Conference support device, conference support system, conference support method, and program

Abstract

The goal is to facilitate consideration by healthcare professionals and reduce omissions and oversights in the review process. [Solution] The meeting support device according to the embodiment comprises an action information acquisition unit, an evaluation result acquisition unit, and an output unit. The action information acquisition unit acquires action information relating to the actions of participants in a meeting. The evaluation result acquisition unit inputs the action information into a generation model and acquires evaluation results from the generation model relating to the progress of consideration of multiple items to be considered in the meeting. The output unit outputs the evaluation results.

Description

Technical Field

[0001] The embodiments disclosed in this specification and the drawings relate to a conference support device, a conference support system, a conference support method, and a program.

Background Art

[0002] Conventionally, in hospitals, medical professionals have been discussing treatment plans suitable for patients through case conferences. Since this case conference involves medical professionals from multiple occupations such as surgeons, oncologists, radiologists, histopathologists, and cancer specialist nurses, medical professionals can discuss treatment plans from various perspectives. Also, by sharing the treatment plans discussed in this conference among medical professionals, medical professionals can provide safer and higher-quality medical services to patients. However, due to scheduling constraints, if only limited types of medical professionals can participate in the case conference, or if the discussion does not converge because the discussion progresses without determining the goal of the conference, there is a risk that the treatment plan cannot be fully discussed in the case conference due to scheduling or discussion progress issues.

[0003] However, in a case conference, it is desirable to discuss step by step so that the discussion is promoted and made more efficient, and it is necessary to present support information such as advice according to the progress of the discussion. Also, it is desired to present support information such as advice according to the progress of the discussion not only in case conferences but also in other conferences such as medical interviews.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] One of the problems that the embodiments disclosed in this specification and drawings aim to solve is to facilitate the review process by healthcare professionals and reduce omissions and oversights in their review. However, the problems that the embodiments disclosed in this specification and drawings aim to solve are not limited to the above problem. Problems corresponding to the effects of each configuration shown in the embodiments described later can also be positioned as other problems. [Means for solving the problem]

[0006] The meeting support device according to the embodiment comprises an action information acquisition unit, an evaluation result acquisition unit, and an output unit. The action information acquisition unit acquires action information relating to the actions of participants in a meeting. The evaluation result acquisition unit inputs the action information into a generation model and acquires evaluation results from the generation model relating to the progress of consideration of multiple items to be considered in the meeting. The output unit outputs the evaluation results. [Brief explanation of the drawing]

[0007] [Figure 1] This is a diagram illustrating the overview of the conference system according to the first embodiment. [Figure 2] This is a block diagram showing an example of the configuration of a conference system according to the first embodiment. [Figure 3] This is a block diagram showing an example of the configuration of a conference support device according to the first embodiment. [Figure 4] This figure shows an example of the data used to train the topic analysis model in the first embodiment. [Figure 5] This figure shows an example of input data input to a topic analysis model and output data output from a topic analysis model in the first embodiment. [Figure 6] This block diagram shows an example of the configuration of a server device according to the first embodiment. [Figure 7] This figure shows an example of input data input to a large-scale language model and output data output from the large-scale language model in the first embodiment. [Figure 8]This figure shows an example of input data input to a large-scale language model and output data output from the large-scale language model in the first embodiment. [Figure 9] This is a flowchart illustrating the contents of the support information output process performed in the conference support device according to the first embodiment. [Figure 10] This figure shows an example of target time-series data in the target information according to the first embodiment. [Figure 11] This figure shows an example of a target radar chart in the target information according to the first embodiment. [Figure 12] This figure shows an example of a first prompt according to the first embodiment. [Figure 13] This figure shows an example of the evaluation results according to the first embodiment. [Figure 14] This diagram illustrates the sequence of steps involved in determining a topic of study using the topic determination function according to the first embodiment. [Figure 15] This figure shows an example of a second prompt according to the first embodiment. [Figure 16] This figure shows an example of support information according to the first embodiment. [Figure 17] This figure shows an example of target information when the length of hospital stay is short in the modified example 1. [Figure 18] This figure shows an example of target information in the case of a long hospital stay in the first variation. [Figure 19] This figure shows an example of the configuration of a conference support device related to Modification 2. [Figure 20] This figure shows an example of a learning method for the topic selection model related to Modification Example 2. [Figure 21] This flowchart illustrates the content of the support information output process performed in the conference support device according to Modification Example 2. [Figure 22] This diagram illustrates the sequence of steps involved in determining a topic of consideration using the topic determination function related to Modification Example 2. [Figure 23] This is a block diagram showing an example of the configuration of a conference support device according to the second embodiment. [Figure 24] It is a diagram showing an example of input data input to a large language model according to the second embodiment and output data output from the large language model. [Figure 25] It is a diagram showing an example of input data input to a large language model according to the second embodiment and output data output from the large language model. [Figure 26] It is a flowchart for explaining the content of support information output processing executed in the conference support device according to the second embodiment. [Figure 27] It is a diagram showing an example of the first prompt according to the second embodiment. [Figure 28] It is a diagram showing an example of the second prompt according to the second embodiment. [Figure 29] It is a diagram showing an example of information output to the output interface in the second embodiment. [Figure 30] It is a diagram showing another example of information output to the output interface in the second embodiment. [Figure 31] It is a block diagram showing an example of the configuration of a conference support device according to the third embodiment. [Figure 32] It is a block diagram showing an example of the configuration of a server device according to the third embodiment. [Figure 33] It is a flowchart for explaining the content of support information output processing executed in the conference support device according to the third embodiment. [Figure 34] It is a diagram showing an example of an evaluation result according to the third embodiment. [Figure 35] It is a block diagram showing an example of the configuration of a conference support device according to the fourth embodiment.

Embodiments for Carrying Out the Invention

[0008] Hereinafter, embodiments of a conference support device, a conference support system, a conference support method, and a program will be described with reference to the drawings. In the following description, components having substantially the same functions and configurations will be denoted by the same reference numerals, and duplicate explanations will be made only when necessary.

[0009] [Summary of the First Embodiment] Figure 1 is a diagram illustrating the overview of the conference system according to the first embodiment. In this embodiment, the conference is a specialist conference to discuss the treatment plan for a patient, i.e., a case conference. For example, as shown in Figure 1, the conference system first (1) acquires patient information PF, which will be described later, and the target information generation function generates target information TF related to the patient's case included in the patient information, based on the conference log data CL of the medical professionals and the patient information PF of the patient associated with the conference log data.

[0010] (2) The behavior log data acquisition function acquires behavior log data from the behavior log DB (Database). (3) The prompt generation function converts the text and typing behavior, or the content of spoken words, voice tone and speed, etc., included in this behavior log data into text and generates prompts.

[0011] (4) The evaluation result acquisition function inputs prompts into a large language model and obtains evaluation result ER from the large language model. (5) The discussion topic determination function generates discussion result SR based on the target information TF and the evaluation result ER, and determines the discussion topic which is the topic to be discussed in the case conference.

[0012] (6) The support information acquisition function inputs the review results, behavior log data, and large-scale language model, and obtains advice on the decided review topic as support information (SF) from the large-scale language model to assist participants in the case review meeting. (7) The meeting support system outputs advice on the decided review topic. This allows the meeting support system to facilitate the review by participants in the case review meeting and reduce omissions and oversights in the review.

[0013] [Detailed description of the first embodiment] Next, the conference support system according to this embodiment will be described in detail. Figure 2 is a block diagram showing an example of the configuration of a conference system according to the first embodiment. As shown in Figure 2, the conference support system 1 is configured to include a conference support device 10 and a server device 30. The conference support device 10 and the server device 30 are connected to each other via a network NW so that they can communicate with each other.

[0014] The conference support device 10 is a terminal device operated by the conference participants. The conference support device 10 acquires the activity log data described later and outputs the support information described later. The conference support device 10 is, for example, a computer used by the conference participants, or a smartphone or tablet carried by the conference participants.

[0015] Here, "meeting" is not limited to a meeting in the narrow sense, but also includes interviews, consultations, counseling, medical examinations, customer service, meetings, and discussions. Furthermore, meetings include medical examinations conducted by doctors with patients, and case conferences, which are expert meetings to discuss treatment plans for patients. Participants in such meetings are, for example, healthcare professionals. These healthcare professionals include, for example, surgeons, oncologists, radiologists, histopathologists, and oncology nurse specialists. It should also be noted that healthcare professionals other than surgeons, oncologists, radiologists, histopathologists, and oncology nurse specialists may be included. Additionally, participants in the meeting may include healthcare professionals, or other participants may be included in place of healthcare professionals. The following explanation will use the example of a case conference where the meeting is a case conference and the participants are healthcare professionals.

[0016] The server device 30 is, for example, a cloud server on a network and has functions for evaluating the progress of the study described later and for generating support information. In this embodiment, the server device 30 is an LLM server having a Large Language Model (LLM) as a generative model, and uses the Large Language Model to generate evaluation results of the study's progress and support information. Note that the Large Language Model can also be referred to as the generative model. Furthermore, the server device 30 is not limited to a cloud server, but may be a server installed within a hospital.

[0017] The term "network" (NW) refers to all information and communication networks that utilize telecommunications technology. Examples of network networks include wireless / wired LANs such as hospital backbone LANs (Local Area Networks), the Internet, as well as telecommunications circuit networks, fiber optic communication networks, cable communication networks, and satellite communication networks.

[0018] Figure 3 is a block diagram showing an example of the configuration of a conference support device according to the first embodiment. As shown in Figure 3, the conference support device 10 is configured to include an input interface 11, an output interface 13, a communication interface 15, a storage circuit 17, and a processing circuit 19. The multiple elements for configuring this conference support device 10 may be housed in a single enclosure or distributed and housed in multiple enclosures.

