Conversation support method

Abstract

To provide a conversation support method, device and program that automatically divides a conversation of a certain speaker into multiple paragraphs (groups of sentences, paragraphs) and classifies the divided paragraphs into paragraph categories.SOLUTION: There is provided a conversation support method using a computer, which comprises: a speech analysis step in which a computer analyzes speech related to a conversation to obtain speech words that are words included in the conversation; a paragraph analysis step in which the speech words are used for analyzing the conversation to obtain a group of paragraphs that are multiple paragraphs included in the conversation; a paragraph classification step in which the group of paragraphs is classified to obtain paragraph classification data; and a paragraph proposal step in which a high-evaluation paragraph that is a paragraph with a higher evaluation is proposed for one or more paragraphs included in the group of paragraphs.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present invention relates to a conversation support method using a computer. [Background technology]

[0002] Patent No. 7017822 describes a conversation support method using a computer. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Patent No. 7017822 Summary of the Invention [Problem to be solved by the invention]

[0004] In order to evaluate speakers and improve conversation, it is desirable to automatically divide a speaker's speech into multiple paragraphs (groups of sentences, paragraphs) and classify the divided paragraphs into paragraph classifications.

[0005] It is also hoped that it will guide people to have better conversations. [Means for solving the problem]

[0006] The above problems are basically solved by having a computer automatically classify the paragraph groups contained in a conversation and obtain paragraph classification data. This conversation support method is a method of automatically supporting conversations using a computer. Specifically, the conversation is supported by dividing a speaker's conversation into paragraphs and classifying which paragraph classification each of the paragraphs belongs to, thereby evaluating the conversation and making it possible to introduce better paragraph groups.

[0007] A first invention relates to a conversation support method using a computer, in which a computer performs various steps. The method includes a speech analysis step, a paragraph analysis step, a paragraph classification step, and a paragraph suggestion step.

[0008] The speech analysis process is a process for analyzing speech related to a conversation and obtaining speech words that are words contained in the conversation.

[0009] The paragraph analysis process is a process for analyzing a conversation using spoken words and obtaining a paragraph group which is a plurality of paragraphs contained in the conversation.

[0010] The paragraph classification process is a process for classifying each paragraph included in a paragraph set and classifying which group each paragraph belongs to. Information regarding which group each paragraph belongs to is also called paragraph classification data.

[0011] An example of the paragraph suggestion step is a step of referring to an evaluation value storage unit that stores evaluation values ​​for groups, reading information about a recommended group that is a group that should come next to the conversation using paragraph classification data, and outputting a recommendation keyword or a sentence including a recommendation keyword stored in association with the recommended group, so that the paragraph that comes next to the conversation becomes a paragraph classified as the recommended group. For example, assume that the conversation is in the first paragraph state and the first paragraph belongs to the first group. The evaluation value storage unit stores an evaluation value for each of multiple groups that come next to the first group. The evaluation value is a value that the better the conversation is, the higher the evaluation value becomes. This value may be stored in the evaluation value storage unit, for example, by aggregating multiple conversations and evaluations from users, and setting a high evaluation value for those that receive a high evaluation. Then, assume that the evaluation value storage unit stores that the group next to the first group, which has the highest evaluation value, is the second group. Then, the second group is the recommended group. The storage unit stores one or more recommendation keywords in association with the recommended group (second group). Therefore, the computer may read out recommendation keywords from the storage unit and output them as is, or may create a sentence using the recommendation keywords (or read out a sentence stored in the storage unit) and output it.

[0012] A preferred example of the present invention is a step in which the paragraph analysis step obtains a paragraph group using phonetic words contained in each paragraph in the paragraph group and information about paragraph analysis keywords for analyzing the paragraphs based on the phonetic words.

[0013] In a preferred embodiment of the present invention, each paragraph included in the paragraph group corresponds to a presentation document or a page of the presentation document.

