialog structure data for each word for each statement held in the dialog structure data, and calculates a total of the TF-IDF values of each word in the statement. A dialog summarization system characterized by a score. An interactive summary program for operating a computer as an interactive summary system that extracts one or more important sentences from interactive contents and generates summary data composed of the important sentences,
Based on dialogue structure data having information on each statement in the dialogue content, score information indicating importance for each statement, and block information in units of continuous statements for each speaker,
Until the predetermined summary condition is satisfied, the statement having the highest score is extracted as the important sentence from the dialog structure data, and the first block from which the important sentence is extracted and the second block in the vicinity thereof are extracted. An important sentence extraction process is performed, in which a score is assigned, and further, the predetermined score is assigned according to a predetermined condition and added to the score of each statement included in the first and second blocks. Summary program.

Images