[0019] The input interface 11 receives various input operations from the operator, converts the received input operations into electrical signals, and outputs them to the processing circuit 19. The input interface 11 includes, for example, a mouse, keyboard, trackball, switch, button, joystick, touch panel, etc. The input interface 11 may also be a user interface that accepts audio input, such as from a microphone. If the input interface 11 is a touch panel, the input interface 11 may also incorporate the display function of a display included in the output interface 13, which will be described later.

[0020] Furthermore, the input interface 11 is not limited to those equipped with physical operating components such as a mouse or keyboard. For example, an electrical signal processing circuit that receives an electrical signal corresponding to an input operation from an external input device located separately from the device and outputs this electrical signal to a control circuit is also included as an example of the input interface 11.

[0021] The output interface 13 includes, for example, a display and a speaker. The display provided by the output interface 13 corresponds to the display unit in this embodiment. The display shows various types of information. For example, the display shows the screen output by the output function of the processing circuit 19, or a GUI (Graphical User Interface) for receiving various input operations from the operator. The display is, for example, an LCD (Liquid Crystal Display), a CRT (Cathode Ray Tube) display, or an organic EL (Electro Luminescence) display. The display of this output interface 13 corresponds to the display unit in this embodiment. The speaker outputs the information input from the processing circuit 19 as sound.

[0022] The communication interface 15 implements various communication protocols depending on the configuration of the network NW. The communication interface 15 enables communication with other devices via the network NW according to the various communication protocols. In this embodiment, the conference support device 10 communicates with the server device 30 via the communication interface 15.

[0023] The memory circuit 17 is a non-transient memory device that stores various types of information and can be implemented using, for example, semiconductor memory elements such as RAM (Random Access Memory) and flash memory, a hard disk, or an optical disc. For example, the memory circuit 17 receives and stores various types of data acquired by the processing circuit 19. The memory circuit 17 also stores programs corresponding to various functions executed by circuits included in the conference support device 10. The memory circuit 17 may also be implemented using a group of servers (cloud) connected to the conference support device 10 via a network NW. The various types of data dealt with in this specification are typically digital data. As shown in Figure 3, this embodiment includes a conference log database 171, an activity log database 173, a patient information database 175, and a topic analysis model 177.

[0024] The meeting log database 171 stores data related to past meetings. This data includes, for example, audio data and document data such as meeting minutes. In this embodiment, since the meetings are case conferences, the data related to past meetings includes information such as patient treatment plans. The meeting log database 171 stores data related to past meetings in association with patient information of the patients who were the subject of the meetings.

[0025] The behavior log database 173 stores behavior log data relating to the actions of participants in the current meeting. This behavior log database 173 stores data related to the current meeting acquired via the input interface 11, output interface 13, or communication interface 15 in real time. When the meeting ends, the behavior log data may be stored in the meeting log database 171 as meeting log data. The behavior log data includes, for example, audio data of the current meeting and document data such as meeting minutes of the current meeting. The behavior log data corresponds to the behavior information according to this embodiment.

[0026] The patient information database 175 stores various types of information about the patient. For example, the patient information database 175 stores the patient's name and identification information, along with information about the patient's attributes. Patient attributes include, for example, the patient's occupation, case, condition, treatment progress, medical history, age, sex, height, weight, and place of residence. The patient information database 175 may also store information from the electronic medical record.

[0027] Topic analysis model 177 is a pre-trained model that takes patient information as input data and outputs target information. This topic analysis model 177 is constructed, for example, by employing a regression model. However, while the topic analysis model 177 is constructed using a regression model as described above, it is not limited to this. For example, topic analysis model 177 may be constructed using a neural network.

[0028] The method for generating the topic analysis model 177 is described below. First, the conference support device 10 obtains conference log data from the conference log database 171. Then, the conference support device 10 inputs the audio data and document data in the conference log data into the large-scale language model of the server device 30, which will be described later, via the communication interface 15, thereby obtaining topic information that sorts which of the multiple topics the statements made by participants in the conference, contained in the audio data and / or document data in the conference log data, belong to. The multiple topics in a case conference in which medical professionals participate include at least one of the following: a topic related to the policy of the case conference, a topic related to the discussion of the case conference, and a topic related to decision-making. Specifically, the multiple topics include the conference goal (GOAL) and the discussion process as topics related to the policy of the case conference, information exchange and misinformation detection as topics related to the discussion of the case conference, and treatment policy and precautions as topics related to decision-making.

[0029] Next, the conference support device 10 calculates the utterance score for each topic for each elapsed time based on the acquired topic information. This utterance score is calculated, for example, based on the utterance rate. In this embodiment, the conference support device 10 calculates the utterance score using utterance score = utterance rate. This utterance rate is calculated using utterance rate = utterance time for the target topic / utterance time for the meeting, or utterance rate = number of utterances for the target topic / total number of utterances for the meeting. The calculated utterance rate is normalized so that it is between a minimum of 0 and a maximum of 1. Then, the conference support device 10 generates a graph of the utterance scores for each topic at each time point based on the calculated utterance scores.

[0030] Figure 4 shows an example of a graph used for training a topic analysis model in the first embodiment. As shown in Figure 4, the conference support device 10 generates a graph G1 of the utterance scores for each topic at each time point, with the horizontal axis representing elapsed time and the vertical axis representing the utterance score. In the example shown in Figure 4, at the beginning of the conference, the utterance scores for topics related to the conference's GOAL and process are high. In the middle of the conference, the utterance scores for topics related to information exchange are high, and a score related to misinformation detection is also calculated. In the end of the conference, the utterance scores for treatment policies are high, and a score related to precautions is also calculated. The conference support device 10 then performs the process of generating a graph of the utterance scores for each topic at each time point for each of the conference log data stored in the conference log database 171.

[0031] Next, the conference support device 10 constructs a topic analysis model 177 by performing machine learning using multiple training data sets, which include graphs of speech scores for each topic at each time point and patient information associated with those graphs. Specifically, the conference support device 10 performs a fitting process on the graph G1 of the calculated speech scores for each topic at each time point, which is one of the training data sets, and performs machine learning to construct the topic analysis model 177 so that the output data approaches the correct data.

[0032] Figure 5 shows an example of input data input to the topic analysis model 177 and output data output from the topic analysis model 177 in the first embodiment. As shown in Figure 5, the topic analysis model 177 takes patient information PF as input data and target information TF as output data. In the example shown in Figure 5, the target information TF is target time series data TTD obtained by fitting the graph G1 of the speech scores of each topic at each time point so that the graph G1 of the speech scores of each topic at each time point follows a normal distribution.

[0033] The processing circuit 19 is a control circuit that performs overall control of the conference support device 10, and is also an arithmetic circuit that performs various calculations. In this embodiment, the processing circuit 19 includes a patient information acquisition function 191, a target information generation function 192, an action log data acquisition function 193, a prompt generation function 194, an evaluation result acquisition function 195, an output function 196, a discussion topic determination function 197, and a support information acquisition function 198.

[0034] Here, for example, the processing functions performed by the components of the processing circuit 19 shown in Figure 3—the patient information acquisition function 191, the target information generation function 192, the behavior log data acquisition function 193, the prompt generation function 194, the evaluation result acquisition function 195, the output function 196, the consideration topic determination function 197, and the support information acquisition function 198—are recorded in the memory circuit 17 in the form of a program that can be executed by a computer. The processing circuit 19 is, for example, a processor. The processor constituting the processing circuit 19 reads the program from the memory circuit 17 and executes it to realize the function corresponding to the read program. In other words, the processing circuit 19 in the state where a program has been read has the functions shown in the processing circuit 19 of Figure 3.

[0035] In Figure 3, the processing circuit 19 is shown to implement the patient information acquisition function 191, the target information generation function 192, the behavior log data acquisition function 193, the prompt generation function 194, the evaluation result acquisition function 195, the output function 196, the consideration topic determination function 197, and the support information acquisition function 198. However, the embodiment is not limited to this. For example, the processing circuit 19 may be composed of a combination of multiple independent processors, with each processor executing its respective program to realize these functions. Furthermore, the processing functions of the processing circuit 19 may be implemented by appropriately distributing or integrating them across one or more processing circuits.

[0036] The patient information acquisition function 191 acquires patient information, which includes at least the patient's case. The patient information acquisition function 191 corresponds to the patient information acquisition unit in this embodiment.

[0037] The goal information generation function 192 generates goal information. The goal information is information regarding the progress of discussions on multiple topics to be considered in the meeting. In this embodiment, the goal information generation function 192 generates goal information using the topic analysis model 177. Specifically, the goal information generation function 192 inputs patient information acquired by the patient information acquisition function 191 into the topic analysis model 177. The goal information generation function 192 also acquires goal information as the output of the topic analysis model 177. This goal information generation function 192 corresponds to the goal information generation function 192 in this embodiment. Furthermore, multiple topics correspond to multiple items in this embodiment.

[0038] The behavior log data acquisition function 193 acquires behavior log data. The behavior log data acquisition function 193 corresponds to the behavior information acquisition unit in this embodiment.

[0039] The prompt generation function 194 generates a first prompt for input to the large-scale language model described later, based on the behavior log data and multiple topics. In this embodiment, the first prompt generated by the prompt generation function 194 includes multiple topics and behavior log data. The prompt generation function 194 also generates a second prompt for input to the large-scale language model, based on the discussion topics described later and behavior log data. In this embodiment, the second prompt generated by the prompt generation function 194 includes the discussion topics and behavior log data. The prompt generation function 194 corresponds to the prompt generation unit.