[0014] The second invention includes a speech analysis step, a paragraph analysis step, a paragraph classification step, and a paragraph suggestion step. The paragraph suggestion step in the second invention is a paragraph suggestion step of referring to an evaluation value storage unit that stores evaluation values ​​for groups, obtaining an evaluation value for the conversation using paragraph classification data, and outputting a keyword stored in association with the group with a higher evaluation value or a sentence including a keyword stored in association with the group with a higher evaluation value when there is a group with a higher evaluation value. For example, assume that a conversation includes groups 1, 2, and 3 in this order. The evaluation value storage unit stores an evaluation value when groups 1, 2, and 3 are included in this order. On the other hand, the evaluation value storage unit stores an evaluation value when groups 1, 2, 4, and 5 are included in this order. When the latter evaluation value is higher than the former evaluation value, the system uses the paragraph classification data to obtain information about the groups included in the conversation (including groups 1, 2, and 3 in this order), and reads out the evaluation value of the conversation from the evaluation value storage unit. The system then reads out information about a group (including group 1, group 2, group 4, and group 5, in this order) that has a higher evaluation value than the evaluation value of the conversation read out from the evaluation value storage unit, and outputs keywords stored in association with each group in this group. The system may create and output a sentence using the keywords read out at this time. The system may also output a sentence stored in association with each group in this group. In this case, if the same group is included in the group included in the conversation and the group included in the group with a high evaluation value, the system may output keywords stored in association with groups not included in the conversation, among the groups included in the group with a high evaluation value, or sentences including such keywords.

[0015] A preferred example of the present invention further includes a step of displaying groups included in the conversation, keywords included in each of the groups included in the conversation, groups with high evaluation values, and keywords included in each of the groups with high evaluation values.

[0016] A preferred example of the present invention further includes a step of displaying the transition of the evaluation value for each group included in the conversation and the transition of the evaluation value for each group with a high evaluation value.

[0017] Another invention of this specification relates to a program for causing a computer to execute any of the above-mentioned inventions, and a non-transitory information recording medium storing the program. Effect of the Invention

[0018] This invention can automatically divide a speaker's speech into multiple paragraphs (groups of sentences, paragraphs) and classify the divided paragraphs into paragraph categories. Furthermore, the present invention can evaluate a conversation for each paragraph and output a better paragraph or keywords contained in the better paragraph for each paragraph, thereby encouraging improvement of the conversation. [Brief description of the drawings]

[0019] [Figure 1] FIG. 1 is a flowchart for explaining a conversation support method using a computer. [Diagram 2] FIG. 2 is a block diagram of a conversation support device that implements the conversation support method. [Diagram 3] FIG. 3 is a conceptual diagram showing examples of keywords contained in each paragraph. [Figure 4] Figure 4 shows a graph visualizing the evaluation of the conversation and the evaluation of the hypothetical conversation after paragraph suggestions. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0020] Hereinafter, the embodiments for carrying out the present invention will be described with reference to the drawings. The present invention is not limited to the embodiments described below, and includes appropriate modifications of the embodiments described below within the scope obvious to those skilled in the art.

[0021] FIG. 1 is a flowchart for explaining a conversation support method using a computer. As shown in Fig. 1, the conversation support method using a computer includes a speech analysis step (S101), a paragraph analysis step (S102), and a paragraph classification step (S103), and may further include a paragraph suggestion step (S104). S means step.

[0022] Fig. 2 is a block diagram of a conversation support device that implements this method. As shown in Fig. 2, the conversation support device 1 that implements this method is a computer-based device, and may include a voice analysis unit 3 that performs a voice analysis step (S101), a paragraph analysis unit 5 that performs a paragraph analysis step (S102), and a paragraph classification unit 7 that performs a paragraph classification step (S103). This device may also have a paragraph suggestion unit 9 that performs a paragraph suggestion step (S104). Each part may be read as each means.

[0023] The computer has an input unit, an output unit, a control unit, a calculation unit, and a storage unit, and each element is connected by a bus or the like so as to be able to exchange information. For example, the storage unit may store a control program or various information. When predetermined information is input from the input unit, the control unit reads out the control program stored in the storage unit. Then, the control unit appropriately reads out the information stored in the storage unit and transmits it to the calculation unit. The control unit also appropriately transmits the input information to the calculation unit. The calculation unit performs calculation processing using the various information received and stores it in the storage unit. The control unit reads out the calculation results stored in the storage unit and outputs them from the output unit. In this way, various processes and each step are executed. The various parts and means execute these various processes. The computer may have a processor, and the processor may realize various functions and various steps. The computer may be standalone. A part of the computer's functions may be distributed to a server and a terminal. In that case, it is preferable that the server and the terminal are able to exchange information through a network such as the Internet or an intranet.

