JP2017016566A - Information processing device, information processing method and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an information processing device for improving the convenience of a participant of a conference more by processing text data acquired during the conference.SOLUTION: The information processing device includes a visualization screen generation part for classifying text data issued by a plurality of speakers during a conference into prescribed categories on the basis of intentions, modality and potential topics of the speakers, and generating a visualization screen being a screen obtained by visualizing classification results classified into the prescribed categories in association with the speakers, and generates any visualization screen of the transition of the intentions of utterances, a transition phase of the potential topics and the transition of the contents of the text data in the case that a state in which a debate is performed is predicted.SELECTED DRAWING: Figure 1

Description

  The present disclosure relates to an information processing apparatus, an information processing method, and a program.

  Systems that support the progress of meetings and the creation, approval, and management of minutes have been developed (for example, Patent Documents 1-4). By using these systems, it is expected that the conference will be more efficient.

JP 2007-108865 A JP 2010-72784 A JP 2005-175627 A JP 2008-77495 A

  However, the systems shown in Patent Documents 1-4 still have room for improvement. For example, in the technologies described in Patent Documents 1 and 2, the main purpose is to share information generated in a meeting such as minutes, etc. among the participants of the meeting, and the information is classified according to the contents. Analysis has not been performed. Patent Document 3 discloses a technique for extracting text data from audio information recording a speech during a meeting and image information obtained by photographing a white board, and automatically generating a minutes based on the text data. These text data are not classified.

  Further, in Patent Document 4, a keyword is extracted from text data acquired during a meeting, a time-series transition of the extracted keyword is analyzed, and the keyword is weighted according to the appearance frequency, A technique for structuring the hierarchical relationship between the keywords is disclosed. In this technology, keywords at each time point of the conference can be grasped, but the speech during the conference is not classified.

  In view of the above circumstances, there is a possibility that the text can be more efficiently processed and the convenience of participants of the conference can be further improved by appropriately processing the text data acquired during the conference. Therefore, the present disclosure proposes a new and improved information processing apparatus, information processing method, and program capable of further improving the convenience of conference participants.

  According to the present disclosure, a visualization screen generation unit that generates a visualization screen that is a screen in which text data issued by a plurality of speakers is classified into a predetermined category and visualized in association with the speaker. An information processing apparatus is provided.

  Further, according to the present disclosure, the processor generates a visualization screen that is a screen in which text data generated by a plurality of speakers is classified into a predetermined category and visualized in association with the speaker. An information processing method is provided.

  In addition, according to the present disclosure, a visualization for generating a visualization screen, which is a screen on which a classification result obtained by classifying text data issued by a plurality of speakers into a predetermined category is visualized in association with the speaker. A program for realizing the screen generation function is provided.

  According to the present disclosure, text data uttered by a plurality of speakers is classified into a predetermined category, and the classification result is visualized in association with the speaker. The predetermined category may be, for example, a speaker's intention. Since the visualized classification results can give the participants of the conference awareness of the progress of the conference, the participants can make the smoother by proceeding with the conference while referring to the classification results. You can proceed with the meeting. Therefore, the convenience of the conference participants can be improved.

  As described above, according to the present disclosure, it is possible to further improve the convenience of conference participants. Note that the above effects are not necessarily limited, and any of the effects shown in the present specification, or other effects that can be grasped from the present specification, together with the above effects or instead of the above effects. May be played.

It is explanatory drawing for demonstrating the outline | summary of the system which concerns on this embodiment. It is a block diagram which shows an example of a function structure of the system which concerns on this embodiment. It is a functional block diagram which shows an example of a function structure of the opinion extraction part shown in FIG. It is explanatory drawing for demonstrating the opinion extraction process by an opinion extraction part. It is a flowchart which shows an example of the process sequence of the information processing method which concerns on this embodiment. It is a flowchart which shows an example of the process sequence of the prediction process by the situation prediction part whether the presentation is performed by one participant, and the process by the visualization screen generation part according to the result of the said prediction process. It is a figure which shows the visualization screen which visualized the transition of opinion type, the transition of a latent topic, and the transition of a keyword which is an example of a visualization screen. It is a flowchart which shows an example of the process sequence of the process which detects the change point of a latent topic. It is a figure which shows the visualization screen which visualized the speech condition of the participant in a meeting which is another example of a visualization screen. It is a figure which shows the visualization screen which visualized the transition of the utterance situation of the participant in a meeting and the opinion type of the said utterance which is another example of a visualization screen. It is a figure which shows the visualization screen which visualized the transition of the content of an utterance and the transition of the opinion type of the said utterance which is another example of a visualization screen. It is a figure which shows the visualization screen on which the result of the positive / negative determination was visualized which is another example of a visualization screen. It is a figure which shows the visualization screen as which the classification result of the text data using a modality was visualized which is another example of a visualization screen. It is a figure which shows the visualization screen in the case where a meeting is performed on social media which is another example of a visualization screen. It is a figure which shows the visualization screen in the case where a meeting is performed on social media which is another example of a visualization screen. It is a block diagram which shows an example of the hardware constitutions of the information processing apparatus which concerns on this embodiment.

  Hereinafter, preferred embodiments of the present disclosure will be described in detail with reference to the accompanying drawings. In addition, in this specification and drawing, about the component which has the substantially same function structure, duplication description is abbreviate | omitted by attaching | subjecting the same code | symbol.

The description will be made in the following order.
1. 1. System overview 2. Functional configuration of system 2-1. Overall configuration 2-2. 2. Details of text data analysis unit 3. Information processing method Display example of visualization screen 4-1. The beginning of the meeting 4-2.1 When presentations are given by participants 4-3. During discussion 4-4. When the meeting is stagnant 4-5. At the time of dispute 4-6. End of the meeting 5. Other display examples 5-1. Visualization screen for positive / negative determination results 5-2. Visualization screen for text data classification results using modalities Other application examples Hardware configuration Supplement

  Here, the present disclosure can be suitably applied to a situation where text data is exchanged by a plurality of users. In the following, as an example of the present disclosure, an embodiment in which the present disclosure is applied to a conference will be mainly described. However, the present disclosure is not limited to such an example, and may be applied to a situation in which messages are exchanged by a plurality of users on social media such as SNS (Social Networking Service).

  The text data exchanged by the participants in the conference includes the contents spoken by the participants, information input by the participants through their own terminals, information handwritten by the participants on the electronic whiteboard, etc. It is. However, in the following explanation, for the convenience of explanation, regardless of the method by which the participant emits text data, the participant issuing text data is also referred to as “speaking” and means the participant who issued the text data. Are also referred to as “speakers”.

(1. System overview)
An overview of a system according to an embodiment of the present disclosure will be described with reference to FIG. FIG. 1 is an explanatory diagram for explaining an overview of a system according to the present embodiment.

  In FIG. 1, the state of the meeting using the electronic white board 20 is schematically illustrated as an example. As shown in FIG. 1, a large number of text data can be issued in a conference. For example, the text data is included in the minutes recorded by the clerk 501 on the electronic whiteboard 20, or various materials used for the conference (presentation materials and URL (Uniform Resource Locator)) linked websites. Etc.), what is included in the speech of the conference participant, what is input via the information processing terminal possessed by the conference participant, and participants other than the secretary 501 use the electronic white board for explanation, etc. 20 may be those described in handwriting.

  In the system 1 according to the present embodiment, various text data issued at the conference are collected. Then, the collected text data is analyzed, and the text data is classified into a predetermined category. The predetermined category may be, for example, a category indicating the intention of a speaker of the text data and / or a category indicating a topic included in the text data. Further, the classification result is visualized and presented to the participant.

  Since the classification result of the text data represents the attribute of the issued text data, the conference participants can see the status of the current conference such as the trend of the conference, for example, by viewing the classification result visualized. You can get various notices. By proceeding according to the notice, the conference can be performed more efficiently.

  The text data analysis method will be described in detail below (2-2. Details of Text Data Analysis Unit). A display example of the visualization screen will be described in detail in the following (4. Display example of the visualization screen).

  Furthermore, in the system 1, a summary of the text data may be generated using the analysis result of the text data. The summary may be, for example, text data collected in association with a classified category. Further, the summary may include a chart that visualizes the classification result of the text data displayed on the visualization screen.

  Since the summary is a summary of various text data issued at the conference, it can be said that the summary can serve as the minutes of the conference. The summary can be said to be useful information when looking back on the contents of the meeting at a later date because it can be said that the summary is compiled from a different viewpoint than the official minutes created by the secretary 501.

  Further, the system 1 may be provided with a function of searching for visualization screens and summaries in past meetings and displaying them again. With this function, for example, when a meeting is started, the contents of the previous meeting can be shared with the participants by referring to the visualization screen and summary of the previous meeting with the participants, so that the meeting can proceed more smoothly. Is possible.

  The overview of the system according to the present embodiment has been described above with reference to FIG.

(2. System functional configuration)
(2-1. Overall configuration)
A detailed functional configuration of the system 1 described with reference to FIG. 1 will be described with reference to FIG. FIG. 2 is a block diagram illustrating an example of a functional configuration of the system 1 according to the present embodiment.

  Referring to FIG. 2, in the system 1 according to the present embodiment, an information processing apparatus 10, an electronic white board 20, and one or a plurality of conference participant terminals 30 are mutually connected via a communication network 40 (network 40). It is configured to be communicably connected.

(Electronic white board 20)
The electronic white board 20 is installed in the conference room and displays various information used for the conference in a display area within the white board surface. For example, the electronic whiteboard 20 can display various materials (for example, presentation materials and websites) presented to the participants at the conference. Moreover, the electronic white board 20 may display various information (for example, comment of a participant etc.) input by the participant of the conference via the conference participant terminal 30 in the display area.

  Moreover, the electronic white board 20 can be comprised so that the participant of a meeting can describe various information by handwriting on the white board surface similarly to a normal white board. For example, during the meeting, the clerk can write the minutes on the electronic white board 20, or other participants can write figures, characters, etc. on the electronic white board 20 for explanation. The electronic white board 20 can acquire various information described on the white board surface as electronic data. Various types of information written on the electronic white board 20 by handwriting (hereinafter referred to as white board input information) are transmitted to the information processing apparatus 10 via the network 40.

  The electronic white board 20 is an example of an input / output device that is installed in a conference room and can be referred to and operated by a plurality of participants. The system 1 may include other display devices or other input devices instead of the electronic white board 20 or together with the electronic white board 20. For example, the system 1 may include a projector, and various types of information used for the conference may be displayed on the projector. Alternatively, depending on the type and scale of the conference, an input / output device such as the electronic white board 20 shared by a plurality of participants may not be provided.