[0040] The evaluation result acquisition function 195 inputs behavior log data into a large-scale language model and obtains evaluation results from the large-scale language model regarding the status of discussion on multiple topics to be considered in the meeting. Specifically, the evaluation result acquisition function 195 inputs prompts into the large-scale language model and obtains evaluation results ER from the large-scale language model. This evaluation result acquisition function 195 inputs behavior log data into the large-scale language model at predetermined intervals and obtains evaluation results ER from the large-scale language model. In this embodiment, the evaluation result acquisition function 195 also obtains evaluation results ER regarding the progress status for each of the multiple topics. The evaluation result acquisition function 195 corresponds to the evaluation result acquisition unit in this embodiment.

[0041] The output function 196 outputs the evaluation result ER. The output function 196 corresponds to the output unit in this embodiment.

[0042] The discussion topic determination function 197 determines the discussion topic, which is the topic to be discussed in the meeting, from among several topics, based on the target information TF and the evaluation result ER. The discussion topic determination function 197 corresponds to the discussion item determination unit in this embodiment.

[0043] The support information acquisition function 198 inputs the discussion topic and behavior log data into a large-scale language model and obtains support information SF from the large-scale language model to support participants in the meeting. Support information SF is, for example, information about advice to help meeting participants consider the discussion topic. Support information SF includes, for example, verbal information, and visual information (including image information) using drawings, images, videos, etc., as information about advice.

[0044] Next, the server device 30 will be described in detail with reference to Figure 6. Figure 6 is a block diagram showing an example of the configuration of the server device 30 in the first embodiment. As shown in Figure 6, the server device 30 is configured to include a communication interface 31, a storage circuit 33, and a processing circuit 35.

[0045] The communication interface 31 implements various communication protocols depending on the configuration of the network NW. The communication interface 31 enables communication with other devices via the network NW according to the various communication protocols. In this embodiment, the server device 30 communicates with the conference support device 10 via the communication interface 31.

[0046] The memory circuit 33 is a non-transient memory device that stores various types of information and can be implemented using, for example, semiconductor memory elements such as RAM and flash memory, a hard disk, or an optical disc. For example, the memory circuit 33 receives and stores various types of data acquired by the processing circuit 35. The memory circuit 33 also stores programs corresponding to various functions executed by circuits included in the server device 30. The memory circuit 33 may also be implemented using a group of servers (cloud) connected to the server device 30 via a network NW.

[0047] As shown in Figure 6, in this embodiment, the memory circuit 33 includes a large-scale language model 331. The large-scale language model 331 is a language model trained to process natural language and outputs a response in text format or the like in response to an input prompt. Examples of large-scale language models 331 include GPT-3 and GPT-4 developed by OpenAI, and BERT developed by Google.

[0048] Figure 7 shows an example of input data input to the large-scale language model 331 and output data output from the large-scale language model 331 in the first embodiment. As shown in Figure 7, the large-scale language model 331 takes a first prompt P1 as input data. The first prompt P1 in this embodiment includes information about instructions, information about topics, and information about content. The information about instructions is, for example, information about instructions to the large-scale language model 331. The information about content is information about actions, i.e., action log data. Also, as shown in Figure 7, the large-scale language model 331 takes evaluation result ER as output data. In the example shown in Figure 7, the evaluation result ER is a first radar chart showing the progress of consideration for each of the multiple topics.

[0049] Figure 8 shows an example of input data input to the large-scale language model 331 and output data output from the large-scale language model 331 in the first embodiment. As shown in Figure 8, the large-scale language model 331 takes the second prompt P2 as input data. The second prompt P2 in this embodiment includes information about instructions, information about a topic, and information about content. The information about instructions is, for example, information about instructions to the large-scale language model 331. The information about a topic includes information about the topic to be considered determined in step S25 and information about the ratio of the evaluation value to the target value. The information about content is information about the conversation. As shown in Figure 8, the large-scale language model 331 also takes support information SF as output data. In the example shown in Figure 8, the support information SF is information about advice for the meeting participants to consider the topic to be considered.

[0050] The processing circuit 35 is a control circuit that performs overall control of the server device 30, and is also an arithmetic circuit that performs various calculations. In this embodiment, the processing circuit 35 includes an information acquisition function 351, an evaluation function 352, a support information generation function 353, and an information output function 354.

[0051] Here, for example, the processing functions performed by the information acquisition function 351, evaluation function 352, support information generation function 353, and information output function 354, which are components of the processing circuit 35 shown in Figure 4, are recorded in the memory circuit 33 in the form of a program that can be executed by a computer. The processing circuit 35 is, for example, a processor. The processor constituting the processing circuit 35 reads the program from the memory circuit 33 and executes it to realize the functions corresponding to the read program. In other words, the processing circuit 35 in the state in which a program has been read has the functions shown in the processing circuit 35 of Figure 6.

[0052] In Figure 6, the processing circuit 35 is shown to implement the information acquisition function 351, the evaluation function 352, the support information generation function 353, and the information output function 354, but the embodiment is not limited to this. For example, the processing circuit 35 may be composed of a combination of multiple independent processors, with each processor executing its respective program to implement these functions. Furthermore, the processing functions of the processing circuit 35 may be implemented by appropriately distributing or integrating them across one or more processing circuits.

[0053] The information acquisition function 351 receives information from the conference support device 10 via the communication interface 31. The information received by the information acquisition function 351 includes, for example, information about the first prompt generated by the prompt generation function 194 and input to the server device 30 by the evaluation result acquisition function 195, and information about the second prompt input to the server device 30 by the support information acquisition function 198. The information acquisition function 351 corresponds to the information acquisition unit in this embodiment.

[0054] The evaluation function 352 uses the large-scale language model 331 to evaluate the progress of discussions on multiple topics considered in the meeting. Specifically, the evaluation function 352 inputs the first prompt P1 acquired by the information acquisition function 351 to the large-scale language model 331. The evaluation function 352 also acquires an evaluation result ER regarding the progress of discussions on multiple topics considered in the meeting as an output from the large-scale language model 331. The evaluation function 352 corresponds to the evaluation unit in this embodiment.

[0055] The support information generation function 353 generates support information SF to assist participants in a meeting using the large-scale language model 331. Specifically, the support information generation function 353 inputs the second prompt P2 acquired by the information acquisition function 351 to the large-scale language model 331. The support information generation function 353 also acquires the support information SF as the output of the large-scale language model 331. The support information generation function 353 corresponds to the support information generation unit in this embodiment.

[0056] The information output function 354 transmits information to the conference support device 10 via the communication interface 31. The information transmitted by the information output function 354 includes, for example, information regarding evaluation results and support information. The information output function 354 corresponds to the information output unit in this embodiment.

[0057] Figure 9 is a flowchart illustrating the content of the support information output process performed in the conference support device 10 according to the first embodiment. This support information output process is performed, for example, when a conference is started.

[0058] As shown in Figure 9, first, the patient information acquisition function 191 in the processing circuit 19 of the conference support device 10 acquires patient information PF (step S11). Specifically, the patient information acquisition function 191 acquires patient information PF from the patient information database 175 that includes at least the cases of patients who are the subject of the conference.

[0059] Next, as shown in Figure 9, the target information generation function 192 in the processing circuit 19 of the conference support device 10 generates target information TF (step S13). Specifically, the target information generation function 192 generates target information TF related to the patient cases included in the patient information PF acquired in step S11. More specifically, the target information generation function 192 generates target information TF by inputting the patient information PF acquired in step S11 into the topic analysis model 177 using the topic analysis model 177.

[0060] Figure 10 shows an example of target time series data TTD in the target information TF according to the first embodiment. As shown in Figure 10, the target information generation function 192 generates target time series data TTD as the target information TF, which shows the time series changes of target values ​​for the progress of consideration for each of the multiple topics. In the example shown in Figure 10, the target time series data TTD includes the first target time series data TTD1 related to the meeting GOAL, the second target time series data TTD2 related to the process, the third target time series data TTD3 related to information exchange, the fourth target time series data TTD4 related to misinformation detection, the fifth target time series data TTD5 related to treatment policies, and the sixth target time series data TTD6 related to precautions. As shown in Figure 10, the first target time series data TTD1, the second target time series data TTD2, the third target time series data TTD3, ​​the fourth target time series data TTD4, the fifth target time series data TTD5, and the sixth target time series data TTD6 are all normally distributed.

[0061] Furthermore, as shown in Figure 10, the target information generation function 192 divides the target time series data TTD into three intervals: the beginning, the middle, and the end. The beginning is the interval from elapsed time t0 to t1, the middle is the interval from elapsed time t1 to t2, and the end is the interval from elapsed time t2 to t3. Of these three divided intervals, the mean of the normal distribution of the first target time series data TTD1 and the mean of the normal distribution of the second target time series data TTD2 are located in the beginning, the mean of the normal distribution of the third target time series data TTD3 and the mean of the normal distribution of the fourth target time series data TTD4 are located in the middle, and the mean of the normal distribution of the fifth target time series data TTD5 and the mean of the normal distribution of the sixth target time series data TTD6 are located in the end. In other words, in the example shown in Figure 10, the meeting's goal and process are early topics, information exchange and misinformation detection are middle topics, and treatment policies and precautions are final topics.

[0062] Furthermore, in step S13, the target information generation function 192 generates a target radar chart as target information, along with the target time series data TTD, showing the target values ​​of the progress of the examination for each of the multiple topics in each of the three intervals of the target time series data TTD. The target radar chart corresponds to the second radar chart in the first embodiment.