[0024] Voice analysis process (S101) The voice analysis process is a process in which a computer analyzes voice related to a conversation and obtains voice words that are words included in the conversation. The conversation may be a business negotiation, an explanation of a product, or a presentation. In the following example, a conversation between multiple people will be described. However, a conversation also includes a statement by one person (such as a presentation). Voice information of the conversation and digital information based on the conversation are input via the input unit (microphone or interface) of the conversation support device (computer). The input information about the conversation is appropriately stored in the memory unit of the computer. The computer reads information about the conversation from the memory unit and a program from the memory unit, analyzes the voice related to the conversation, and obtains voice words that are words included in the conversation. At this time, the dictionary included in the memory unit may be referred to as appropriate. In addition, if the conversation is based on a presentation, the memory unit may store a specific dictionary that stores terms related to the presentation, and the voice analysis process may refer to the specific dictionary to obtain the correct voice words. The following are examples of voice words.

[0025] MR) Thank you for your continued support, Professor. Today, I would like to introduce the newly released DPP4 inhibitor Aipro tablets. Dr.) (But) there are so many available, why is it being released again? I don't plan on using any more new DPP4 inhibitors. MR) It is true that there are already several DPP4 inhibitors on the market, but our Aipro tablets have some advantages over other drugs. Which DPP4 inhibitor are you currently using? Dr.) There are so many options, so I use different ones depending on the patient's lifestyle. I've been prescribing Putin tablets more often recently because they have a strong effect after just one dose. MR) I see. Our AiPro tablets have extremely high selectivity for DPP4. It has 80 times the selectivity compared to Putin tablets. This is why in terms of effectiveness, it tends to have a stronger blood sugar lowering effect than Putin tablets. Dr.) But the medical representative over there said, "Selectivity has no effect on efficacy." Is it more effective than Putin tablets? MR) Yes. The change in HbA1c from baseline with Aipro tablets was 0.9%. On the other hand, Putin tablets had a rate of about 0.5%. High selectivity means high efficacy and, because smaller doses are required, high safety. Doctor) I see. It seems to be very effective, so I'll consider it. MR) Please consider prescribing this to patients who have trouble controlling their blood sugar levels.

[0026] In the above, the computer may distinguish between speakers (MR or Dr) based on the frequency of the voice related to the conversation, or may distinguish based on the voice input unit (microphone, etc.), or may not distinguish at all.

[0027] Paragraph analysis process (S102) The paragraph analysis step is a step in which a computer uses speech words to analyze a conversation and obtain a paragraph group, which is a plurality of paragraphs included in the conversation. A paragraph is a set of sentences with a certain meaning. Paragraphs may be divided into paragraphs each time a speaker changes. A conversation is composed of a paragraph group consisting of n paragraphs, for example, the first paragraph, the second paragraph, ... For example, one or more keywords included in the m-th paragraph are stored in a storage unit, and when the computer interprets that the keyword is included in a predetermined number or more of speech words, it may analyze that the m-th paragraph has been spoken. Since a conversation usually progresses as the first paragraph, the second paragraph, ..., after the computer interprets that the first paragraph has been spoken, the computer may read out a keyword related to the second paragraph and analyze whether it is included in the speech words. In this way, a computer can automatically analyze a conversation into a plurality of paragraphs.

[0028] When multiple groups of Web interviews are conducted, there will be similar conversations among them. The computer evaluates the conversations using the evaluation method described below, and groups and classifies the conversations using the speech words contained in the similar conversations stored in the memory unit, dividing them into conversations that go well and those that do not. In this way, the computer may obtain the paragraph groups contained in the conversations, the classification of each paragraph, and the keywords contained in each classification, and store them in the memory unit.

[0029] Also, for example, words contained in each page of the presentation materials may be extracted, and a page-related word database may be prepared that stores information about each page (e.g., page B of presentation A) and words and related words associated with each page.The paragraph analysis unit of the computer may then compare the speech words with the page-related word database to determine which page the conversation or presentation is on, and thereby divide the speech words into paragraphs.