(Meeting participant terminal 30)
The conference participant terminal 30 is an information processing terminal possessed by each participant of the conference. The conference participant terminal 30 may be a notebook PC (Personal Computer), a tablet PC, a smartphone, an electronic paper terminal, a wearable device (for example, a spectacle-type terminal, a head mounted display (HMD), or the like). . In the illustrated example, two conference participant terminals 30 are shown for convenience, but the system 1 may include one or more arbitrary number of conference participant terminals 30.

  In the present embodiment, the conference to which the system 1 is applied may be held by all the participants gathered in the same room, or participation in a remote place using a so-called video conference system or the like. It may be held with the participation of a person. Accordingly, the conference participant terminal 30 may also be used by a participant in the conference room, or may be used by a participant participating in the conference from a remote location.

  For example, a conference participant causes the display area of the electronic whiteboard 20 to be displayed in the display area of the electronic whiteboard 20 by linking the display area of the electronic whiteboard 20 with the display screen of the conference participant terminal 30 of the conference. Can be presented to participants. In addition, a conference participant may input information such as a comment to his / her conference participant terminal 30 and display the information on the electronic white board 20. Various types of information input through the conference participant terminal 30 (hereinafter referred to as terminal input information) are transmitted to the information processing apparatus 10 through the network 40.

(Information processing apparatus 10)
The information processing apparatus 10 analyzes the text data issued in the meeting, classifies the text data into a predetermined category, and visualizes the classification result visualized in association with the speaker of the text data Is generated. The information processing apparatus 10 may be a local apparatus installed in the conference room, or may be installed in a place away from the conference room, such as a server installed on a so-called cloud.

  The information processing apparatus 10 functions as an utterance analysis unit 110, a text data collection unit 120, a text data DB (Data Base) 130, a text data analysis unit 140, an analysis result DB 150, and a situation prediction unit 160. , A visualization screen generation unit 170, a summarization unit 180, and a search unit 190.

  Here, the speech analysis unit 110, the text data collection unit 120, the text data analysis unit 140, the situation prediction unit 160, the visualization screen generation unit 170, the summarization unit 180, and the search unit 190 are, for example, a CPU (Central Processing Unit), a DSP (DSP). The control means is configured by various processors such as a digital signal processor (ASIC) and an application specific integrated circuit (ASIC), and controls the operation of the information processing apparatus 10 by executing predetermined arithmetic processing. These functions are realized by the processors constituting the text data collection unit 120, the text data analysis unit 140, the situation prediction unit 160, the visualization screen generation unit 170, the summarization unit 180, and the search unit 190 operating according to a predetermined program. The

  In addition, the text data DB 130 and the analysis result DB 150 are configured by various storage devices such as a magnetic storage device such as an HDD (Hard Disk Drive), a semiconductor storage device, an optical storage device, or a magneto-optical storage device. Storage means for storing.

  The utterance analysis unit 110 acquires information about the voice uttered by the participant during the conference (hereinafter referred to as voice input information), and analyzes the participant's utterance included in the voice input information. Note that the voice input information can be acquired by a voice input device (not shown) such as a microphone installed in the conference room.

  For example, the utterance analysis unit 110 uses the voice activity analysis (VAD) technology to extract the utterance interval of the participant from the observed audio during the conference. There may be sections in the conference where no participants are speaking and there are only noises such as environmental sounds, but these sections can be spoken by using voice section analysis technology. Can be prevented from being misrecognized as an utterance section of a person. Known techniques for identifying speaker utterances and noise include, for example, the PARADE method and the SKIF method (for example, “Akiko Araki et al.,“ Conference voice speaker identification using voice segment detection and direction information). System and its evaluation ", Acoustical Society of Japan Proceedings, March 2008").

  For example, the utterance analysis unit 110 uses a speaker identification technique to identify whether the person who is speaking is the same person or a different person. Conventionally, various speaker identification techniques using machine learning have been proposed. For example, in the GMM-UBM method, an average utterance model (UBM: Universal Back Ground Model) is introduced, and it is determined whether the voice of a certain speaker X belongs to another known speaker Y. . Specifically, for the voice of the speaker X, the determination is made based on the difference between the similarity regarding the speaker Y and the similarity regarding the UBM. In addition, a method has been proposed in which a plurality of microphones are used to detect the direction of speech and identify a speaker.

  For details of the speaker identification technique, see, for example, “Akiko Araki et al.,“ Conference voice speaker identification system using voice interval detection and direction information and its evaluation ”, Acoustical Society of Japan Proceedings, March 2008. Tsuki ", Takafumi Koshinaka et al.,“ International trends in speaker recognition ”, Journal of the Acoustical Society of Japan, 2013, 69, 7, pp. 342-348 "and the like.

  The utterance analysis unit 110 uses the above-described voice segment detection technology, speaker identification technology, and the like, and a segment in which no participant speaks (hereinafter also referred to as a non-speak segment), or a plurality of participants speak at the same time. And the like (hereinafter also referred to as an utterance collision section) or the like can be extracted.

  Further, for example, the utterance analysis unit 110 identifies a person who is speaking using speaker recognition technology. Thereby, the text data related to the utterance included in the voice input information and the speaker of the text data are associated with each other.

  In addition, as a speech area analysis method, a speaker identification method, and a speaker recognition method, for example, various known methods as described in the above-mentioned documents may be used.

  The utterance analysis unit 110 provides the analysis result of the voice input information to the text data collection unit 120 and the analysis result DB 150.

  The text data collection unit 120 collects text data related to the conference by acquiring all information related to the conference and extracting text data from the information. For example, the text data collection unit 120 acquires terminal input information from the conference participant terminal 30 and extracts text data from the terminal input information. At this time, the text data collection unit 120 can extract the text data in a state associated with the speaker by using an ID or the like assigned to each conference participant terminal 30.

  Further, for example, the text data collection unit 120 acquires information on various materials presented to the participants in the conference (hereinafter referred to as conference material information), and extracts text data from the conference material information. At this time, the text data collection unit 120, for example, based on information such as the creator of the conference material, the text data is associated with the person who describes the characters in the conference material (that is, the speaker). It may be extracted. The conference material information may be manually input to the information processing apparatus 10 by any of the participants in the conference, or the conference material is stored in a predetermined storage area accessible by the information processing apparatus 10. The information processing apparatus 10 may automatically acquire the information.

  In addition, for example, the text data collection unit 120 acquires whiteboard input information from the electronic whiteboard 20, and uses various known methods such as OCR (Optical Character Recognition) and a stroke-based character recognition method (Stroke Based Character Recognition). Then, text data is extracted from the whiteboard input information by analyzing the handwritten characters input to the electronic whiteboard 20. At this time, the text data collection unit 120 may extract the text data in a state associated with the person describing the character (that is, the speaker), for example, by analyzing the handwriting of the written character. . Alternatively, the system 1 may be provided with an imaging device that captures the state of the conference room, and the text data collection unit 120 writes characters on the electronic whiteboard 20 based on the video captured by the imaging device. Participants may be identified and associated with the text data. Various known methods may be used as a handwritten character recognition method, a handwriting analysis method, and a person recognition method based on video.

  For example, the text data collection unit 120 acquires voice input information and extracts text data from the voice input information. At this time, the text data collection unit 120 may extract the text data in a state associated with the speaker based on the result of the speaker recognition process performed by the utterance analysis unit 110. Various known methods may be used as a method for extracting text data from speech.

  The text data collection unit 120 stores the extracted text data in the text data DB 130. At this time, the text data collection unit 120 stores the text data in association with the speaker.

  Here, as an example, the case where the text data collection unit 120 extracts text data from the terminal input information, the conference material information, the white board input information, and the voice input information has been described, but the present embodiment is limited to such an example. Not. The information to be extracted by the text data collection unit 120 may be information related to the conference, and the text data collection unit 120 can extract the text data from any information. For example, if a normal white board is used instead of the electronic white board 20, the text data collection unit 120 acquires image information obtained by photographing the white board, and is written on the white board based on the image information. Text data may be extracted.

  The text data DB 130 stores the text data collected by the text data collection unit 120. At this time, the text data DB 130 can store the text data in association with the name of the meeting in which the text data is spoken, the time when the text data was spoken, the speaker of the text data, and the like. The text data DB 130 may store not only text data extracted during one meeting, but also a history of text data extracted in each of a plurality of meetings.

  The text data analysis unit 140 analyzes text data stored in the text data DB 130. The text data analysis unit 140 includes, as its function, an opinion extraction unit 141 that extracts opinions from text data, and a topic analysis unit 142 that analyzes topics included in the text data.

  In the present embodiment, the text data analysis unit 140 may be configured to execute at least one of the opinion extraction process and the topic analysis process in order to analyze the text data. In FIG. 1, for convenience, the text data analysis unit 140 is illustrated as having both an opinion extraction unit 141 and a topic analysis unit 142, but actually, the text data analysis unit 140 What is necessary is just to be comprised so that it may have at least one of these functions.

  In the opinion extraction process, for example, the intention of the speaker is extracted from the text data in terms of the polarity of the opinion (positive, negative, moderate), modality, etc., and the category corresponding to the intention is extracted. The text data is classified into (opinion type). In the topic analysis process, topics are extracted from words included in the text data by using a method such as LDA (Lent Dirichlet Allocation), for example. Details of the opinion extraction process and the topic analysis process in the text data analysis unit 140 will be described later (2-2. Details of the text data analysis unit).

  The text data analysis unit 140 stores the analysis result of the text data in the analysis result DB 150.

  The analysis result DB 150 stores the analysis result of the text data by the text data analysis unit 140. At this time, the analysis result DB 150 displays the analysis result of the text data, the text data that is the basis of the analysis result, the name of the meeting in which the text data is spoken, the time when the text data is spoken, and the statement of the text data. Can be stored in association with a person or the like. In addition, the analysis result DB 150 may store not only analysis results for text data extracted during one meeting, but also a history of analysis results for text data extracted in each of a plurality of meetings.

  The analysis result DB 150 stores the utterance analysis result by the utterance analysis unit 110. At this time, similarly to the analysis result of the text data, the analysis result DB 150 displays the analysis result of the utterance, the utterance content that is the basis of the analysis result, the name of the meeting where the utterance was made, the time when the utterance was made, It can be stored in association with the utterer of the utterance. Further, the analysis result DB 150 may store not only the analysis result for the utterances uttered during one meeting, but also the history of the analysis results for the utterances uttered in each of a plurality of meetings.