[0063] Figure 11 shows an example of a target radar chart in the target information according to the first embodiment. As shown in Figure 11, the target information generation function 192 generates a first target radar chart TRC1 showing the target value at the beginning of the meeting, a second target radar chart TRC2 showing the target value in the middle of the meeting, and a third target radar chart TRC3 showing the target value at the end of the meeting as target information TF. The first target radar chart TRC1, the second target radar chart TRC2, and the third target radar chart TRC3 each output the progress of discussion for each of the multiple topics on a hexagon, and each of the multiple topics output on the hexagon is associated with each of the vertices of the hexagon and is placed at each of the vertices of the hexagon in the order of discussion in the meeting. In the example shown in Figure 11, the first target radar chart TRC1, the second target radar chart TRC2, and the third target radar chart TRC3 each have multiple topics associated with each vertex of a hexagon: the topic TP1 of the meeting's GOAL, the topic TP2 of the process, the topic TP3 of information exchange, the topic TP4 of misinformation detection, the topic TP5 of treatment strategy, and the topic TP6 of precautions. Furthermore, in accordance with the order of discussion in the meeting, the topics are arranged clockwise from the meeting's GOAL at each vertex of the hexagon: process, information exchange, misinformation detection, treatment strategy, and precautions.

[0064] Furthermore, the target information generation function 192 calculates the target value in the target radar chart TRC by scoring the target time series data TTD with discrete values ​​according to the utterance score. As a result, as shown in Figure 11, the target radar chart TRC includes target value data TD that indicates the target value.

[0065] The first target radar chart TRC1 is a radar chart that shows the target values ​​for the progress of discussions on multiple topics at elapsed time t1. In the first target time series data TTD1 and the second target time series data TTD2 at the beginning of the meeting shown in Figure 10, the utterance scores for the meeting's GOAL and process are high. Therefore, in the first target value data TD1 of the first target radar chart TRC1 shown in Figure 11, the target values ​​for the meeting's GOAL and process are high. The second target radar chart TRC2 is a radar chart that shows the target values ​​for the progress of discussions on multiple topics at elapsed time t2. In the third target time series data TTD3 and the fourth target time series data TTD4 at the middle of the meeting shown in Figure 10, the utterance scores for information exchange and misinformation detection are high. Therefore, in the second target value data TD2 of the second target radar chart TRC2 shown in Figure 11, the target values ​​for information exchange and misinformation detection are higher than those for the first target radar chart TRC1. Furthermore, the third objective radar chart TRC3 is a radar chart that shows the target values ​​for the progress of discussions on multiple topics at elapsed time t3. In the fifth objective time series data TTD5 and the sixth objective time series data TTD6 at the end of the meeting shown in Figure 10, the speaking scores for treatment policies and precautions are high. Therefore, in the third objective value data TD3 of the third objective radar chart TRC3 shown in Figure 10, the target values ​​for information exchange and misinformation detection are higher compared to the second objective radar chart TRC2.

[0066] Next, as shown in Figure 9, the action log data acquisition function 193 in the processing circuit 19 of the conference support device 10 acquires action log data (step S15). Specifically, the action log data acquisition function 193 acquires action log data from the action log database 173.

[0067] Next, as shown in Figure 9, the prompt generation function 194 in the processing circuit 19 of the conference support device 10 generates the first prompt P1 (step S17). Specifically, the prompt generation function 194 generates the first prompt P1 based on the activity log data and multiple topics.

[0068] Figure 12 shows an example of a first prompt P1 according to the first embodiment. As shown in Figure 12, in the example shown in Figure 12, the first prompt P1 includes information as instructions, such as "Please answer with the topic that applies from the following." The information regarding topics includes, for example, the meeting's goal, process, information exchange, misinformation detection, treatment plan, and precautions. The prompt generation function 194 also substitutes information regarding content, for example, by converting some or all of the behavior log data acquired in step S15 into text data.

[0069] Next, as shown in Figure 9, the evaluation result acquisition function 195 in the processing circuit 19 of the conference support device 10 inputs the first prompt P1 to the large-scale language model 331 (step S19). Specifically, the evaluation result acquisition function 195 transmits the first prompt P1 generated by the prompt generation function 194 to the server device 30. The information acquisition function 351 of the server device 30 then receives the first prompt P1 transmitted from the conference support device 10. Subsequently, the evaluation function 352 inputs the first prompt P1 to the large-scale language model 331.

[0070] Next, as shown in Figure 9, the evaluation result acquisition function 195 acquires the evaluation result (step S21). Specifically, first, the evaluation function 352 of the server device 30 acquires the evaluation result ER from the large-scale language model 331. Then, the information output function 354 transmits the evaluation result ER to the conference support device 10. Subsequently, the evaluation result acquisition function 195 acquires the evaluation result ER by receiving the evaluation result ER.

[0071] Figure 13 is a diagram showing an example of evaluation results according to the first embodiment. In the example shown in Figure 13, the evaluation result acquisition function 195 acquires an evaluation radar chart ERC1 as the evaluation result ER, which shows the evaluation results of the progress of the consideration of each of the multiple items. This evaluation radar chart ERC1 is a radar chart that shows the evaluation results of the progress of the consideration of each of the multiple items at the elapsed time t2. The evaluation radar chart ERC1 outputs the progress of the consideration of each of the multiple topics on a hexagon, and each of the multiple topics output on the hexagon is associated with each of the vertices on the hexagon and is placed at each of the vertices of the hexagon in accordance with the order of consideration in the meeting. In the example shown in Figure 13, the evaluation radar chart ERC1 associates multiple topics—topic TP1 (meeting goal), topic TP2 (process), topic TP3 (information exchange), topic TP4 (misinformation detection), topic TP5 (treatment plan), and topic TP6 (precautions)—with each of them corresponding to a vertex of a hexagon. Furthermore, in accordance with the order of discussion during the meeting, the topics are arranged clockwise from the meeting goal: process, information exchange, misinformation detection, treatment plan, and precautions, each corresponding to a vertex of the hexagon.

[0072] Furthermore, the evaluation radar chart ERC1 includes evaluation value data ERD1, which shows the evaluation value of the progress of consideration for each of the multiple items. The evaluation value for each of the multiple topics included in this evaluation value data ERD is calculated based on the behavior log data using the large-scale language model 331. Specifically, the large-scale language model 331 generates the relevant topic information based on the behavior log data, which is information about the content included in the first prompt P1, and calculates a speech score based on the behavior log data based on the generated relevant topic information. Then, the large-scale language model 331 calculates the evaluation value in the evaluation radar chart ERC1 by scoring it with discrete values ​​according to the calculated speech score based on the behavior log data. In other words, the method for calculating the evaluation value and the target value is the same. The evaluation radar chart ERC1 corresponds to the first radar chart in this embodiment.

[0073] Next, as shown in Figure 9, the output function 196 in the processing circuit 19 of the conference support device 10 outputs the evaluation result ER to the topic determination function 197 (step S23). Specifically, the output function 196 outputs the evaluation result ER acquired in step S21 to the topic determination function 197 along with the target information generated in step S13.

[0074] Next, as shown in Figure 9, the discussion topic determination function 197 in the processing circuit 19 of the conference support device 10 determines the discussion topic (step S25). Specifically, the discussion topic determination function 197 determines the discussion topic based on the target information TF generated in step S13 and the evaluation result ER acquired in step S21. More specifically, the discussion topic determination function 197 determines the discussion topic based on the target value included in the target information TF and the evaluation value included in the evaluation result ER, such that the evaluation value approaches the target value.

[0075] Figure 14 is a diagram illustrating the sequence of steps for determining discussion topics by the discussion topic determination function 197 according to the first embodiment. The evaluation result ER shown in Figure 14 is an evaluation radar chart ERC1 showing the progress of discussion of multiple topics in the meeting at elapsed time t2, the target information TF is a radar chart showing the target value of the progress of discussion for each of the multiple topics at elapsed time t3, and the third target radar chart TRC3 is a radar chart showing the target value of the progress of discussion for each of the multiple topics at elapsed time t3. In the example shown in Figure 14, the discussion topic determination function 197 generates a discussion result SR by comparing the evaluation radar chart ERC1 and the third target radar chart TRC3, and determines topics where there is a difference between the target value and the evaluation value as discussion topics. Specifically, the discussion topic determination function 197 determines topics where the evaluation value is small compared to the target value by comparing the evaluation radar chart ERC1 and the third target radar chart TRC3 as discussion topics. In the example shown in Figure 14, the discussion topic determination function 197 determines that treatment policy and precautions are the topics to be discussed in the meeting.

[0076] Next, as shown in Figure 9, the prompt generation function 194 generates a second prompt P2 (step S27). Specifically, the prompt generation function 194 generates the second prompt P2 based on the discussion topic determined in step S25 and the action log data acquired in step S15.

[0077] Figure 15 shows an example of a second prompt P2 according to the first embodiment. In the example shown in Figure 15, the second prompt P2 includes information as instructions, such as "Please answer with the topic that applies from the following." In the example shown in Figure 15, the second prompt P2 includes information as information regarding the topic to be considered and the ratio of the evaluation value to the target value, such as precautions (40%) and treatment plan (30%). The prompt generation function 194 also substitutes the conversation content of the meeting participants from the behavior log data acquired in step S15 as text data as information regarding the content. In the example shown in Figure 15, the prompt generation function 194 substitutes the conversation content of the meeting participants from the behavior log data acquired in step S15 as text data as information regarding the content, but is not limited to this. That is, the information regarding the content is arbitrary and may include information obtained by converting part or all of the behavior log data into text data.

[0078] Next, as shown in Figure 9, the support information acquisition function 198 in the processing circuit 19 of the conference support device 10 inputs the second prompt P2 to the large-scale language model 331 (step S29). Specifically, the support information acquisition function 198 transmits the second prompt P2 generated by the prompt generation function 194 to the server device 30. The support information generation function 353 of the server device 30 then receives the second prompt P2 transmitted from the conference support device 10. Subsequently, the support information generation function 353 inputs the second prompt P2 to the large-scale language model 331.