[0030] A preferred example of the paragraph analysis step is to obtain a group of paragraphs using information on keywords contained in each of a plurality of paragraphs contained in a conversation and speech words. It is preferable that the keywords contained in each paragraph are obtained by collating a plurality of conversations and analyzing the keywords contained in each paragraph. That is, the system stores in a storage unit conversations and presentations given by a plurality of people using a certain presentation material or a certain topic (e.g., a certain product). Then, speech words contained in the conversations and presentations are obtained and stored in the storage unit. At that time, the speech words may be stored in association with each page of the presentation material. Then, the system may obtain paragraphs and keywords (speech words) for each classification contained in the paragraphs by obtaining matches and mismatches of speech words contained in the plurality of conversations or presentations, and store them in the storage unit. Also, the system may obtain speech words used for each page of the presentation material, analyze the patterns to obtain the classification of the page, and store characteristic speech words for each classification as keywords in the storage unit.

[0031] The method of dividing a paragraph may be to analyze a conversation between multiple speakers and distinguish a set in which different terms appear as a paragraph. This paragraph division may be performed automatically by a computer by comparing the recorded voice words, and classification words (keywords) for classifying each paragraph may be stored in a memory unit. For example, the utterance "excellent" may be treated as an NG word. In other words, the classification words may include NG words, and NG words may be stored in the memory unit as a type of keyword. An NG word is a term that is not desirable to use (must not be used).

[0032] FIG. 3 is a conceptual diagram showing examples of keywords contained in each paragraph. In the example of FIG. 3, the paragraphs following the paragraphs shown in topic groups 1 and 2 (and the keywords classified into those paragraphs) are different. The computer can then compare the spoken words contained in the paragraphs following topic groups 1 and 2 with the keywords stored in the memory unit, and classify the paragraphs following the paragraphs shown in topic groups 1 and 2 of that conversation. It is also possible to analyze who is speaking in a conversation, and analyze and classify paragraphs based only on the speech of a certain person.

[0033] It is preferable that each of the multiple paragraphs corresponds to a presentation material or a page of the presentation material. In this case, a keyword for classifying the presentation material or each page of the presentation material into paragraphs is stored in the storage unit, and the paragraphs are analyzed (and classified) using the keyword. For example, the first page of the presentation material can be classified into paragraphs 1 to 3. Then, in relation to the first paragraph, keyword A 1,1,1 ,Keyword A 1,1,2 ,Keyword A 1,1,3 ,Keyword A 1,1,4 , ... 1,2,1 ,Keyword A 1,2,2 ,Keyword A1,2,3 ,Keyword A 1,2,4 ,... are stored. Some of the keywords related to these different paragraphs may be the same. By using such a group of keywords, the paragraphs contained in the conversation can be analyzed and classified.

[0034] An example of the paragraph analysis process is as follows: By using these classification keywords, speech analysis can be performed more accurately.

[0035] Topic Group 1 (1st paragraph) MR) Thank you for your continued support, Professor. Today, I would like to introduce the newly launched DPP4 inhibitor, Aipro Tablets. Dr.) (But) there are so many available, why is it being released again? I don't plan on using any more new DPP4 inhibitors. MR) It is true that there are already several DPP4 inhibitors on the market, but our company's Aipro tablets have some advantages over other drugs. Which DPP4 inhibitor are you currently using? Dr.) There are so many options, so I use different ones depending on the patient's lifestyle. I've been prescribing Putin tablets more often recently because they have a strong effect after just one dose.

[0036] In this example, the terms "new release," "DPP4 inhibitor," "Aipro tablets," "other drugs," and "compared to" are stored in the memory as keywords related to topic group 1 of the first paragraph. The computer compares the spoken words with the keywords and analyzes the above conversational portion as the first paragraph (topic group 1). In the above example, the quotation marks are spoken words that match the keywords related to topic group 1.

[0037] Topic Group 2 (2nd paragraph) MR) I see. Our company's "EyePro Tablets" have extremely high selectivity for DPP4, and are 80 times more "selective" than "Putin" tablets. This is why in terms of effectiveness, it tends to have a stronger blood sugar lowering effect than Putin tablets. Dr.) But the medical representative over there said, "Selectivity has no effect on efficacy." Is it more effective than Putin tablets?