  The situation prediction unit 160 predicts the meeting situation based on the text data stored in the analysis result DB 150 and / or the analysis result of the utterance. The status of the conference includes, for example, that it is immediately after the conference is started, that the conference is stagnant, that the discussion is inconsistent, and that the conference is approaching the end. The situation prediction unit 160 provides the predicted result to the visualization screen generation unit 170.

  Based on the text data stored in the analysis result DB 150 and / or the analysis result of the utterance, the visualization screen generation unit 170 generates a visualization screen that is a screen on which the analysis result is visualized. For example, the visualization screen generation unit 170 determines a predetermined category (for example, a category indicating the intention of the speaker of the text data (ie, opinion type) and / or a topic included in the text data based on the analysis result by the text data analysis unit 140. Text data can be classified into categories, and a screen in which the classification result is visualized in association with the speaker of the text data can be generated. In addition, the visualization screen generation unit 170 visualizes the utterance situation (for example, the utterance interval, non-utterance interval, and / or utterance collision interval of each participant) during the conference based on the analysis result by the utterance analysis unit 110. A screen can be generated.

  Information about the visualization screen generated by the visualization screen generation unit 170 is transmitted to the electronic white board 20 and / or the conference participant terminal 30 via the network 40, and the electronic white board 20 and / or the conference participant terminal 30. Displayed on the display screen. Since the visualization screen is a screen in which the intention of the speaker, the topic included in the text data, and / or the utterance status during the conference are visualized in association with the speaker, these visualization screens are displayed during the conference. It is possible to give the participant various awareness regarding the progress of the conference.

  For example, by referring to the visualization screen, each participant objectively reads information such as the status of active discussion, whether there are many positive opinions, and the position of each participant. be able to. For this reason, it becomes possible to obtain awareness such as confirmation that the content currently being discussed does not deviate from the agenda of the original meeting, and that the discussion is converging. Also, if there are a lot of text data that ask questions or questions, the contents of the explanatory materials and the preliminary preparation of the meeting may be insufficient, so participants will improve these points for the next meeting It can be a trigger. Or, there may be a lot of text data that asks questions or questions due to differences in comprehension depending on the field or position of the participants, so it should be used as a reference when creating materials at the next meeting. Can do.

  Note that the visualization screen may be displayed on the electronic whiteboard 20 and / or all conference participant terminals 30 so that all participants can refer to it. For example, only a specific participant can refer to the visualization screen. It may be displayed only on the conference participant terminals 30 of some participants such as facilitators.

  Further, at this time, the visualization screen generation unit 170 generates a visualization screen having contents according to the conference status based on the conference status predicted by the status prediction unit 160 at a timing according to the conference status. Also good. Based on the situation of the meeting, a more appropriate visualization screen is generated, and the visualization screen is presented to the meeting participants at a more appropriate timing, so that the participants can be more useful about the progress of the meeting. It becomes possible to provide information.

  The details of the process of predicting the meeting status by the status prediction unit 160 and the example of the visualization screen generated by the visualization screen generation unit 170 are described below (4. Display example of visualization screen) and (5. Other tables). This will be described in detail in the example).

  The summarization unit 180 creates a summary of the meeting based on the text data stored in the analysis result DB 150 and the analysis result of the utterance and the visualization screen generated by the visualization screen generation unit 170. For example, the summary unit 180 creates a summary by associating text data with a classified category. The summarization unit 180 may create a summary including a chart that visualizes the classification result of the text data displayed on the visualization screen. Further, the summary unit may create a summary including minutes recorded on the electronic white board 20 by a secretary, materials displayed during the meeting, and the like.

  The conference summary created by the summary unit 180 is distributed to the electronic whiteboard 20 and / or the conference participant terminal 30 via the network 40. The participant can refer to the summary of the conference via the electronic white board 20 and / or the conference participant terminal 30. By referring to the summary, the participant can easily check the transition of opinions in the conference, the standing position of each participant on the agenda, and the like afterwards.

  Here, the minutes recorded by the secretary on the electronic white board 20 and the official minutes created by the secretary after the meeting may have a mixed subjectivity of the secretary. On the other hand, since the summary created by the summarization unit 180 is automatically generated based on the analysis result of the text data, it can be said that the contents of the meeting are summarized from a more objective viewpoint. . Therefore, the participants of the conference can grasp the content of the conference from various perspectives by referring to both the official minutes and the summary created by the summary unit 180.

  As a method for creating a conference summary, for example, various methods generally known as a method for automatically generating a conference summary may be used.

  The search unit 190 searches the text data stored in the analysis result DB 150 and the analysis result of the utterance, and the visualization screen generated by the visualization screen generation unit 170, and displays the analysis result and / or the visualization screen in the past meeting. This is distributed to the electronic whiteboard 20 and / or the conference participant terminal 30 via the network 40. For example, the participants of the conference can easily grasp the content of the past conference by checking the analysis result in the past conference and the content of the visualization screen as necessary. Therefore, it is possible to more smoothly carry out discussions taking over the contents of past meetings. In addition, the search unit 190 searches the minutes recorded in the electronic whiteboard 20 in the past conference, the materials displayed during the conference, etc., and distributes them to the electronic whiteboard 20 and / or the conference participant terminal 30. May be.

  The search instruction may be input by the participant via the electronic whiteboard 20 and / or the conference participant terminal 30. For example, in response to an instruction from the participant, the search unit 190 can perform a search from the viewpoint of the meeting name, the date and time of the meeting, the participants of the meeting, the contents of text data spoken during the meeting, and the like. . For example, by searching for analysis results and contents of visualization screens at meetings of similar themes held in the past, knowledge about the contents of the past meetings can be shared among all participants. Can be avoided.

  Heretofore, the functional configuration of the system 1 according to the present embodiment has been described. As described above, according to the present embodiment, a speaker's intention (affirmation, denial, suggestion, question, command, etc.) is extracted from the speech in the conference and visualized. Therefore, a lot of awareness can be given to the participants, the conference can proceed efficiently, and the convenience of the participants can be improved. For example, if negative remarks occur frequently, it will trigger confirmation of whether the discussion has deviated from the main topic. In addition, for example, when positive opinions increase, it becomes a guide to proceed to the step of obtaining the participant's agreement on the agenda. In addition, for example, when a question erupts, it is imagined that there is some deficiency in explanation or material. In addition, it can be seen that there is a difference in understanding depending on the position and skills of the participants, which can be used as a reference for the next document preparation.

  The configuration of the system 1 is not limited to the illustrated example. The system 1 should just be comprised so that each function demonstrated above as a whole is realizable, and the specific structure may be arbitrary. For example, each function illustrated as a function of the information processing apparatus 10 in FIG. 2 may not necessarily be executed by one apparatus, and may be executed by cooperation of a plurality of apparatuses.

  Note that a computer program for realizing each function of the system 1 according to the present embodiment as described above can be produced and mounted on a PC or the like. In addition, a computer-readable recording medium storing such a computer program can be provided. The recording medium is, for example, a magnetic disk, an optical disk, a magneto-optical disk, a flash memory, or the like. Further, the above computer program may be distributed via a network, for example, without using a recording medium.

(2-2. Details of text data analysis unit)
A more detailed function of the text data analysis unit 140 shown in FIG. 2 will be described.

(Opinion extraction process)
A more detailed function of the opinion extraction unit 141 of the text data analysis unit 140 shown in FIG. 2 will be described with reference to FIGS. FIG. 3 is a functional block diagram illustrating an example of a functional configuration of the opinion extraction unit 141 illustrated in FIG. FIG. 4 is an explanatory diagram for explaining opinion extraction processing by the opinion extraction unit 141.

  The opinion extraction unit 141 includes, as its functions, a sentence division unit 201, a syntax / semantic analysis unit 203, a business term collection 205, a dictionary generation unit 207, a business term dictionary 209, an opinion / evaluation expression extraction unit 211, Have. The sentence division unit 201, the syntax / semantic analysis unit 203, the dictionary generation unit 207, and the opinion / evaluation expression extraction unit 211 are configured by a processor such as a CPU, and the program executes predetermined arithmetic processing according to a predetermined program. These functions can be realized. Further, the business term collection 205 and the business term dictionary 209 can be configured by a storage device such as an HDD.

  The sentence division unit 201 decomposes the text data into words by, for example, morphological analysis processing. The syntax / semantic analysis unit 203 performs syntax analysis processing and semantic analysis processing on the text data divided by the sentence division unit 201 while referring to the business term dictionary 209. By performing the morphological analysis process, the syntax analysis process, and the semantic analysis process, the relationship structure between the clauses in the text data is analyzed. Note that various known methods used in the field of so-called natural language processing may be used as specific methods of morphological analysis processing, syntax analysis processing, and semantic analysis processing.

  For example, let us consider a case where the sentence division unit 201 and the syntax / semantic analysis unit 203 analyze the text data “I want more Web production personnel”. In this case, as shown in FIG. 4, the text data is divided into phrases “Web production”, “Human resources”, “More” and “I want more”. "Phrase" and "case" are analyzed. In addition, regarding the dependency structure, the phrase “Web production” is related to the phrase “personnel”, and the phrases “personnel” and “more” are both phrased “I want more.” It is analyzed that it is involved.

  The business terminology 205 is a database in which terms that are frequently used in business are collected. A plurality of business glossaries 205 may be prepared for each business field. The business field may be, for example, the business field of the company where the conference is held, such as “electrical machinery”, “IT”, “construction”, or more specifically “design development”, “production”, “accounting”. ", Etc., may be a field related to occupations within a single company.

  The dictionary generation unit 207 extracts a term corresponding to the business field to which the conference currently being analyzed belongs from the business term collection 205 and generates a business term dictionary 209. As described above, the dictionary generation unit 207 can generate the business term dictionary 209 suitable for the conference for each conference.

  The business term dictionary 209 is a database in which words and parts of speech are stored. The business term dictionary 209 only needs to be configured to include specific expression information and evaluation expression words that are generally required to perform syntax analysis processing and semantic analysis processing, and the configuration may be arbitrary.

  The opinion / evaluation expression extraction unit 211 is based on the words included in the text data analyzed by the sentence division unit 201 and the syntax / semantic analysis unit 203, the dependency structure of the text data, and the predicate term structure of the text data. The evaluation expression and opinion are extracted from the text data. Further, the opinion / evaluation expression extraction unit 211 classifies the opinion into a category (opinion type) according to the intention of the speaker based on the extracted evaluation expression.

  Here, in this embodiment, “opinion” means a combination of “evaluation expression” of text data to be analyzed and “evaluation object (object)” of the evaluation expression. The evaluation expression is an expression representing the intention of the speaker included in the text data. For example, in the above example, “I want you to increase” corresponds to the evaluation expression. Further, “personnel” corresponds to the object of the evaluation expression. That is, the opinion extracted from the text data “I want you to increase the number of Web production personnel” can be ““ I want to increase the number of personnel ””.