[0079] Next, as shown in Figure 9, the support information acquisition function 198 acquires support information SF (step S31). Specifically, first, the support information generation function 353 of the server device 30 acquires support information SF from the large-scale language model 331. Then, the information output function 354 transmits the support information SF to the conference support device 10. Subsequently, the support information acquisition function 198 acquires the support information SF by receiving it.

[0080] Next, as shown in Figure 9, the output function 196 outputs support information SF (step S33). Specifically, the output function 196 outputs the support information SF acquired in step S31 via the output interface 13.

[0081] Figure 16 shows an example of support information SF according to the first embodiment. As shown in Figure 16, the output function 196 outputs information as support information SF, which provides advice to the participant to consider the topic of discussion determined in step S25. Specifically, in the example shown in Figure 16, the output function 196 outputs to the display of the output interface 13 information SF1, which provides advice to the participant to consider precautions, stating, "Since it is a circumferential lesion, it may be safer to have both radiation therapy before surgery rather than surgery," and information SF2, which provides advice to the participant to consider treatment strategies, stating, "How about conducting an assessment of the effectiveness of chemotherapy before surgery?"

[0082] Next, as shown in Figure 9, the activity log data acquisition function 193 determines whether the meeting has ended or not (step S35). Specifically, the activity log data acquisition function 193 determines whether the meeting has ended or not by determining whether the activity log data for the meeting has been updated or not.

[0083] Then, in step S35, if the meeting has not ended (step S35: No), the activity log data acquisition function 193 determines whether a predetermined time has elapsed (step S37). Specifically, in step S15, the activity log data acquisition function 193 determines whether a predetermined time has elapsed since acquiring the activity log data. Then, in step S37, if the predetermined time has not elapsed (step S37: No), the process in step S37 is repeated and the function waits until the predetermined time has elapsed.

[0084] On the other hand, in step S37, if a predetermined time has elapsed (step S37: Yes), the activity log data acquisition function 193 acquires the activity log data (step S15). In other words, the conference support device 10 repeatedly executes the process from step S15 until the conference ends.

[0085] On the other hand, if the meeting ends in step S35 (step S35: No), the support information output process according to this embodiment is terminated.

[0086] As described above, the conference support device 10 in the conference support system 1 according to the first embodiment acquires patient information, generates target information TF based on the patient information, acquires behavior log data, inputs the behavior log data into the large-scale language model 331, acquires evaluation results ER from the large-scale language model 331, determines discussion topics based on the target information TF and evaluation results ER, inputs the discussion topics and behavior log data into the large-scale language model 331, acquires support information from the large-scale language model 331, and outputs support information. This facilitates discussion among medical professionals participating in the conference and reduces omissions and oversights in the discussion.

[0087] Furthermore, the conference support device 10 according to the first embodiment determines the topics to be discussed based on the evaluation result ER and the target information TF. Since the topics to be discussed are determined according to the elapsed time, support information corresponding to the elapsed time can be output, thereby promoting and streamlining the discussion of multiple topics in the conference.

[0088] JPEG2026132710000002.jpg63166

[0089] [Variation 1] In the conference support device 10 of the conference support system 1 according to the first embodiment described above, the patient information PF includes the patient's case and generates target information TF related to the patient's case, but it is not limited to this. In modification 1, the patient information PF may include information other than the patient's case along with the patient's case, and the conference support device 10 may generate target information TF based on the patient information PF. The following describes the parts that differ from the first embodiment described above.

[0090] Figure 17 is a diagram showing an example of target information when the length of hospital stay is short in Modification Example 1, and corresponds to Figure 11. Figure 18 is a diagram showing an example of target information when the length of hospital stay is long in Modification Example 1, and corresponds to Figure 11. In the examples shown in Figures 17 and 18, the target information generation function 192 inputs information on the length of hospital stay along with the patient case information PF into the topic analysis model 177, and obtains the target information TFa and TFb output from the topic analysis model 177. In other words, the target information TFa and TFb in this modification shown in Figures 17 and 18 includes target values ​​for multiple topics corresponding to the length of hospital stay, along with the patient case information.

[0091] In the example shown in Figure 17, the target information TFa for patients with a short hospital stay is shown. As shown in Figure 17, the target information TFa includes the first target radar chart TRC1a, which is a radar chart showing the target value of the progress of discussion for each topic at the beginning of the meeting, elapsed time t1; the second target radar chart TRC2a, which is a radar chart showing the target value of the progress of discussion for each topic at the middle of the meeting, elapsed time t2; and the third target radar chart TRC3a, which is a radar chart showing the target value of the progress of discussion for each topic at the end of the meeting, elapsed time t3. In addition, in the example shown in Figure 18, the target information TFb for patients with a long hospital stay is shown. As shown in Figure 18, the target information TFb includes a first target radar chart TRC1b, which is a radar chart showing the target values ​​of the progress of the examination of multiple topics during the elapsed time t1, i.e., in the initial stages; a second target radar chart TRC2b, which is a radar chart showing the target values ​​of the progress of the examination of multiple topics during the elapsed time t2, i.e., in the middle stages; and a third target radar chart TRC3b, which is a radar chart showing the target values ​​of the progress of the examination of multiple topics during the elapsed time t3, i.e., in the final stages.

[0092] The first target radar chart TRC1a shown in Figure 17 and the first target radar chart TRC1b shown in Figure 18 are identical. This indicates that the target values ​​for each of the multiple topics in the early stages of the conference do not change even if the length of hospital stay differs.

[0093] On the other hand, comparing the second target radar chart TRC2a shown in Figure 17 with the second target radar chart TRC2b shown in Figure 18, the target value for information exchange in TRC2a is higher than that in TRC2b shown in Figure 18. This indicates that, due to the short length of hospital stay, it is desirable for each participant in the meeting to exchange information in the middle of the meeting.

[0094] Furthermore, comparing the third objective radar chart TRC3a shown in Figure 17 with the third objective radar chart TRC3b shown in Figure 18, the target value for precautions is higher in TRC3a than in TRC3b. This indicates that, due to the shorter length of hospital stay, it is desirable to share precautions among the conference participants. On the other hand, the target value for treatment strategy is higher in TRC3b than in TRC3a. This indicates that, due to the longer length of hospital stay, precautions are shared among the conference participants, and therefore it is desirable to decide on the treatment strategy towards the end of the conference.

[0095] As described above, in the conference support device 10 according to Modification 1, the target information generation function 192 can obtain target information TF corresponding to the length of hospital stay, which is output from the topic analysis model 177, by inputting patient information PF, which includes the patient's length of hospital stay, along with the patient's case, to the topic analysis model 177. This facilitates medical professionals in considering the patient's condition and reduces omissions in their considerations.

[0096] In the above-described modification 1, the patient information PF input to the topic analysis model 177 includes information on the patient's length of hospital stay along with the patient's case. However, the information included in the patient information input to the topic analysis model 177 is not limited to information on the patient's length of hospital stay. That is, the information included in the patient information PF input to the topic analysis model 177 is arbitrary, and the patient information may include, for example, information such as medical history, treatment progress, and attribute data. In this case, the topic analysis model 177 outputs target information TF according to the information such as medical history, treatment progress, and attribute data.

[0097] [Variation 2] In the conference support device 10 of the conference support system 1 according to the first embodiment described above, the topic selection function 197 is configured to select topics with a difference between the target value and the evaluation value as topics for consideration, but it is not limited to this. In Modification 2, the topic selection function 197 may use a reinforcement learning method to select topics for consideration. The following describes the parts that differ from the first embodiment described above.

[0098] Figure 19 is a diagram showing an example of the configuration of the conference support device 10 according to Modification 2, and corresponds to Figure 3. As shown in Figure 19, the configuration of the memory circuit and the function of the discussion topic determination function differ from those of the first embodiment, so in this embodiment, they are referred to as the memory circuit 17a and the discussion topic determination function 197a. Note that the configuration and functions other than the memory circuit 17a and the discussion topic determination function 197a are the same as those of the first embodiment described above, so their explanation is omitted.

[0099] The memory circuit 17a in this embodiment differs from the memory circuit 17 of the conference support device 10 according to the first embodiment in that it includes a discussion topic selection model 179. The configuration other than the discussion topic selection model 179 is the same as that of the memory circuit 17 of the first embodiment described above, so a description is omitted.

[0100] The topic selection model 179 is a trained model that takes the evaluation result ER as input data and outputs the probability of each of the multiple topics included in the evaluation result ER. The probability of each of the multiple topics output by the topic selection model 179 indicates which topic should be considered next among the multiple topics. For example, among the probabilities of each of the multiple topics, the topic with the highest probability corresponds to the topic that should be considered next. This topic selection model 179 is trained using reinforcement learning. This topic selection model 179 is constructed using, for example, a neural network. Hereafter, this topic selection model 179 will also be called an agent.

[0101] The discussion topic determination function 197a according to this embodiment uses the discussion topic selection model 179 to determine which of several discussion topics will be discussed in the meeting.

[0102] Figure 20 is a diagram illustrating the sequence of steps for generating the topic selection model 179 in modified example 2. As shown in Figure 20, first, the prompt generation function 194 obtains meeting log data from the meeting log database 171. Next, the prompt generation function 194 generates a first prompt P1 based on the meeting log data. Specifically, the prompt generation function 194 generates a first prompt P1 based on the meeting log data that includes multiple topics and audio data and / or document data up to t=0. Next, the evaluation result acquisition function 195 inputs the first prompt P1, which includes multiple topics and audio data and / or document data up to t=0, to the large-scale language model 331 of the server device 30, thereby acquiring the evaluation result ER at t=0 output from the large-scale language model as the state in reinforcement learning.