[0038] Topic Group 3 (3rd paragraph) MR) Yes. The change in HbA1c from the baseline with AiPro tablets was 0.9%. On the other hand, Putin tablets had a rate of about 0.5%. High selectivity means high effectiveness and, since smaller doses are required, high safety. Doctor) I see. It seems to be very effective, so I'll consider it. MR) Please consider prescribing this to patients who have trouble controlling their blood sugar.

[0039] Paragraph classification process (S103) The paragraph classification process is a process in which a computer classifies paragraph groups and obtains paragraph classification data. Paragraph classification data means information about which group each paragraph group belongs to when it is divided into several groups. By using this information, the paragraphs can be classified.

[0040] A preferred example of the paragraph classification step is to classify each of a plurality of paragraphs contained in a conversation into two or more paragraph classes according to a plurality of keywords contained in each of the plurality of paragraphs contained in the conversation, and to classify each of a plurality of paragraphs contained in the conversation according to which keyword of the two or more classified paragraph classes the speech words match. In the example described above, the paragraphs are classified in the paragraph analysis step.

[0041] The paragraph classification step may be performed simultaneously with the paragraph analysis step, or may be performed separately. An example of performing the paragraph classification step and the paragraph analysis step separately is to have a paragraph classification storage unit that stores paragraph classification keywords for determining which group each paragraph belongs to, and to analyze which group each paragraph belongs to by comparing speech words contained in each paragraph with the paragraph classification keywords.

[0042] For example, in the above example, the paragraph classification unit may refer to the paragraph classification storage unit using the speech words included in the first paragraph to determine that the first paragraph belongs to the first group. The paragraph classification unit may refer to the paragraph classification storage unit using the speech words included in the second paragraph to determine that the second paragraph belongs to the second group. The paragraph classification unit may refer to the paragraph classification storage unit using the speech words included in the third paragraph to determine that the third paragraph belongs to the third group.

[0043] Paragraph proposal process (S104) The paragraph suggestion process is a process in which the computer proposes a highly rated paragraph, which is a paragraph with a higher evaluation, for one or more paragraphs included in the paragraph group based on the paragraph classification data. This process makes it possible to recommend examples of speakers with higher evaluations. For example, when a certain paragraph is classified into five classifications, if there is a classification with a higher evaluation value than the classification spoken by the speaker, the classification with the higher evaluation value can be suggested, thereby promoting better conversation. In this case, the evaluation value may be stored in the storage unit in association with each paragraph classification, and after determining which classification the paragraph belongs to, the paragraph with the higher evaluation value or keywords related to the paragraph may be output. An example of the output is to display the paragraph with the higher evaluation value or keywords related to the paragraph on the display unit. Then, the speaker of the conversation can grasp the paragraph with the higher evaluation value or keywords related to the paragraph for each paragraph, and can use them in the next explanation or presentation.

[0044] For example, the system has an evaluation value storage unit that stores evaluation values ​​for groups. Then, the paragraph suggestion unit receives information from the paragraph classification unit that the first paragraph belongs to the first group. The evaluation value storage unit stores the second group as the group that comes next to the first group. In other words, the evaluation value storage unit stores that a conversation in which the second group comes next to the first group has a higher evaluation than a conversation in which another group comes next to the first group. Then, the paragraph suggestion unit outputs the recommendation keyword stored in association with the second group or a sentence including the recommendation keyword during a presentation (conversation) about the first group or after a presentation (conversation) about the first group has been made. Then, the recommendation keyword or a sentence including the recommendation keyword is output to the terminal. In this way, the system can support the presenter to talk about the second group after the first group.

[0045] The evaluation value storage unit also stores an evaluation value when the third group is spoken after the first group and the second group. The evaluation value storage unit also stores an evaluation value when the fourth and fifth groups are spoken after the first and second groups. If the latter is higher, the paragraph suggestion unit receives information from the paragraph classification unit that the second paragraph following the first group belongs to the second group. Then, the paragraph suggestion unit outputs the recommendation keyword stored in association with the fourth group or a sentence including the recommendation keyword during a presentation (conversation) about the second group or after a presentation (conversation) about the second group has been made. Then, the recommendation keyword or a sentence including the recommendation keyword is output to the terminal. In this way, the system can support the presenter to speak about the fourth group after the second group.