  Specifically, the opinion / evaluation expression extraction unit 211 identifies a phrase corresponding to the predicate and a phrase related to the phrase corresponding to the predicate by performing predicate term structure analysis processing. Then, an evaluation expression of the text data is extracted from the phrase corresponding to the predicate. In addition, a phrase related to the phrase corresponding to the predicate is extracted as an object of the evaluation expression.

  Further, the opinion type can be, for example, “affirmation”, “denial”, “necessary”, “request”, “suggestion”, and the like. The opinion type can be determined based on the evaluation expression. For example, in the case where the analysis is performed on the text data “I want to increase the number of Web production personnel” above, the opinion / evaluation expression extraction unit 211 is the evaluation expression “I want more. . ”Is a phrase representing the opinion types“ necessary ”and“ aspiration ”. Then, the opinion / evaluation expression extraction unit 211 classifies opinions including the evaluation expression (that is, “I want more people” and “I want more people”) into opinion types “necessary” and “wish”.

  In this way, the opinion / evaluation expression extraction unit 211 finally extracts an opinion and an opinion type representing the category to which the opinion belongs from the text data. Since it can be said that the opinion is extracted from the main part of the text data, it can be said that the extraction processing of the opinion and opinion type by the opinion / evaluation expression extraction unit 211 is processing for classifying the text data into opinion types.

  The opinion / evaluation expression extraction unit 211 extracts opinions from each text data issued during the meeting collected by the text data collection unit 120, and classifies the extracted opinions into opinion types. Then, the extracted opinion and the opinion type are linked and stored in the analysis result DB 150. The visualization screen generation unit 170 refers to the analysis result DB 150 and generates a visualization screen in which the classification result, that is, the opinion type for each text data (or each opinion) is visualized.

  Here, as a method for judging the opinion type from the evaluation expression, various known methods can be used. Here, as an example, a method using positive / negative determination and a method using modality will be described.

(Method using positive / negative judgment)
In general, a technique called positive / negative determination is known in which an evaluation expression word representing affirmative (positive) and negative (negative) is used to determine the affirmation degree and the negation degree of text data. The opinion / evaluation expression extraction unit 211 uses the positive / negative determination to determine whether the text data or opinion is affirmative or negative, and the opinion type to which the text data or opinion belongs is “affirmation”, “denial”, or “ Can be determined to be neutral.

  In the positive / negative determination, it is necessary to create an evaluation expression word dictionary in advance. The evaluation expression word dictionary may be created manually, but since manual evaluation creates differences in the evaluation expression by domain and takes time to create, there is also a method of automatically creating the evaluation expression word dictionary in advance. Proposed (for example, “Tetsuya Nasukawa, Hiroshi Kanayama, Yuta Tsuboi, Hideo Watanabe,“ Natural Language Processing Applying Favored Context ”, The 11th Annual Conference of the Association for Language Processing, March 2005, pp. 153-156 ”,“ Choichi Sugano, “Method for extracting text sensibility expression contained in“ voice of customer ””, IPSJ Research Report Natural Language Processing (NL153-14), 2002, pp. 105-112 ” Etc.).

  In addition, there are evaluation evaluation words whose evaluation (that is, opinion type) varies depending on phrases and words around the evaluation expression word and context. For example, an evaluation expression word such as “good” or “bad” may be a word directly representing “affirmation” or “denial”, but in the case of an evaluation expression word “long”, the “long” The evaluation may vary depending on the context in which the word is used. For example, the evaluation of text data “long charging time” may be “negative”, but the evaluation of text data “long available time” may be “positive”. Therefore, a positive / negative determination method that analyzes the structure of a sentence and the co-occurrence of words before and after and considers the object of positive / negative determination has also been proposed (for example, “Shigeru Fujimura et al.” "Extraction method", IEICE 16th Data Engineering Workshop (DEWS 2005), 2005 ", etc.). The opinion / evaluation expression extraction unit 211 may perform positive / negative determination in consideration of the above-described “object” using the method.

(Method using modality)
Modality is the attitude or mood of a speaker in a sentence or utterance. For example, for the verb “go”, by changing the usage or attaching an attached word, “will go” (inference), “go” (command), “should” (duty, assert), “do not go” "(Negation)" is given various meanings representing the attitude of the speaker. In general, a technique for extracting such a speaker's attitude and mood (ie, modality) from an evaluation expression included in text data has been developed.

  The opinion / evaluation expression extraction unit 211 can extract a modality from text data or an evaluation expression included in an opinion using the technology, and can employ the modality as the text data or the opinion type of the opinion. The modalities are, for example, “assumed”, “difficult”, “easy”, “prohibited”, “question”, “aspiration”, “claim”, “solicitation”, “possible”, “unexpected”, “degree”. , “Easy”, “Impossible” and so on.

  Here, as described above, the opinion / evaluation expression extraction unit 211 extracts “object” as a part of the opinion. It can be said that the “object” represents a target of the modality. Thus, in the present embodiment, modalities can be extracted including the target ones.

  For example, if the text data is “there is no schedule,” the target is “schedule” and the modality is “deny”.

  For example, in the case of text data “Let's postpone the sale”, the object is “release”, and the modalities are “aspiration” and “claim”.

  For example, if the text data is “Accuracy is not achieved with the current method”, the target is “Accuracy” and the modality is “Negative”.

  For example, in the case of text data “Would you like to be involved in the project?”, The target is “Brest”, and the modalities are “Assumption” and “Proposal”.

  The detailed function of the opinion / evaluation expression extraction unit 211 has been described above.

(Topic analysis process)
A more detailed function of the topic analysis unit 142 of the text data analysis unit 140 shown in FIG. 2 will be described. The topic analysis unit 142 analyzes the topic of the text data.

  As a method for analyzing the topic of text data, various known methods can be used. For example, LDA is known as a method for estimating a latent topic of a document from words appearing in the document. In addition, a method for determining a topic change point from a series of sentence flows using LDA has been proposed (for example, “David M. Blei et al.,“ Lent Dirichlet Allocation ”, Journal of Machine Learning Research 3, 2003, pp. 993-1022 "," Akira Nakamura et al., "Effects of multiple model integration in LDA online adaptation based on topic change point detection", Proc. Of the 15th Annual Conference of the Language Processing Society, 3, 4, 2009 ").

  LDA is a document classification model based on the premise that a document has a plurality of topics. In LDA, a latent topic of a document is defined by a hidden variable Z (multiple dimensions), and a topic probability P (Z | D) of each document D and a word appearance probability P (W | Z) is learned. Then, P (Z | D) can be estimated for an unknown document.

  In LDA, it is necessary to learn a word co-occurrence relationship in advance using a large number of documents in a corresponding field. For example, the business documents related to the target industry are used in advance to learn about the field to which the industry belongs, and at the same time, additional learning is performed using text data collected during the meeting. It is also possible to update the learning model.

  In the present embodiment, text data can be classified according to potential topics estimated using LDA. For example, if words such as “hard disk”, “partition”, and “C drive” are included in the text data, the dimension score related to these words is high for the text data. Also, for example, if words such as “Internet”, “Wi-Fi (registered trademark)”, “router” are included in the text data, the dimension score related to these words is high for the text data. Become. Therefore, by classifying the text data according to the result of the dimension score, the text data having a similar topic is classified into the same category. Thus, according to LDA, text data can be classified using multi-topic attributes. It is also possible to detect a topic change point. The visualization screen generation unit 170 can generate a visualization screen in which the result of classifying text data by latent topics in this way is visualized.

  The detailed function of the topic analysis unit 142 has been described above.

(3. Information processing method)
With reference to FIG. 5, a processing procedure of the information processing method according to the present embodiment will be described. FIG. 5 is a flowchart showing an example of a processing procedure of the information processing method according to the present embodiment. Note that the processing in each step shown in FIG. 5 corresponds to the processing executed in the system 1 shown in FIG. Since the details of the processing executed in the system 1 have already been described with reference to FIG. 2, the following description of the information processing method according to the present embodiment describes only an overview of the processing in each step. The description of the details of each process is omitted.

  Referring to FIG. 5, in the information processing method according to the present embodiment, first, the participant's utterance is analyzed based on the voice input information (step S101). The processing in step S101 corresponds to the processing executed by the utterance analysis unit 110 shown in FIG.

  Next, text data related to the meeting is collected (step S103). The processing in step S103 corresponds to the processing executed by the text data collection unit 120 shown in FIG.

  Next, the collected text data is analyzed (step S105). The processing in step S105 corresponds to the processing executed by the text data analysis unit 140 shown in FIG.

  Next, based on the utterance analysis result in step S101 and / or the text data analysis result in step S103, the status of the conference is predicted (step S107). The process in step S107 corresponds to the process executed by the situation prediction unit 160 shown in FIG. Details of the processing in step S107 will be described later in (4. Display example of visualization screen).

  Next, based on the utterance analysis result in step S101 and / or the analysis result of the text data in step S103, a visualization screen that is a screen on which the analysis result is visualized is generated (step S109). The processing in step S109 corresponds to the processing executed by the visualization screen generation unit 170 shown in FIG. In the process in step S109, a visualization screen may be generated based further on the prediction result of the meeting status in step S107.

  Next, the generated visualization screen is displayed (step S111). In the process in step S111, a visualization screen is displayed on the electronic white board 20 and / or the conference participant terminal 30 shown in FIG.

  The processing procedure of the information processing method according to the present embodiment has been described above with reference to FIG.

(4. Visualization screen display example)
A display example of the visualization screen according to the present embodiment will be described. In the present embodiment, the status of the conference is predicted, and the display of the visualization screen presented to the participant can be sequentially switched according to the status. Accordingly, here, assuming a typical meeting progress, a display example of the visualization screen will be described while explaining how the display of the visualization screen changes sequentially according to the progress of the meeting. . In addition, a specific example of the process of predicting the meeting status will also be described.

(4-1. The beginning of the meeting)
For example, the situation prediction unit 160 illustrated in FIG. 2 may be input with information about the time since the start of the conference, and the situation prediction unit 160 determines the current situation based on the information about the time. , Immediately after the meeting is started, that is, the beginning of the meeting can be predicted.

  In the case of the beginning of the meeting, the visualization screen generation unit 170 generates, for example, the visualization screen displayed at the time of the previous meeting again. The visualization screen displayed during the previous meeting is presented to the participant so that the participant can intuitively grasp the contents of the previous meeting in a short time. It will be possible to participate smoothly in the proceedings. The visualization screen is preferably a screen on which the contents of the previous meeting can be easily grasped, such as a screen on which changes in the time series of topics and keywords shown in FIG.