[0103] Next, the discussion topic determination function 197a acquires the topic to which the support information SF belongs for the evaluation result ER acquired at t=0 as an action in reinforcement learning. Next, the discussion topic determination function 197a determines the reward by determining whether or not the discussion was stimulated as a result of the output of the support information SF. Specifically, the reward is determined by quantifying the typing behavior in the meeting log data after the output of the support information SF, or by determining the tone or speed of the voice, and determining a positive reward if the meeting became lively, and a negative reward if the meeting became stagnant. Then, the discussion topic determination function 197a associates the state, action, and reward and registers them in the association table T1.

[0104] The processing circuit 19 generates the association table T1 by executing the above-described generation process at predetermined time intervals such as t=1, t=2, etc. Then, the topic selection function 197 generates the topic selection model 179 by training the topic selection model 179 using the association table T1.

[0105] Figure 21 is a flowchart illustrating the content of the support information output processing performed in the conference support device 10 according to the modified example 2, and corresponds to Figure 9. Note that the processing of steps S11 to S23 shown in Figure 21 is the same as in the first embodiment.

[0106] As shown in Figure 21, the discussion topic determination function 197a in the processing circuit 19 of the conference support device 10 determines the discussion topic (step S25a). Specifically, the discussion topic determination function 197a determines the discussion topic using the discussion topic selection model 179.

[0107] Figure 22 is a diagram illustrating the sequence of steps for determining a topic of consideration by the topic determination function 197a according to Modification 2. As shown in Figure 22, the topic of consideration determination function 197a inputs the evaluation radar chart ERC1, which is the evaluation result ER acquired in step S21, into the topic of consideration selection model 179, obtains the probability PR for each of the multiple topics included in the evaluation result ER output by the topic of consideration selection model 179, and determines the topic of consideration based on the probability for each of the multiple topics. Note that the processing in steps S27 to S37 from step S25a onward is the same as in the first embodiment.

[0108] As described above, in the conference support device 10 according to the modified example 2, the discussion topic determination function 197a determines the discussion topic using the discussion topic selection model 179, so that the discussion topic can be determined even when it is not possible to generate target information TF.

[0109] In the modified example 2 described above, the discussion topic determination function 197a determines the reward by identifying whether or not the meeting was activated, but it is not limited to this. That is, the method of determining the reward is arbitrary, and if target information TF has been generated, the discussion topic determination function 197a may determine a positive reward if the evaluation result ER is approaching the target information TF.

[0110] [Second Embodiment] In the first embodiment described above, the conference support device 10 acquires evaluation results ER and support information SF regardless of the participant's attributes and provides them to the participant, but it is not limited to this. In the third embodiment, the conference support device 10 may acquire evaluation results ER and support information SF according to the participant's attributes and provide them to the conference participant. The differences from the first embodiment described above will be explained below.

[0111] Figure 23 is a block diagram showing an example of the configuration of the conference support device 10 according to the second embodiment, and corresponds to Figure 3. As shown in Figure 23, the conference support device 10 according to this embodiment is configured by adding a participant information acquisition function 199 to the processing circuit 19 compared to the conference support device 10 according to the first embodiment described above. Furthermore, since the functions of the evaluation result acquisition function and the support information acquisition function differ from those of the first embodiment, in this embodiment they are referred to as the evaluation result acquisition function 195a and the support information acquisition function 198a. The participant information acquisition function 199 corresponds to the participant information acquisition unit in this embodiment. Furthermore, the configuration and functions other than the evaluation result acquisition function 195a, the support information acquisition function 198a, and the participant information acquisition function 199 are the same as those in Figure 3 of the first embodiment described above, so their explanation is omitted.

[0112] In this embodiment, the evaluation result acquisition function 195a inputs participant attribute information along with behavioral information into the large-scale language model 331 and acquires evaluation results from the large-scale language model 331 according to the participant attribute information. Here, participant attribute information is information about the attributes of the participants. Participant attribute information includes, for example, the attributes of participants such as surgeons, oncologists, radiologists, histopathologists, and oncology nurses.

[0113] In this embodiment, the support information acquisition function 198a inputs participant attribute information along with behavioral information into the large-scale language model 331 and acquires support information from the large-scale language model 331 according to the participant attribute information.

[0114] The participant information acquisition function 199 acquires participant information. Here, participant information is information about the participants. This participant information includes at least participant attribute information, such as the participant's name and the number of participants. Specifically, the participant information acquisition function 199 acquires participant information from the user via the input interface 11. Alternatively, the participant information acquisition function 199 may acquire participant information by analyzing voice data and document data stored in the behavior log database 173.

[0115] Figure 24 is a diagram showing an example of input data input to the large-scale language model 331 and output data output from the large-scale language model 331 in the second embodiment, and corresponds to Figure 8. As shown in Figure 24, the large-scale language model 331 takes the first prompt P1a as input data. The first prompt P1a in this embodiment includes participant attribute information as perspective information, along with information regarding instructions, topics, and content. The information regarding instructions, topics, and content is the same as in the first embodiment. Also, as shown in Figure 24, the large-scale language model 331 outputs evaluation result ER corresponding to the participant attribute information as output data.

[0116] Figure 25 is a diagram showing an example of input data input to the large-scale language model 331 and output data output from the large-scale language model 331 in the second embodiment, and corresponds to Figure 8. As shown in Figure 25, the large-scale language model 331 takes the second prompt P2a as input data. The second prompt P2a in this embodiment includes participant attribute information as perspective information, along with information regarding instructions, topics, and content. The information regarding instructions, topics, and content are the same as in the first embodiment. Also, as shown in Figure 25, the large-scale language model 331 outputs support information SFa corresponding to the participant attribute information as output data.

[0117] Figure 26 is a flowchart illustrating the content of the support information output process performed in the conference support device 10 according to the second embodiment, and corresponds to Figure 9. This support information output process is performed, for example, when a conference is started. Note that the process in step S11 shown in Figure 26 is the same as in the first embodiment, so its explanation is omitted.

[0118] As shown in Figure 26, the participant information acquisition function 199 in the processing circuit 19 of the conference support device 10 acquires participant information (step S41). Specifically, the participant information acquisition function 199 acquires participant information, including participant attribute information, from the user via the input interface 11.

[0119] Next, as shown in Figure 26, the target information generation function 192 generates target information TF (step S13a). Specifically, the target information generation function 192 generates target information corresponding to the participant attribute information acquired in step S41. More specifically, the target information generation function 192 uses the topic analysis model 177 to input the participant attribute information along with the patient information PF acquired in step S11 into the topic analysis model 177, thereby generating target information TF corresponding to the participant attribute information. Note that the processing in step S15, which follows step S13a, is the same as in Figure 9, so its explanation is omitted.

[0120] Next, as shown in Figure 26, the prompt generation function 194 generates the first prompt P1a (step S17a). Specifically, the prompt generation function 194 generates the first prompt P1a based on the activity log data, multiple topics, and participant attribute information included in the participant information.

[0121] Figure 27 is a diagram showing an example of the first prompt P1a according to the second embodiment, and corresponds to Figure 12. In the example shown in Figure 27, the first prompt P1a includes information regarding instructions, such as "Please answer with the topic that applies from the following content." The topic information includes, for example, the meeting's GOAL, process, information exchange, misinformation detection, treatment policy, and precautions. The prompt generation function 194 also substitutes information regarding content, such as converting part or all of the behavior log data acquired in step S15 into text data. The prompt generation function 194 also substitutes information regarding perspectives, such as participant attribute information included in the participant information acquired in step S41. Note that the processing in step S19 after step S17a is the same as in Figure 9, so the explanation is omitted.

[0122] Next, as shown in Figure 26, the evaluation result acquisition function 195a acquires the evaluation results (step S21a). Specifically, the evaluation result acquisition function 195a acquires evaluation results based on participant attribute information. Note that the processing in steps S23 and S25, which follow step S21a, is the same as in Figure 9, so the explanation is omitted.

[0123] Next, as shown in Figure 26, the prompt generation function 194 generates a second prompt P2a (step S27a). Specifically, the prompt generation function 194 generates the second prompt P2a based on step S41, using both the discussion topic determined in step S25 and the action log data acquired in step S15.

[0124] Figure 28 shows an example of a second prompt P2a according to the second embodiment. In the example shown in Figure 28, the second prompt P2a includes information regarding instructions, such as "Please answer with the topic that applies from the following content." In addition, in the example shown in Figure 28, the second prompt P2a includes information regarding the topic to be considered and the ratio of the evaluation value to the target value, such as precautions (40%) and treatment plan (30%). The prompt generation function 194 also converts the conversation content of the meeting participants from the behavior log data acquired in step S15 into text data and substitutes it as information regarding the content. In addition, in the example shown in Figure 28, the prompt generation function 194 substitutes participant attribute information included in the participant information acquired in step S41 as information regarding the perspective. Note that the processing from step S27a onward to step S29 is the same as in Figure 9, so the explanation is omitted.

[0125] Next, as shown in Figure 26, the support information acquisition function 198a acquires support information (step S31a). Specifically, the support information generation function 353 of the server device 30 acquires support information SFa corresponding to participant attribute information from the large-scale language model 331. Subsequently, the information output function 354 transmits the support information SFa corresponding to participant attribute information to the conference support device 10. Then, the support information acquisition function 198 acquires the support information SFa corresponding to participant attribute information by receiving the support information SFa corresponding to participant attribute information.

[0126] Next, as shown in Figure 26, the output function 196 outputs the support information SFa (step S33a). Specifically, the output function 196 outputs the support information SFa acquired in step S31 via the output interface 13.