[0046] Methods and devices for evaluating conversations (including presentations) are well known. For example, Patent No. 7049010 describes a presentation evaluation system. This system includes a voice analysis unit that analyzes the content of the conversation, a presentation material related information storage unit that stores information about the presentation material, the information about the presentation material including information for identifying each page of the presentation material, a keyword storage unit that stores keywords for each page of the presentation material, a related word storage unit that stores related words to the keywords, and an evaluation unit that evaluates the content of the conversation analyzed by the voice analysis unit or the people who conducted the conversation. The evaluation unit then identifies each page of the presentation materials based on the information on the presentation materials stored in the presentation material related information storage unit, reads out keywords related to the identified page of the presentation materials from the keyword storage unit, reads out related words of the keywords related to the identified page of the presentation materials from the related word storage unit, and reads out evaluation values ​​related to the number of keywords for each page of the presentation materials included in the content of the conversation analyzed by the voice analysis unit, the number of related words of the keywords related to the identified page of the presentation materials, the combination of keywords related to the identified page of the presentation materials, or the combination of related words of the keywords related to the identified page of the presentation materials from the storage unit using the number of keywords for each page of the presentation materials included in the content of the conversation analyzed by the voice analysis unit, the number of related words of the keywords related to the identified page of the presentation materials, the combination of keywords related to the identified page of the presentation materials, or the combination of related words of the keywords related to the identified page of the presentation materials, obtains the read-out evaluation values, and, if there are multiple read-out evaluation values, adds up the read-out evaluation values ​​to obtain an evaluation value for evaluating the content of the conversation or the conversation participants using the read-out evaluation values. In this way, the conversation (presentation) and each paragraph can be evaluated. The obtained evaluation value may be stored in a storage unit as appropriate.

[0047] FIG. 4 is a graph visualizing the evaluation of a conversation and the evaluation of a virtual conversation after a paragraph proposal. In this example, a conversation progressed as shown in topic group 1, topic group 2, and topic group 3 as a result of computer analysis. The third paragraph of topic group 3 would have a higher evaluation value if it progressed as shown in topic group 4 and topic group 5. The computer stores the evaluation value of the classification included in each paragraph, so it reads out the evaluation value of topic group 3 included in the third paragraph and the evaluation values ​​of topic groups 4 and 5 included in the third paragraph, compares them, and outputs the keywords included in topic groups 4 and 5. The computer also reads out the evaluation value of each paragraph, finds the change in the evaluation value of the actual conversation and the change in the evaluation value when a virtual conversation based on the paragraph proposal is held, graphs them, and outputs them. In this way, it is possible to encourage conversations and presentations according to better paragraphs, and since the effect can be visually confirmed by visualization, it is also possible to make the speaker more persuasive.

[0048] Furthermore, examples of presentations based on the same presentation materials are collected, divided into paragraphs, and then divided into groups. Then, for presentations (conversations) that have been well received, the transition of groups when the presentation is given is stored in the storage unit. In this way, the evaluation value storage unit that stores the evaluation values ​​for the groups can be updated. For example, the evaluation value storage unit stores the transition of evaluation values ​​(see FIG. 4) in association with presentations that have progressed from groups 1, 2, 4, and 5. The evaluation value storage unit also stores the transition of evaluation values ​​(see FIG. 4) in association with presentations that have progressed from groups 1, 2, and 3. In this way, by having a certain presentation be presented (conversation) by multiple people, dividing the presentations into multiple groups, and storing the evaluation values, the flow of the groups of talk of those who have given good presentations can be updated, and if a certain speaker's presentation is lower than that evaluation value, it can be guided to a group that is given a higher evaluation value. Also, in the above example, if the system determines that the presentation has progressed from groups 1 and 2, the keywords included in group 4 can be displayed on the presenter's terminal to encourage the presentation to progress to groups 4 and 5, and thus it is possible to guide the presenter to give a presentation that is highly evaluated during the presentation.