(When a presentation is given by 4-2.1 participants)
At the beginning of a meeting, a single participant often makes a presentation to raise an agenda. During the presentation, it is likely that only one participant is speaking.

  Accordingly, the situation prediction unit 160 determines whether one speaker has continued speaking for a predetermined time or more based on the result of speaker identification processing by the speech analysis unit 110. Based on the determination result, the situation prediction unit 160 can predict that the current situation is a situation where a presentation is being performed by one speaker. When such a prediction is made by the situation prediction unit 160, the presentation material should be presented to the participant. For example, the visualization screen generation unit 170 does not generate a visualization screen. However, even if the visualization screen is not generated, in the system 1, other processes such as speech analysis and text data collection are executed as needed.

  FIG. 6 shows a processing procedure of a prediction process of whether a presentation is being performed by one participant by the situation prediction unit 160 and a process by the visualization screen generation unit 170 according to the result of the prediction process. FIG. 6 is a flowchart showing an example of a processing procedure of a prediction process of whether a presentation is being performed by one participant by the situation prediction unit 160 and a process by the visualization screen generation unit 170 according to the result of the prediction process. It is. In each process shown in FIG. 6, the processes in steps S201 to S205 and step S211 correspond to the processes performed by the situation prediction unit 160, and the processes in steps S207 and S209 are performed by the visualization screen generation unit 170. It corresponds to the processing.

  Referring to FIG. 6, first, each speaker's utterance section is detected for each unit time frame, and these are integrated (step S201). Next, it is determined whether or not the total of all the speaker utterance sections (all utterance sections) has reached a specified time or more (step S203).

  Here, in the processing in step S205, which will be described later, it is determined whether or not one speaker is speaking continuously for a long time, based on the ratio of each speaker's utterance interval to the total utterance interval. Therefore, when the entire utterance section is short, even if a certain speaker utters only for a very short time, the proportion of the utterance section of the speaker in the entire utterance section becomes large. There is a risk that an erroneous determination is made in S205. Therefore, in this embodiment, when it is determined in step S203 that the total utterance time is shorter than the specified time, the process does not proceed to step S205, but returns to step S201 to include the next unit time frame. Integration processing is further performed. Only when it is determined in step S203 that the total utterance time is equal to or longer than the specified time, the process proceeds to step S205.

  In step S205, whether the utterance rate S (X) of the speaker X (X is a number indicating the speaker: 1,..., N (N is the total number of participants in the conference)) exceeds the threshold Ts. Whether it is determined. Here, the utterance rate S (X) is S (X) = (the utterance time of the speaker X) / (all utterance sections). The threshold value Ts is a threshold value for determining whether or not the speaker X is giving a presentation. The threshold value Ts can be appropriately set in consideration of the past actual value of the utterance rate of a speaker who performs the presentation in a general presentation.

  When it is determined in step S205 that S (X) exceeds the threshold value Ts, in the section to be determined (that is, the section in which all utterance sections are calculated), the speaker X is almost the same. Therefore, the current situation is predicted to be a situation where the speaker X is giving a presentation. Accordingly, in this case, the process proceeds to step S207, and the visualization screen is not generated, and the presentation material is displayed to the conference participants (step S207).

  On the other hand, when it is determined in step S205 that S (X) is equal to or less than the threshold value Ts, a plurality of speakers are used in the determination target section (that is, a section in which all utterance sections are calculated). Therefore, the current situation is unlikely to be the situation where any one speaker is giving a presentation. Accordingly, in this case, the process proceeds to step S209, where a visualization screen corresponding to a normal discussion period (for example, a visualization screen corresponding to another prediction result by the situation prediction unit 160) is generated, and is sent to the conference participants. Is displayed.

  When the process in step S207 or step S209 is completed, the integration period for calculating all utterance periods is updated (step S211), the process returns to step S201, and the series of processes described above is repeatedly executed.

(4-3. During discussion)
When the presentation ends, a discussion on the agenda can be held by a plurality of participants. For example, the situation prediction unit 160 can predict the end of the presentation and the start of the discussion by calculating the utterance rate of each speaker based on the result of the speaker identification process by the utterance analysis unit 110.

  In the present embodiment, at the time of discussion, the visualization screen generation unit 170 visualizes the transition of opinion types (for example, results of positive / negative determination and modality), the transition of latent topics, and the transition of keywords as shown in FIG. A screen is suitably generated. FIG. 7 is a diagram illustrating an example of a visualization screen, which is a visualization screen in which a transition of opinion types, a transition of latent topics, and a transition of keywords are visualized.

  Referring to FIG. 7, in the visualization screen, the flow of time is plotted on the vertical axis, and for the text data spoken during the meeting, the speaker who spoke the text data, the opinion type of the text data (positive / negative determination) Result, modality), latent topics of the text data, and keywords included in the text data can be arranged and displayed in the order in which the text data was spoken. The visualization screen generation unit 170 can include an opinion type on the visualization screen based on the analysis result by the opinion extraction unit 141 illustrated in FIG. Further, the visualization screen generation unit 170 can include the latent topics and keywords on the visualization screen based on the analysis result by the topic analysis unit 142 illustrated in FIG.

  In the illustrated example, regarding the opinion type, a column is provided for each opinion type, and the corresponding column is filled in, so that the opinion type of the text data is visually expressed. Further, as shown in the drawing, the fill may be shaded depending on the degree of intention of the speaker represented by the opinion type corresponding to the text data.

  Participants in the conference can intuitively grasp the transition of the opinion type during the discussion by referring to the filled state in the opinion type column. For example, if the frequency of opinion types such as “Negative” and “Question” decreases, it is considered that opinions about the agenda of the meeting and comments about the question have decreased. It is an indicator that participants' agreement is being formed on the agenda.

  In the illustrated example, “affirmation”, “denial”, “aspiration”, “question”, and “suggestion” are displayed as opinion types, but the types of opinion types displayed on the visualization screen are such examples. Without limitation, various opinion types that are considered beneficial for the progress of the meeting may be displayed on the visualization screen. Which opinion type is included in the visualization screen may be appropriately set by the designer of the system 1 or the participant of the conference.

  In addition, in the visualization screen, as with the modality, a column is provided for each type of the latent topic, and the corresponding topic is filled in so that the latent topic indicated by the text data is visually expressed. Yes. Further, the filling may be shaded depending on the level of the level of the latent topic corresponding to the text data.

  Note that the type of latent topic may correspond to the hidden variable Z in the LDA described above. In the illustrated example, probability values corresponding to the hidden variables Z (in this example, five dimensions, Z1 to Z5) obtained as a result of topic analysis are displayed in each column. The probability value represents the probability that the text data belongs to the latent topic corresponding to each hidden variable Z. The shade of each column corresponds to the magnitude of the probability value.

  Participants in the conference can intuitively grasp the topic transition during the discussion by referring to the state of filling in the latent topic column. For example, if the tendency of the latent topic has changed significantly at a certain time, the participant can know that the direction of the meeting may have changed at that time. In addition, the participant can know a specific change in the topic by referring to the keyword column.

  In the illustrated example, the transition of opinion types, the transition of latent topics, and the transition of keywords are all displayed, but the visualization screen displayed at the time of discussion is not limited to such an example. At the time of discussion, it is preferable to generate a visualization screen that can visually grasp the flow of discussion. For example, the visualization screen displayed at the time of discussion includes the transition of opinion types, the transition of potential topics, and the It is sufficient that at least one of the transitions is included. In addition, if there is room in the screen space, a specific opinion may be visualized.

  Here, regarding the change in the tendency of the latent topic, in the present embodiment, the point in time when the tendency of the latent topic has changed significantly (potential topic change point) may be explicitly displayed on the visualization screen. In the example shown in the figure, the change point of the latent topic is indicated by a thick broken line on the visualization screen.

  The change point of the latent topic can be detected by the processing procedure shown in FIG. 8, for example. FIG. 8 is a flowchart showing an example of a processing procedure for detecting a change point of a latent topic. In each process shown in FIG. 8, the processes in steps S301 to S307 and S313 correspond to the processes performed by the situation prediction unit 160, and the processes in steps S309 and S311 are performed by the visualization screen generation unit 170. It corresponds to the processing. That is, the change point of the latent topic is predicted by the situation prediction unit 160, and the visualization screen generation unit 170 may generate a visualization screen including the change point based on the prediction result.

  Referring to FIG. 8, first, words included in text data within a unit time frame are acquired (step S301). As the word, for example, a text data division result by morphological analysis performed by the sentence division unit 201 illustrated in FIG. 3 may be used.

  Next, it is determined whether a predetermined time has elapsed (step S303). This is because if the period of time during which text data is collected is excessively short, the number of collected text data is small, and it is difficult to detect changes in potential topics with high accuracy.

  If it is determined in step S303 that the predetermined time has not elapsed, the process returns to step S301 to further acquire words including the text data collected in the next unit time frame. On the other hand, if it is determined in step S303 that the predetermined time has elapsed, the process proceeds to step S305.

In step S305, the topic feature vector Z is extracted based on the words in the text data using LDA. The topic feature vector Z is a vector representation of the hidden variable Z in the LDA described above. The topic feature vector Z is extracted as needed for each collected text data within a period in which words are acquired. Hereinafter, the period during which one topic feature vector Z is extracted is also referred to as a topic frame for convenience, and the topic feature vector Z extracted in the topic frame n (n = 1, 2,...) and it is described as Z _n.

Then, based on the extracted topic feature vector Z _n, whether topics change point is detected is judged (step S307). Various known methods may be used as the topic change point detection process. For example, linear distance and topic feature vector _{Z n} and topic feature vector _{Z n-1, Kallback-Leibler} using information amount, etc., topics extracted in the previous topic frame feature vector _{Z n-1} And the similarity Sim (Z _n , Z _n-1 ) with the pick feature vector Z _n is calculated, and is the similarity Sim (Z _n , Z _n-1 ) smaller than a predetermined threshold value Tt? By determining whether or not, a topic change point can be detected.

For example, in the present embodiment, the similarity Sim (Z _n , Z _n-1 ) may be defined by a simple format such as the following formula (1). In addition, as the threshold value Tt, an arbitrary value generally used for determining the similarity of latent topics may be set.