[0127] Figure 29 shows an example of information output to the output interface 13 in the second embodiment. As shown in Figure 29, the output function 196 outputs support information SFa along with the behavioral log data ALD to the display on the output interface 13. In the example shown in Figure 29, the output function 196 outputs text data, which is either audio data or document data converted into text, as the behavioral log data ALD. The display on the output interface 13 shows the contents of the most recent case review meeting included in the behavioral log data ALD, including "I have an opinion on the treatment plan" and "I think surgical resection is the first priority." The output function 196 also outputs three types of support information SFa: the first support information SF1a, the second support information SF2a, and the third support information SF3a. The first support information SF1a, from the surgeon's perspective, provides advice on topics related to precautions, such as "What are the risks to the patient's condition? Is the patient able to withstand surgery?" The second support information SF2a, from the radiologist's perspective, provides advice on treatment plans, such as "How does this compare to radiotherapy?" Furthermore, the third support information SF3a outputs information regarding treatment policies from the perspective of an oncologist, specifically advice on "What about drug therapy?". Note that the processing in steps S35 and S37, which follow step S33a, is the same as in Figure 9, so the explanation is omitted.

[0128] Then, in step S35, if the meeting has ended (step S35: No), the support information output process according to this embodiment is terminated.

[0129] As described above, the conference support device 10 in the conference support system 1 according to the second embodiment acquires patient information PF and participant information, generates target information TF based on participant attribute information in the participant information together with the patient information PF, acquires behavior log data ALD, inputs participant attribute information together with the behavior log data ALD into the large-scale language model 331, acquires evaluation result ER from the large-scale language model 331 according to the participant attribute information, determines discussion topics based on the target information TF and evaluation result ER, inputs the discussion topics and behavior log data into the large-scale language model 331, acquires support information SFa from the large-scale language model 331 according to the participant attribute information, and outputs support information SFa according to the participant attribute information. As a result, support information SFa according to the attributes of the participants can be output, which facilitates discussion among medical professionals who are participants in the conference and reduces omissions and oversights in the discussion.

[0130] In step S33a of the support information output processing according to the second embodiment described above, the output function 196 outputs the behavior log data ALD and the support information SFa to the output interface 13. However, the information output to the output interface 13a in step S33a is not limited to this. The output function 196 may also output information indicating the progress of the review along with the behavior log data and the support information.

[0131] Figure 30 is a diagram showing another example of the information output to the output interface 13 in the second embodiment, and corresponds to Figure 29. As shown in Figure 30, the output function 196 outputs information CF indicating the progress of the study for both the behavior log data ALD and the support information SFb. Note that the behavior log data ALD is the same as in the example shown in Figure 29, so its explanation is omitted. In the example shown in Figure 30, the output function 196 outputs information regarding advice on "What about drug therapy?" which is from the perspective of an oncologist regarding treatment policies, and information regarding the content of the next advice, as support information SFb. In addition, in the example shown in Figure 30, the output function 196 outputs information CF indicating the progress of the study, which includes target time series data and target radar chart, and information superimposed on the target time series data TTD to show the current position.

[0132] Furthermore, in the conference support device 10 of the conference support system 1 according to the second embodiment described above, the evaluation result ER and support information SFa according to the participant attribute information are obtained by including participant attribute information in the first prompt P1a and the second prompt P2a. However, the method of obtaining the evaluation result ER and support information SFa according to the participant attribute information is not limited to this. That is, the method of obtaining the evaluation result ER and support information SFa according to the participant attribute information is arbitrary, and a large-scale language model may be generated for each participant's attribute, and the evaluation result ER and support information SFa according to the participant attribute information may be obtained by inputting the first prompt P1a and the second prompt P2a into the large-scale language model according to the participant's attribute.

[0133] [Third Embodiment] In the first and second embodiments described above, the conference support device 10 provides support information to conference participants by outputting support information to the output interface 13, but it is not limited to this. In the third embodiment, the conference support device 10 may provide evaluation results ER to conference participants by outputting evaluation results to the output interface 13. The differences from the first embodiment described above will be explained below.

[0134] Figure 31 is a block diagram showing an example of the configuration of the conference support device 10 according to the third embodiment, and corresponds to Figure 3. As shown in Figure 31, the configuration of the memory circuit and processing circuit differs from that of the first embodiment, so in this embodiment, they are referred to as memory circuit 17a and processing circuit 19b. Note that the configuration other than memory circuit 17a and processing circuit 19b is the same as in Figure 3, so the explanation is omitted.

[0135] The memory circuit 17a according to this embodiment differs from the memory circuit 17 of the conference support device 10 according to the first embodiment in that it comprises only the action log database 173, as shown in Figure 31. The configuration of the memory circuit 17a other than comprising only the action log database 173 is the same as that of the first embodiment, and therefore its description is omitted. Furthermore, the details of the action log database 173 are the same as those of the action log database 173 shown in Figure 3 above, and therefore their description is omitted.

[0136] The processing circuit 19b according to this embodiment differs from the processing circuit 19 of the conference support device 10 according to the first embodiment in that, as shown in Figure 31, it includes an action log data acquisition function 193, a prompt generation function 194, an evaluation result acquisition function 195, and an output function 196. The configuration of the processing circuit 19b, other than having only the action log data acquisition function 193, the prompt generation function 194, the evaluation result acquisition function 195, and the output function 196, is the same as in the first embodiment and will therefore not be described. Furthermore, the details of the action log data acquisition function 193, the prompt generation function 194, the evaluation result acquisition function 195, and the output function 196 are the same as in the first embodiment described above and will therefore not be described.

[0137] Figure 32 is a block diagram showing an example of the configuration of the server device 30 according to the third embodiment, and corresponds to Figure 6. As shown in Figure 32, the configuration of the processing circuit is different from that of the first embodiment, so in this embodiment, it is referred to as the processing circuit 35a. Note that the configuration other than the processing circuit 35a is the same as that of Figure 6, so its explanation is omitted.

[0138] The processing circuit 35a according to this embodiment differs from the processing circuit 35 of the server device 30 according to the first embodiment in that it includes an information acquisition function 351, an evaluation function 352, and an information output function 354. The configuration of the processing circuit 35a other than the inclusion of the information acquisition function 351, the evaluation function 352, and the information output function 354 is the same as in the first embodiment, so its description is omitted. Furthermore, the details of the information acquisition function 351, the evaluation function 352, and the information output function 354 are the same as in the first embodiment described above, so their description is omitted.

[0139] Figure 33 is a flowchart illustrating the content of the support information output processing performed in the conference support device 10 in the third embodiment, and corresponds to Figure 9. Note that the processing of steps S15 to S21 shown in Figure 33 is the same as in the first embodiment.

[0140] Next, as shown in Figure 33, the output function 196 outputs the evaluation result ER to the output interface 13 (step S23a). Specifically, the output function 196 outputs the evaluation result ER acquired in step S21 to the display of the output interface 13.

[0141] Figure 34 shows an example of an evaluation result ER output to the output interface 13 of the conference support device 10 according to the third embodiment. As shown in Figure 34, the output function 196 outputs the evaluation radar chart ERC1 as the evaluation result ER to the display of the output interface 13. Conference participants can check the progress of discussions on multiple topics by checking the evaluation result ER output to the display of the output interface 13. Note that the processing from step S23a onward to steps S35 to S37 is the same as in the first embodiment.

[0142] As described above, in the conference support device 10 according to the third embodiment, the output function 196 outputs the evaluation result ER to the output interface 13, allowing medical professionals who are participants in the conference to check the evaluation result ER. This enables them to check the progress of the discussion of multiple topics, facilitating the discussion by medical professionals and reducing any omissions or oversights in the discussion.

[0143] [Fourth Embodiment] In the first to third embodiments described above, the large-scale language model 331 was provided within the server device 30, but this is not limited to that. In the fourth embodiment, the large-scale language model 331 is provided within the conference support device 10. The differences from the first embodiment described above will be explained below. In this embodiment, the server device 30 and the network NW may be omitted.

[0144] Figure 35 is a block diagram showing an example of the configuration of the conference support device 10 according to the fourth embodiment, and corresponds to Figure 3. As shown in Figure 35, the conference support device 10 according to the fourth embodiment differs from the conference support device 10 according to the first embodiment in that, in addition to the configuration of the conference support device 10 according to the first embodiment described above, the memory circuit 17 is equipped with a large-scale language model. Note that the communication interface 15 may be omitted. Furthermore, the configuration other than the large-scale language model is the same as that of the first embodiment described above, so its explanation is omitted.

[0145] The large-scale language model 181 is located within the memory circuit 17. Details of the large-scale language model 181 are the same as those of the large-scale language model 331 described above, so their explanation is omitted.

[0146] The operation example of the conference support device 10 according to the fourth embodiment will be described below with reference to Figure 9. First, the processing of steps S11 to S17 is the same as in the first embodiment.

[0147] Next, in step S19, the evaluation result acquisition function 195 in the processing circuit 19 of the conference support device 10 inputs the first prompt P1 to the large-scale language model 181 in the memory circuit 17 of the conference support device 10. Also, in step S21, the evaluation result acquisition function 195 acquires the evaluation result output from the large-scale language model 181. The processing in steps S23 to S27 is the same as in the first embodiment.

[0148] Furthermore, in step S29, the support information acquisition function 198 in the processing circuit 19 of the conference support device 10 inputs the second prompt P2 to the large-scale language model 181. Also, in step S31, the support information acquisition function 198 acquires the support information output from the large-scale language model 181. Note that the processing in steps S33 to S37 is the same as in the first embodiment.

[0149] As described above, the conference support device 10 according to the fourth embodiment can facilitate the review process by medical professionals and reduce omissions and errors in the review, even in so-called on-premise large-scale language models 181. For example, the conference support device 10 according to this embodiment can be used even in locations without a communication environment.