[0049] Furthermore, if the actual conversation (presentation) progresses from group 1 to group 2 to group 3, the paragraph suggestion unit reads out information on the presentation that progressed to groups 1, 2, 4, and 5, which have higher evaluation values ​​than the evaluation values ​​associated with the presentation that progressed to groups 1, 2, and 3, from the evaluation value storage unit. Then, as shown in FIG. 3, the flow of groups that can obtain a higher evaluation than the actual conversation is output. In this way, after the actual conversation, support can be provided for having a conversation with a higher evaluation. Furthermore, by outputting the transition of evaluation values ​​as shown in FIG. 4, it is possible to encourage a more persuasive and good conversation.

[0050] The program of the present invention is for causing a computer to implement the above-mentioned method. The program of the present invention is a program for causing a computer to execute a method including a speech analysis step, a paragraph analysis step, and a paragraph classification step. The program of the present invention may further cause the computer to execute a paragraph suggestion step.

[0051] The non-transitory information recording medium of the present invention is a non-transitory information recording medium that can be read by a computer that stores the above-mentioned program. Examples of the non-transitory information recording medium are a CD-ROM, a DVD, and a USB memory.

[0052] This invention can be implemented, for example, as an application program in a user's terminal. For example, a dialogue / role-playing data from a web interview conducted by a certain user (A) is used and re-conducted using an app. Then, the terminal on which this application is implemented recommends (keywords in) the highly rated paragraphs of others (B and C) and displays them on the display unit of the terminal. Then, when the user selects an example of a reply from B, the paragraph in A's conversation that was highly rated by others is changed to the paragraph spoken by B. By listening to the conversation after this change, it is possible to learn conversations that will get better evaluations. This processing can be performed by using a computer. In this way, conversations for each paragraph are stored in the storage unit, and when there is a paragraph of another person that is highly rated (highly rated paragraph), if there is a paragraph in a certain conversation that is rated lower than the paragraph of the other person (lowly rated paragraph), the computer may display the keywords contained in the highly rated paragraph of the other person on the display unit, or may replace the low-rated paragraph with a highly rated paragraph and store it. At this time, the voice (frequency band) of the replaced highly rated paragraph may be converted to the voice (frequency band) of a certain speaker and stored in the storage unit. In this way, it becomes possible to play back the conversation in which the low-rated paragraph has been replaced with the highly rated paragraph. [Industrial Applicability]

[0053] This invention can be used in the information industry. [Explanation of symbols]

[0054] 1. Conversation support device 3. Audio Analysis Section 5 Paragraph Analysis 7 Paragraph Classification Section 9. Paragraph Suggestions

Claims

1. A conversation support method using a computer, comprising: The computer, a speech analysis step for analyzing speech relating to a conversation to obtain speech words which are words contained in the conversation; a paragraph analysis step of analyzing the conversation using the speech words to obtain a paragraph group that is a plurality of paragraphs included in the conversation; a paragraph classification step of classifying each paragraph included in the paragraph group and obtaining paragraph classification data regarding to which group each of the paragraphs belongs; a paragraph suggestion step of referring to an evaluation value storage unit that stores evaluation values ​​for groups, using the paragraph classification data to read out information on a recommended group that is a group that should come next in the conversation, and outputting a recommendation keyword or a sentence including a recommendation keyword that is stored in association with the recommended group, thereby encouraging the paragraph that comes next in the conversation to be classified as the recommended group; A method, wherein each paragraph in the set of paragraphs corresponds to a presentation material or a page of a presentation material.

2. To the computer, a speech analysis step for analyzing speech relating to a conversation to obtain speech words which are words contained in the conversation; a paragraph analysis step of analyzing the conversation using the speech words to obtain a paragraph group that is a plurality of paragraphs included in the conversation; a paragraph classification step of classifying each paragraph included in the paragraph group and obtaining paragraph classification data regarding to which group each of the paragraphs belongs; a paragraph suggestion step of referring to an evaluation value storage unit that stores evaluation values ​​for groups, using the paragraph classification data to read out information on a recommended group that is a group that should come next in the conversation, and outputting a recommendation keyword or a sentence including a recommendation keyword that is stored in association with the recommended group, thereby encouraging the paragraph that comes next in the conversation to be classified as the recommended group; Each paragraph included in the paragraph group corresponds to a presentation document or a page of the presentation document; A program for carrying out the method.

3. A non-transitory information recording medium storing the program according to claim 2.

Images