If it is determined that the similarity Sim (Z _n , Z _n-1 ) is smaller than the threshold value Tt, it is considered that the topic has changed relatively greatly at that time, so that the time changes the topic. It can be regarded as a point. In this case, it is considered that a topic change point has been detected, and the process proceeds to step S309. In step S309, for example, as shown in FIG. 7, a visualization screen including the detected topic change point is generated along with the change of the modality, the latent topic, and the keyword.

On the other hand, when it is determined that the similarity Sim (Z _n , Z _n-1 ) is _{equal to} or greater than the threshold value Tt, it is considered that no major topic change has occurred. In this case, it is considered that a topic change point has not been detected, and the process proceeds to step S311. In step S311, for example, the visualization screen shown in FIG. 7 has no display related to the topic change point, that is, the visualization screen showing the normal opinion type, latent topic, and keyword transition in which the topic change point is not displayed.

  When the processing in step S309 or step S311 is completed, the integration interval for acquiring a word for detecting a topic change point is updated (step S313), the process returns to step S301, and the series of processing described above is repeatedly executed.

(4-4. When the meeting is stagnant)
As the conference progresses, it is difficult for the participants to give opinions, and the conference may stagnate. For example, the situation prediction unit 160 can predict that the conference is stagnant by detecting a non-speech segment based on the result of the speech segment analysis process performed by the utterance analysis unit 110. For example, the situation prediction unit 160 predicts that the current situation is a situation where the conference is stagnant when the non-speech section continues for a predetermined time or more.

  When the situation prediction unit 160 predicts that the conference is stagnant, the visualization screen generation unit 170 preferably uses a visualization screen that visualizes the utterance status of participants during the conference, as shown in FIG. 9, for example. To generate. FIG. 9 is a diagram illustrating a visualization screen that is another example of the visualization screen, in which the utterance status of the participant during the conference is visualized.

  Referring to FIG. 9, on the visualization screen, the horizontal axis indicates the time from the conference start time, and the section where the utterance is spoken for each participant is filled and displayed. In addition, a display showing the current time is also made. The visualization screen generation unit 170 can generate the visualization screen based on the results of the speech segment analysis processing and speaker recognition processing by the utterance analysis unit 110.

  By referring to the visualization screen, the conference participants can intuitively know the participants who have not spoken so far. For example, other participants, such as facilitator participants, can encourage such participants to speak up so far. In this way, according to the visualization screen, since the frequency of the participant's previous speech is visualized and presented to the participant, an opportunity to speak can be given to the participant who has little speech. It can be expected that the discussion will become more active.

  When the meeting is stagnant, the visualization screen generation unit 170 may generate the visualization screen shown in FIG. 10 instead of the visualization screen shown in FIG. FIG. 10 is a diagram illustrating a visualization screen that is another example of the visualization screen, in which the utterance status of participants during the conference and the transition of the opinion type of the utterance are visualized.

  Referring to FIG. 10, in the visualization screen, similarly to the visualization screen shown in FIG. 9, the horizontal axis takes time from the conference start time, and the section where the utterance is spoken is displayed for each participant. . However, in the visualization screen shown in FIG. 10, the color of the fill is changed according to the type of opinion type of the utterance. In the example shown in the figure, different colors are assigned to the statements belonging to “affirmation”, “denial”, and “neutral”. In addition, in the visualization screen shown in FIG. 10, an utterance collision section and a non-utterance section are also displayed. The visualization screen generation unit 170 can generate the visualization screen based on the results of the opinion extraction processing by the opinion extraction unit 141 and the speech segment analysis processing and speaker recognition processing by the utterance analysis unit 110.

  By referring to the visualization screen shown in FIG. 10, the conference participants can intuitively grasp the participants who have not spoken so far as in the case of referring to the visualization screen shown in FIG. And ask the participants for their opinions. At this time, in the visualization screen shown in FIG. 10, since each participant's opinion type with respect to the utterance, that is, the intention, is also visualized, the facilitating participant or the like stands on the agenda of the participant who does not speak much. Based on the position, the participant can be asked to speak. Therefore, since it is possible to ask for a more appropriate participant at a more appropriate timing, further activation of the discussion can be expected.

(4-5. Discussion disputes)
Contrary to the above (4-4. At the time of stagnation of the conference), as the conference progresses, opinions frequently come out from the participants, and the discussion may be confused. For example, the situation prediction unit 160 can predict that the conference is stagnant by detecting an utterance collision section based on the results of the speech section analysis process and the speaker identification process performed by the utterance analysis section 110. For example, the situation prediction unit 160 predicts that the current situation is a situation in which the discussion is inconsistent when the utterance collision section has occurred a predetermined number of times or more. Note that the utterance collision section can be detected by analyzing the sound collected by an on-microphone with a microphone attached to each speaker or estimating the direction of the speaker using a microphone array composed of a plurality of microphones. (For example, “Yuo et al.,“ Evaluation method of speech section detection and speaker direction estimation from multi-person dialogue using microphone array ”, 30th Anniversary Conference of the Robotics Society of Japan, RSJ2012AC3D1- 4, September 2012 ”,“ Kofi Bokeye et al., “OVERLAPPED SPEECH DETECTION FOR IMPROVED SPEAKER DIARIZATION IN MULTIPARTY MEETINGS”, Acoustics and Spiges al Processing, 2008. ICASSP, 2008. IEEE International Conference on, March 31-April 4, 2008, p.

  When the situation prediction unit 160 predicts that the discussion is inconsistent, the visualization screen generation unit 170 visualizes the transition of the content of the speech and the transition of the opinion type of the speech as shown in FIG. Generate a visualization screen. FIG. 11 is a diagram illustrating a visualization screen that is another example of the visualization screen, in which the transition of the content of the speech and the transition of the opinion type of the speech are visualized.

  Referring to FIG. 11, on the visualization screen, the horizontal axis indicates the time from the conference start time, and typical utterance content is displayed as a balloon in association with the utterance time. In the figure, detailed description is omitted for the sake of simplicity, but the contents of the text data spoken by the participant can be displayed in the balloon.

  In addition, each text data may be provided with a display indicating the opinion type classified by the opinion extraction unit 141. In the illustrated example, different types of colors are added around the text data to indicate that the opinion type of the text data is one of “affirmation”, “denial”, and “neutral”. .

  In addition, on the visualization screen, a graph indicating the ratio of opinion types of text data uttered between a meeting start time and the time at a representative time can be displayed together. In the illustrated example, the ratio of text data belonging to “affirmation”, “denial”, and “neutral” in “15 min”, “30 min”, and “45 min” is displayed in a pie chart.

  The balloon and the pie chart may be displayed at all times. For example, the balloon and the pie chart at the time corresponding to the position of the pointer by moving the pointer on the display indicating the time axis by the participant. A pop-up may be displayed.

  The visualization screen generation unit 170 can generate the visualization screen based on the result of opinion extraction processing by the opinion extraction unit 141.

  By referring to the visualization screen, the conference participant can intuitively grasp the transition of the content of the speech during the conference and the transition of the opinion type of the speech during the conference. For example, facilitating participants check these transitions to confirm whether the direction of discussion is not deviated from the original schedule, and correct the direction if necessary. So that other participants can be encouraged.

  If the discussion is inconsistent, the visualization screen generation unit 170 may generate the visualization screen shown in FIG. 7 described above instead of the visualization screen shown in FIG. As described above, on the visualization screen shown in FIG. 7, the transition of keywords and the transition of opinion types along the time series are displayed. Therefore, the visualization screen can also be an effective screen for allowing participants to confirm the direction of discussion.

(4-6. End of the meeting)
The situation prediction unit 160 may predict that the meeting is approaching the final stage based on the result of the opinion extraction process by the opinion extraction unit 141. In general, when a conference ends, it is considered that an agreement is often formed on the opinions of the participants. Accordingly, for example, the situation prediction unit 160 determines that the current situation is a meeting when text data whose opinion type is “affirmation” is greater than or equal to a predetermined ratio for text data remarked within a predetermined time. Predict that the situation is approaching the end.

  When the situation prediction unit 160 predicts that the meeting is approaching the final stage, the visualization screen generation unit 170 generates, for example, the visualization screen illustrated in FIG. 10 described above. Since the visualization screen shown in FIG. 10 shows the transition of opinion types, the participants of the conference intuitively grasp the transition of the opinion type of the speech made during the conference by referring to the visualization screen. can do. At this time, as shown in FIG. 10, it is preferable that the visualization screen shows a transition of opinion types including at least “affirmation” and “denial”. For example, participants can recognize that an agreement is being formed between participants by confirming that the number of comments belonging to “affirmation” has increased on the visualization screen, and that the meeting is approaching the end. Can do. For example, the facilitating participant can use this as a trigger to guide the discussion toward the end of the conference, and the conference can proceed smoothly.

  When the meeting is approaching the final stage, the visualization screen generation unit 170 may generate the visualization screen shown in FIG. 7 described above instead of the visualization screen shown in FIG. In the visualization screen illustrated in FIG. 7, as in the visualization screen illustrated in FIG. 10, opinion type transitions including at least “affirmation” and “denial” are displayed. Therefore, the visualization screen can also be an effective screen for suggesting the end of the conference to the participants.

(5. Other display examples)
Other display examples other than the display example described in the above (4. Display example of visualization screen) will be described.

(5-1. Visualization screen for positive / negative determination results)
With reference to FIG. 12, an example of the visualization screen about the result of positive / negative determination by the opinion extraction part 141 is demonstrated. FIG. 12 is a diagram illustrating a visualization screen in which the result of positive / negative determination is visualized, which is another example of the visualization screen.

  Referring to FIG. 12, on the visualization screen, the text data corresponding to “affirmation” and “denial” are corresponded as the result of the positive / negative determination of the text data uttered within a predetermined time during the meeting. A bar graph showing the percentage of text data is displayed. In the lower row, combinations of the object and the predicate in the text data determined to be “No” are tabulated, and the appearance frequency is displayed as a bar graph. The visualization screen generation unit 170 can generate the visualization screen based on the result of opinion extraction processing by the opinion extraction unit 141.

  In the example shown in the figure, “power consumption is large”, “data is difficult to understand”, “cost burden is too large”, “data is broken”, “reliability is low”, etc. The combination of the object and the predicate is extracted. It can also be seen that among these combinations, the combination of “high power consumption” is said most frequently during the conference. In the illustrated example, only the result for “No” is displayed, but the result for “Yes” may be displayed in the same manner by an appropriate switching operation by the participant.

  Since the combination of the target object and the predicate is a summary of text data, the participants who have referred to the visualization screen can use the text data related to “affirmation” or “denial” issued during the meeting. You can check the summary list. Also, by displaying the appearance frequency, the participant can grasp what kind of content is most discussed in the meeting.