[0150] [Other variations] In the first to fourth embodiments described above, the generative model may be a so-called small language model (SLM) or a multimodal model. A small language model is a language model with fewer parameters than a large language model. A multimodal model is a model trained using combinations of multiple different types of data.

[0151] Furthermore, in the first to fourth embodiments described above, the utterance score is calculated based on the proportion of utterances, but it is also possible to calculate it based on the specificity of the utterances along with the proportion of utterances. Specifically, the conference support device 10 calculates the utterance score using the formula: utterance score = proportion of utterances + specificity of utterances. This specificity of utterances is calculated, for example, by inputting a prompt containing information such as "Please assign a rank to each utterance according to its specificity" along with the action log data into the large-scale language model 331, thereby assigning a score related to the specificity of the utterances. The calculated specificity of utterances is normalized so that the minimum value is 0 and the maximum value is 1. The conference support device 10 may also generate a graph of the utterance scores for each topic at each calculated time point.

[0152] Furthermore, the method for calculating the specificity score of a statement is not limited to using the large-scale language model 331. In other words, the method for calculating the specificity score of a statement is arbitrary, and for example, it may be calculated based on a dictionary in which the score of each word is recorded. This dictionary is, for example, a database (MRC Psycholinguistic Database) in which the specificity score and imagery score of each word are recorded. Here, specificity indicates that it has a clear and tangible reality. Imagery indicates how easy it is to visualize a word.

[0153] Furthermore, in the first to fourth embodiments described above, the output function 196 can also output the target information TF and evaluation result ER along with the support information SF to the output interface 13. By outputting the target information TF and evaluation result ER along with the support information SF, the output function 196 allows healthcare professionals to verify the accuracy of the content of the support information SF.

[0154] Furthermore, in the first to fourth embodiments described above, the target information generation function 192 may generate a mixed distribution as the target time series data TTD in the target information TF, which is a mixture of the normal distributions of multiple topics.

[0155] Furthermore, in the first to fourth embodiments described above, the meeting may be an interview, consultation, medical interview, customer service, meeting, or consultation. If the meeting is a medical interview, the multiple topics may be exploration, convergence, denial, affirmation, result, and cause.

[0156] Furthermore, in the first to fourth embodiments described above, the evaluation value is calculated by the large-scale language model 331 and the evaluation result ER is output, but this is not limited to this. For example, the evaluation function 352 may input a first prompt P1 to the large-scale language model 331 to obtain the relevant topic information of the behavior log data, calculate a speech score based on the relevant topic information, and calculate the evaluation value in the evaluation radar chart ERC by scoring discrete values ​​according to the calculated speech score.

[0157] In the above description, the term "processor" refers to circuits such as a CPU (Central Processing Unit), a GPU (Graphics Processing Unit), an Application Specific Integrated Circuit (ASIC), or a programmable logic device (e.g., a Simple Programmable Logic Device (SPLD), a Complex Programmable Logic Device (CPLD), and a Field Programmable Gate Array (FPGA)). The processor functions by reading and executing programs stored in memory circuits 17 and 33. Alternatively, instead of storing programs in memory circuits 17 and 33, the processor may be configured to directly incorporate programs into its circuitry. In this case, the processor functions by reading and executing the programs incorporated into the circuitry. The processor is not limited to being configured as a single circuit; it may also be configured by combining multiple independent circuits to form a single processor and realize its functions. Furthermore, the multiple components shown in Figures 2 and 6 may be integrated into a single processor to realize its functions.

[0158] According to at least one embodiment described above, it is possible to facilitate consideration by healthcare professionals and reduce omissions or oversights in the consideration process.

[0159] Although several embodiments have been described above, these embodiments are presented only as examples and are not intended to limit the scope of the invention. The novel apparatus and methods described herein can be implemented in a variety of other forms. Furthermore, various omissions, substitutions, and modifications can be made to the embodiments of the apparatus and methods described herein, without departing from the spirit of the invention. The appended claims and equivalents are intended to include such embodiments and modifications that are included in the scope and spirit of the invention. [Explanation of Symbols]

[0160] 1...Conference support system, 10...Conference support device, 11...Input interface, 13, 13a...Output interface, 15...Communication interface, 17, 17a...Memory circuit, 19, 19b...Processing circuit, 30...Server device, 31...Communication interface, 33...Memory circuit, 35, 35a...Processing circuit, 171...Conference log database, 173...Behavior log database, 175...Patient information database, 177...Topic analysis model, 179...Topic selection model for consideration 181...Large-scale language model, 191...Patient information acquisition function, 192...Goal information generation function, 193...Behavioral log data acquisition function, 194...Prompt generation function, 195, 195a...Evaluation result acquisition function, 196...Output function, 197, 197a...Consideration topic determination function, 198, 198a...Support information acquisition function, 199...Participant information acquisition function, 331...Large-scale language model, 351...Information acquisition function, 352...Evaluation function, 353...Support information generation function, 354...Information output function

Claims

1. A behavioral information acquisition unit that acquires behavioral information regarding the actions of participants in a meeting, An evaluation result acquisition unit inputs the aforementioned behavioral information into a generation model and obtains evaluation results from the generation model regarding the progress of consideration of multiple items to be considered in the meeting, An output unit that outputs the evaluation results, A conference support device equipped with the following features.

2. The conference support device according to claim 1, wherein the evaluation result acquisition unit inputs the behavior information to the generation model at predetermined intervals and acquires the evaluation result from the generation model.

3. The conference support device according to claim 1, wherein the evaluation result acquisition unit acquires the evaluation results relating to the progress status for each of the plurality of items.

4. The evaluation result acquisition unit acquires a first radar chart showing the progress of the examination for each of the multiple items as the evaluation result, The output unit outputs the first radar chart to the display unit. The conference support device according to claim 1.

5. The first radar chart displays the progress of the examination for each of the multiple items on a polygon. The conference support device according to claim 4, wherein each of the plurality of items output on the polygon is associated with each of the vertices of the polygon and is arranged at each of the vertices of the polygon in the order of discussion in the conference.

6. The meeting support device according to claim 1, wherein the plurality of items include at least one of items relating to the policy of the meeting, items relating to the discussions in the meeting, and items relating to the decision-making of the meeting.

7. The system further comprises a prompt generation unit that generates prompts for input to the generation model based on the behavioral information and the plurality of items, The conference support device according to claim 1, wherein the evaluation result acquisition unit inputs the prompt to the generation model and acquires the evaluation result from the generation model.

8. The system further comprises a target information generation unit that generates target information regarding the progress of the aforementioned considerations for the aforementioned multiple items, The output unit outputs the target information along with the evaluation results. The conference support device according to claim 1.

9. The aforementioned meeting is a specialist meeting to consider treatment plans for patients. The system further includes a patient information acquisition unit that acquires patient information including at least the patient's case, The conference support device according to claim 8, wherein the target information generation unit generates the target information related to the patient's case included in the patient information.

10. The meeting support device according to claim 9, wherein the target information generation unit generates target time-series data showing the time-series changes in target values ​​of the progress status of the consideration for each of the plurality of items as the target information.

11. The conference support device according to claim 10, wherein the target information generation unit divides the target time series data into a plurality of intervals and generates a second radar chart showing the target value of the progress of the examination for each of the plurality of items in each of the plurality of intervals.

12. The meeting support device according to claim 8, further comprising a unit for determining items to be considered in the meeting from among the plurality of items, based on the target information and the evaluation results.

13. The aforementioned target information includes target values ​​that indicate the progress of the aforementioned examination for each of the aforementioned items, The evaluation results include evaluation values ​​that show the progress of the examination for each of the aforementioned items, The conference support device according to claim 12, wherein the item determination unit determines, among the plurality of items, the item for which there is a difference between the target value and the evaluation value as an item for consideration.

14. The system further includes a support information acquisition unit that inputs the aforementioned consideration items and the aforementioned behavioral information into the generation model and acquires support information from the generation model to support the participants in the meeting, The conference support device according to claim 12, wherein the output unit outputs the support information.

15. The conference support device according to claim 1, wherein the generative model is a large-scale language model.

16. The conference support device according to claim 7, wherein the evaluation result acquisition unit inputs participant attribute information relating to the attributes of the participants, along with the behavioral information, into the generation model, and acquires the evaluation result corresponding to the participant attribute information from the generation model.

17. The conference support device according to claim 14, wherein the support information acquisition unit inputs participant attribute information relating to the attributes of the participants, along with the consideration items and the behavioral information, into the generation model, and acquires the support information corresponding to the participant attribute information from the generation model.

18. A behavioral information acquisition unit that acquires behavioral information regarding the actions of participants in a meeting, An evaluation unit that uses a generative model to evaluate the progress of the discussion on multiple items to be considered in the aforementioned meeting, An evaluation result acquisition unit inputs the aforementioned behavioral information into a generation model and obtains evaluation results from the evaluation unit regarding the progress of consideration on multiple items to be considered in the meeting, An output unit that outputs the evaluation results, A meeting support system equipped with the following features.

19. Steps to obtain behavioral information regarding the behavior of participants in a meeting, The steps include inputting the aforementioned behavioral information into a generative model and obtaining evaluation results from the generative model regarding the progress of consideration on multiple items to be considered in the meeting, The step of outputting the evaluation result, A meeting support method that includes the following features.

20. On the computer, Steps to obtain behavioral information regarding the behavior of participants in a meeting, The steps include inputting the aforementioned behavioral information into a generative model and obtaining evaluation results from the generative model regarding the progress of consideration on multiple items to be considered in the meeting, The step of outputting the evaluation result, A program that executes something.

Images