(5-2. Visualization screen for text data classification results using modalities)
With reference to FIG. 13, an example of the visualization screen about the classification result of the text data using the modality by the opinion extraction unit 141 will be described. FIG. 13 is a diagram illustrating a visualization screen that is another example of the visualization screen, in which the classification result of text data using modalities is visualized.

  Referring to FIG. 13, in the visualization screen, as the classification result using the modalities of the text data uttered within a predetermined time during the meeting, the text data corresponding to each modality (that is, opinion type) is displayed on the visualization screen. A bar graph showing the percentage is displayed. By a selection operation such as moving the pointer to a location corresponding to each modality on the bar graph, a numerical value of a specific ratio of the modality may be displayed as a pop-up. In the example shown in the drawing, the ratio of text data related to “desire” is displayed in a pop-up manner. In the lower row, the combinations of the object and the predicate in the text data classified as “desire” are tabulated, and the appearance frequency is displayed as a bar graph. The visualization screen generation unit 170 can generate the visualization screen based on the result of opinion extraction processing by the opinion extraction unit 141.

  In the example shown in the figure, “I want you to create a document”, “I want to request a correction”, “I want to check the performance”, “I want to bear the cost”, “ A combination of an object and a predicate such as “I want to ask” is extracted. In addition, among these combinations, it can be seen that the combination of “I want you to create a document” is said most frequently during the meeting. In the example shown in the figure, only the result for “Wish” is displayed, but the result for other modalities may be displayed in the same manner by an appropriate switching process by the participant.

  Since the combination of the “object” and the predicate is a summary of text data, the participant who refers to the visualization screen checks the list of statements related to each modality issued during the meeting. be able to. Further, by displaying the frequency, the participant can grasp what kind of content is most discussed in the conference.

(6. Other application examples)
In recent years, there are cases where conferences are held on social media such as SNS using so-called chat tools. Here, a case where the system 1 is applied to a conference held on such social media will be described.

  When a conference is held on social media, the system 1 acquires a message, an image, a URL indicating a link to a document, etc. posted by the conference participant, and a link destination indicated by the message, the image, or the URL. Collect text data. Then, the text data is analyzed, opinions and topics are extracted, and a visualization screen on which the results are visualized is generated.

  FIG. 14 and FIG. 15 show an example of a visualization screen when a meeting is held on social media. 14 and 15 are diagrams illustrating a visualization screen when a meeting is performed on social media, which is another example of the visualization screen.

  Referring to FIGS. 14 and 15, a meeting is performed using a so-called chat tool, and messages and images posted by each participant are displayed in chronological order. The balloons displaying the messages are colored according to the opinion type corresponding to each message extracted by the opinion extraction unit 141. In the illustrated example, different colors are assigned to messages belonging to “affirmation”, “denial”, and “suggestion”. Further, as shown in the figure, an icon for emphasizing that the message indicates “suggestion” may be displayed together with the message belonging to “suggestion”. The visualization screen generation unit 170 can generate the visualization screen based on the result of opinion extraction processing by the opinion extraction unit 141.

  The participants of the conference can visually grasp the opinion type of each participant's message by referring to the visualization screen, so that they can intuitively recognize each participant's thoughts and the trend of the conference. Can do. In addition, a message including a call to another participant such as “Suggestion” is highlighted and displayed, so that the other participant becomes more aware of the “Suggestion” from someone, Is expected to be more active. Note that, in the illustrated example, only three types of opinion, “affirmation”, “denial”, and “proposition” are targeted, but other opinion types may be visualized in the same manner.

  Here, in a chat tool of general social media whose use is not specified for a meeting (for example, “Facebook (registered trademark)” of Facebook, etc.), posted messages and images are only displayed in chronological order. Therefore, it is not necessarily suitable for meetings. In such a chat tool, for example, messages are sequentially displayed in the vertical direction, so it is difficult to access past information, message groups cannot be displayed separately for each session, and dependency between messages There may be inconveniences such as it is difficult to understand (for which message a message is issued).

  Therefore, chat tools specialized for conferences have been developed (for example, “Air-Campus (registered trademark)” of Business Breakthrough, “Co-meeting (registered trademark)” of Co-meeting, etc.) . These tools are equipped with various functions for smooth meetings, such as the ability to discuss multiple topics simultaneously using a hierarchical discussion chart and the real-time simultaneous editing note function. It can solve the problem.

  However, in these tools, the intention of the speaker is not visualized. In this embodiment, as shown in FIGS. 14 and 15, since the intention of the speaker is visualized and displayed so that each participant can intuitively grasp it, the meeting can be performed more smoothly. Become.

  Note that the visualization screen when the system 1 is applied to a conference held on social media is not limited to that shown in FIGS. 14 and 15, and even in this case, the above-described FIGS. A visualization screen similar to FIG. 12 may be generated.

(7. Hardware configuration)
A hardware configuration of the information processing apparatus according to the present embodiment will be described with reference to FIG. FIG. 16 is a block diagram illustrating an example of a hardware configuration of the information processing apparatus according to the present embodiment. Note that the information processing apparatus 900 illustrated in FIG. 16 can realize the information processing apparatus 10 illustrated in FIG. 2.

  The information processing apparatus 900 includes a CPU 901, a ROM (Read Only Memory) 903, and a RAM (Random Access Memory) 905. The information processing apparatus 900 may include a host bus 907, a bridge 909, an external bus 911, an interface 913, an input device 915, an output device 917, a storage device 919, a communication device 921, a drive 923, and a connection port 925. The information processing apparatus 900 may include a processing circuit such as a DSP or an ASIC in place of or in addition to the CPU 901.

  The CPU 901 functions as an arithmetic processing unit and a control unit, and controls all or a part of the operation in the information processing apparatus 900 according to various programs recorded in the ROM 903, the RAM 905, the storage apparatus 919, or the removable recording medium 929. The ROM 903 stores programs used by the CPU 901, calculation parameters, and the like. The RAM 905 temporarily stores programs used in the execution of the CPU 901, parameters at the time of execution, and the like. The CPU 901, the ROM 903, and the RAM 905 are connected to each other by a host bus 907 configured by an internal bus such as a CPU bus. In the present embodiment, for example, each function of the information processing apparatus 10 described above can be realized by the CPU 901 operating according to a predetermined program.

  The host bus 907 is connected to an external bus 911 such as a PCI (Peripheral Component Interconnect / Interface) bus via a bridge 909.

  The input device 915 is configured by a device operated by a user, such as a mouse, a keyboard, a touch panel, a button, a switch, and a lever. The input device 915 may be, for example, a remote control device (so-called remote controller) that uses infrared rays or other radio waves, or an external connection device such as a mobile phone or a PDA that supports the operation of the information processing device 900. It may be 931. Furthermore, the input device 915 includes an input control circuit that generates an input signal based on information input by the user using the above-described operation means and outputs the input signal to the CPU 901, for example. A user of the information processing apparatus 900 can input various data and instruct a processing operation to the information processing apparatus 900 by operating the input device 915. In the present embodiment, for example, information related to a dictionary used in opinion extraction processing by the opinion extraction unit 141 may be input by the user via the input device 915.

  The output device 917 is a device that can notify the user of the acquired information visually or audibly. Examples of such devices include CRT display devices, liquid crystal display devices, plasma display devices, EL display devices, display devices such as lamps, audio output devices such as speakers and headphones, printer devices, and the like. For example, the output device 917 outputs results obtained by various processes performed by the information processing apparatus 900. Specifically, the display device visually displays results obtained by various processes performed by the information processing device 900 in various formats such as text, images, tables, and graphs. In the present embodiment, for example, the analysis result by the speech analysis unit 110 and the text data analysis unit 140, the visualization screen generated by the visualization screen generation unit 170, and the like may be displayed on the display device. On the other hand, the audio output device converts an audio signal composed of reproduced audio data, acoustic data, and the like into an analog signal and outputs it aurally.

  The storage device 919 is a data storage device configured as an example of a storage unit of the information processing device 900. The storage device 919 includes, for example, a magnetic storage device such as an HDD, a semiconductor storage device, an optical storage device, or a magneto-optical storage device. The storage device 919 stores programs executed by the CPU 901, various data, various data acquired from the outside, and the like. In the present embodiment, the storage device 919 can constitute the text data DB 130 and the analysis result DB 150, and the text data collected by the text data collection unit 120 and the analysis by the utterance analysis unit 110 and the text data analysis unit 140. Results etc. can be stored.

  The communication device 921 is a communication interface configured by a communication device or the like for connecting to the network 927, for example. The communication device 921 is, for example, a communication card for wired or wireless LAN (Local Area Network), Bluetooth (registered trademark), or WUSB (Wireless USB). Further, the communication device 921 may be a router for optical communication, a router for ADSL (Asymmetric Digital Subscriber Line), a modem for various communication, or the like. The communication device 921 can transmit and receive signals and the like according to a predetermined protocol such as TCP / IP, for example, with the Internet or other communication devices. The network 927 connected to the communication device 921 is configured by a wired or wireless network, and may be, for example, the Internet, a home LAN, infrared communication, radio wave communication, satellite communication, or the like. In the present embodiment, for example, the network 927 corresponds to the network 40 illustrated in FIG. The information processing apparatus 10 can communicate with the electronic white board 20 and / or the conference participant terminal 30 via the network 927 by the communication apparatus 921.

  The drive 923 is a recording medium reader / writer, and is built in or externally attached to the information processing apparatus 900. The drive 923 reads information recorded on a removable recording medium 929 such as a mounted magnetic disk, optical disk, magneto-optical disk, or semiconductor memory, and outputs the information to the RAM 905. The drive 923 can also write information to a removable recording medium 929 such as a mounted magnetic disk, optical disk, magneto-optical disk, or semiconductor memory. The removable recording medium 929 is, for example, a DVD medium, an HD-DVD medium, a Blu-ray (registered trademark) medium, or the like. Further, the removable recording medium 929 may be a compact flash (registered trademark) (CompactFlash: CF), a flash memory, an SD memory card (Secure Digital memory card), or the like. The removable recording medium 929 may be, for example, an IC card (Integrated Circuit card) on which a non-contact IC chip is mounted, an electronic device, or the like. In this embodiment, various types of information processed by the CPU 901 may be read from the removable recording medium 929 or written to the removable recording medium 929 by the drive 923.

  The connection port 925 is a port for directly connecting a device to the information processing apparatus 900. Examples of the connection port 925 include a USB (Universal Serial Bus) port, an IEEE 1394 port, a SCSI (Small Computer System Interface) port, and the like. As another example of the connection port 925, there are an RS-232C port, an optical audio terminal, an HDMI (registered trademark) (High-Definition Multimedia Interface) port, and the like. By connecting the external connection device 931 to the connection port 925, the information processing apparatus 900 acquires various data directly from the external connection device 931 or provides various data to the external connection device 931. In the present embodiment, various types of information processed by the CPU 901 may be acquired from the external connection device 931 or output to the external connection device 931 via the connection port 925.

  Heretofore, an example of the hardware configuration of the information processing apparatus 900 according to the present embodiment has been shown. Each component described above may be configured using a general-purpose member, or may be configured by hardware specialized for the function of each component. Therefore, it is possible to change the hardware configuration to be used as appropriate according to the technical level at the time of carrying out this embodiment.

  Note that a computer program for realizing each function of the information processing apparatus 900 according to the present embodiment as described above can be produced and mounted on a PC or the like. In addition, a computer-readable recording medium storing such a computer program can be provided. The recording medium is, for example, a magnetic disk, an optical disk, a magneto-optical disk, a flash memory, or the like. Further, the above computer program may be distributed via a network, for example, without using a recording medium.

(8. Supplement)
The preferred embodiments of the present disclosure have been described in detail above with reference to the accompanying drawings, but the technical scope of the present disclosure is not limited to such examples. It is obvious that a person having ordinary knowledge in the technical field of the present disclosure can come up with various changes or modifications within the scope of the technical idea described in the claims. Of course, it is understood that it belongs to the technical scope of the present disclosure.

  In addition, the effects described in the present specification are merely illustrative or illustrative, and are not limited. That is, the technology according to the present disclosure can exhibit other effects that are apparent to those skilled in the art from the description of the present specification in addition to or instead of the above effects.

  For example, each of the visualization screens exemplified in the above embodiment displays the classification result of the text data and the speaker of the text data in a one-to-one correspondence. It is not limited to such an example. For example, a visualization screen may be generated in which an opinion type that is a classification result of text data is visualized in association with an attribute of a speaker (for example, gender, age, nationality, residence, etc.). In the visualization screen, since the opinion type corresponding to the attribute of the speaker on a certain agenda is statistically displayed, it is different from the embodiment described above for the participants who referred to the visualization screen. Can be noticed. For example, by referring to the visualization screen and grasping the tendency of the opinion type according to the attribute of the speaker, the meeting can be performed more smoothly.

  In addition, the visualization screens exemplified in the above embodiment are all displayed with the text data classification result associated with the speaker of the text data (that is, the text data classification result is the text data speech. However, the present embodiment is not limited to such an example. For example, a visualization screen may be generated in which text data classification results are summarized from different viewpoints such as the time when the text data was spoken or the content of the text data. For example, it may be possible to obtain a new awareness about the transition of opinion types according to the progress of the conference by collectively displaying the opinion types that have appeared frequently at the beginning, middle, and end of the conference. In addition, for example, by arranging and displaying text data including a predetermined word according to the opinion type, there is a possibility that a new awareness may be obtained about the relationship between the word and the opinion type.

The following configurations also belong to the technical scope of the present disclosure.
(1) A visualization screen generation unit that generates a visualization screen that is a screen in which text data issued by a plurality of speakers is classified into a predetermined category and visualized in association with the speaker. Processing equipment.
(2) The information processing apparatus according to (1), wherein the text data is classified into a category according to the intention of the speaker.
(3) The information processing apparatus according to (2), wherein the text data is classified into one of positive, negative, and neutral.
(4) The information processing apparatus according to (2), wherein the text data is classified using a modality.
(5) The information processing apparatus according to any one of (1) to (4), wherein the text data is classified according to a latent topic included in the text data.
(6) The text data is text data uttered by a participant during a meeting, and the visualization screen generation unit, in the case where it is predicted that the current situation is the beginning of the meeting, The information processing apparatus according to any one of (1) to (5), wherein the generated visualization screen is generated.
(7) The text data is text data spoken by a participant during a meeting, and the visualization screen generation unit is predicted that the current situation is a situation in which a plurality of participants are discussing. In the case, the visualization screen on which at least one of the transition of the intention of the speaker of the text data, the transition of the latent topic that the text data has, and the transition of the content of the text data is visualized, The information processing apparatus according to any one of (1) to (5).
(8) The information processing apparatus according to (7), wherein the visualization screen generation unit causes the visualization screen on which the transition of the latent topic is visualized to display the time when the topic of discussion has changed.
(9) The text data is text data remarked by a participant during a meeting, and the visualization screen generation unit, when it is predicted that the current situation is a situation where the meeting is stagnant, The information processing apparatus according to any one of (1) to (5), wherein a section in which each participant speaks during a meeting and the visualization screen on which the intention of each participant's speech is visualized are generated. .
(10) In the above (9), the current situation is predicted to be a situation where the meeting is stagnant when a non-speaking section in which no participant speaks continues for a predetermined time or more. The information processing apparatus described.
(11) The text data is text data remarked by a participant during a meeting, and the visualization screen generation unit, when it is predicted that the current situation is a situation in which the discussion is confused, The information processing apparatus according to any one of (1) to (5), wherein the visualization screen on which a transition of an intention of a speaker of the text data and a transition of the content of the text data are visualized is generated.
(12) When the utterance collision section in which a plurality of participants in the conference speak at the same time occurs more than a predetermined number of times, the current situation is predicted to be a situation in which the discussion is confused (11) The information processing apparatus described in 1.
(13) The text data is text data uttered by a participant during a meeting, and the visualization screen generation unit, when it is predicted that the current situation is a situation where the meeting is approaching the final stage The information processing apparatus according to any one of (1) to (5), wherein the visualization screen on which an intention of a speaker of the text data including at least positive and negative is visualized.
(14) When the number of text data that is estimated to be positive by the speaker with respect to the total number of text data within a predetermined time has reached a predetermined ratio or more, the current situation is at the end of the meeting. The information processing apparatus according to (13), wherein the information processing apparatus is predicted to be in an approaching state.
(15) The visualization screen generation unit is a predetermined combination of a clause corresponding to a predicate and an object to be a target of the previous description unit of text data related to any of the classified speaker's intentions The information processing apparatus according to any one of (2) to (4), wherein the visualization screen is generated in which the appearance frequency within the time is visualized.
(16) The information processing apparatus according to any one of (1) to (15), wherein the text data is text data exchanged by a user on social media.
(17) The processor includes generating a visualization screen that is a screen in which text data issued by a plurality of speakers is classified into a predetermined category and visualized in association with the speaker. Processing method.
(18) A visualization screen generation function for generating, on a computer, a visualization screen which is a screen in which text data issued by a plurality of speakers is classified into a predetermined category and visualized in association with the speaker. A program to make it happen.

DESCRIPTION OF SYMBOLS 1 System 10 Information processing apparatus 20 Electronic white board 30 Conference terminal 40 Network 110 Speech analysis part 120 Text data collection part 130 Text data DB
140 Text Data Analysis Unit 141 Opinion Extraction Unit 142 Topic Analysis Unit 150 Analysis Result DB
160 Situation Prediction Unit 170 Visualization Screen Generation Unit 180 Summarization Unit 190 Search Unit 201 Sentence Division Unit 203 Syntax / Semantic Analysis Unit 205 Business Glossary 207 Dictionary Generation Unit 209 Business Term Dictionary 211 Opinion / Evaluation Expression Extraction Unit

Claims (18)

A visualization screen generating unit that generates a visualization screen that is a screen in which text data issued by a plurality of speakers is classified into a predetermined category and visualized in association with the speaker;
An information processing apparatus comprising: The text data is classified into a category according to the intention of the speaker.
The information processing apparatus according to claim 1. The text data is classified as either positive, negative, or neutral.
The information processing apparatus according to claim 2. The text data is classified using a modality.
The information processing apparatus according to claim 2. The text data is classified according to potential topics that the text data has.
The information processing apparatus according to claim 1. The text data is text data spoken by a participant during a meeting,
The visualization screen generation unit generates the visualization screen generated in the previous conference when the current situation is predicted to be the beginning of the conference,
The information processing apparatus according to claim 1. The text data is text data spoken by a participant during a meeting,
When the current screen state is predicted to be a state where discussions are being conducted by a plurality of participants, the visualization screen generation unit may change the intention of the speaker of the text data and the potential of the text data. Generating the visualization screen in which at least one of the transition of the topic and the transition of the content of the text data is visualized;
The information processing apparatus according to claim 1. The visualization screen generation unit causes the visualization screen on which the transition of the latent topic is visualized to be displayed together with the time when the topic of discussion has changed,
The information processing apparatus according to claim 7. The text data is text data spoken by a participant during a meeting,
When it is predicted that the current situation is a situation in which the conference is stagnant, the visualization screen generation unit visualizes a section where each participant speaks during the conference, and the intention of each participant's speech Generating the visualized screen,
The information processing apparatus according to claim 1. If a non-speaking section in which no participant in the meeting has spoken continues for a predetermined time or longer, the current situation is predicted to be a situation where the meeting is stagnant,
The information processing apparatus according to claim 9. The text data is text data spoken by a participant during a meeting,
When it is predicted that the current situation is a situation in which the discussion is inconsistent, the visualization screen generation unit visualizes the transition of the intention of the speaker of the text data and the transition of the content of the text data. Generating the visualization screen,
The information processing apparatus according to claim 1. When the number of utterance collision sections where a plurality of participants in the conference speak at the same time occurs more than a predetermined number of times, the current situation is predicted to be a situation where the discussion is in conflict.
The information processing apparatus according to claim 11. The text data is text data spoken by a participant during a meeting,
The visualization screen generation unit visualizes the speaker's intention of the text data including at least affirmation and denial when the current situation is predicted to be a situation where a meeting is about to end Generate
The information processing apparatus according to claim 1. The current situation is approaching the end of the meeting when the number of text data estimated to be positive by the speaker is greater than a certain percentage of the total number of text data in a given time Predicted to be the situation,
The information processing apparatus according to claim 13. The visualization screen generation unit is a predetermined time of a combination of a clause corresponding to the predicate and the target object of the previous description unit of the text data related to any of the classified speaker's intentions Generating the visualization screen in which the appearance frequency in is visualized,
The information processing apparatus according to claim 2. The text data is text data exchanged by a user on social media.
The information processing apparatus according to claim 1. A processor generates a visualization screen that is a screen in which text data issued by a plurality of speakers is classified into a predetermined category and visualized in association with the speaker;
Including an information processing method. On the computer,
A visualization screen generation function for generating a visualization screen that is a screen in which text data issued by a plurality of speakers is classified into a predetermined category and visualized in association with the speaker;
A program to realize

Images