Human-machine fusion group discussion method and device based on large language model

Abstract

The application provides a kind of method and device for group discussion based on large language model, which is applied to discussion system.The discussion system integrates artificial intelligence (AI) experts, which are constructed by large language model.The method comprises the following steps: monitoring group discussion to obtain the discussion results of human experts and AI experts participating in group discussion on discussion problems; determining the discussion views of human experts and AI experts on discussion problems based on the respective discussion results of human experts and AI experts; and performing decision fusion on the discussion views of human experts and AI experts to obtain the problem decision of discussion problems.The method combines the powerful capacity of model with human wisdom to jointly deal with complex problems in group discussion, improves the efficiency of discussion, overcomes the problem that machine computing capacity and human creativity cannot be fully utilized in traditional group discussion, and makes group discussion better adapt to complex problems to find the most feasible solution for complex problems.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a human-computer integrated group discussion method and apparatus based on a large language model. Background Technology

[0002] Group discussions have always been a key means of driving innovation and solving complex problems. By pooling the wisdom and experience of multiple individuals, they provide diverse perspectives and approaches to problem-solving. In the process of sharing knowledge, experience, and ideas, groups can generate more creative solutions and find the most consensus-based and feasible solutions through thorough discussion and debate.

[0003] However, traditional group discussions often face challenges when dealing with highly complex problems. This is because the complexity of the problem can lead to stalemates and difficulties in the discussion, resulting in incomplete and inaccurate solutions. Furthermore, the increasing number of participants and their diverse roles, along with the diversity of fields involved, further complicates collaborative work. In addition, traditional meeting-style group discussions are limited by time and space, further restricting the possibility of in-depth discussions and potentially reducing participant engagement. Moreover, current discussion software often has limited functionality and struggles to flexibly meet the needs of complex problem discussions, exhibiting limitations in information processing, decision-making time, and participant experience, leading to unsatisfactory problem-solving and participant experience. Summary of the Invention

[0004] This invention provides a human-computer integrated group discussion method and apparatus based on a large-scale language model, which addresses many problems existing in the conference-style group discussion in the prior art. By leveraging the powerful capabilities of large-scale language models, it constructs a human-computer integrated group discussion mode, combining the powerful capabilities of large-scale language models with human intelligence to jointly address complex problems in group discussions. While improving the efficiency of discussions, it optimizes solutions to complex problems, enabling group discussions to better adapt to the needs of solving complex problems and innovative thinking, making them more dynamic and effective.

[0005] This invention provides a human-computer integrated group discussion method based on a large-scale language model, applied to a discussion system. The discussion system integrates artificial intelligence (AI) experts, which are constructed using a large-scale language model. The method includes:

[0006] The group discussion is monitored to obtain the discussion results of the human experts and the AI ​​experts participating in the group discussion on the discussion questions of the group discussion;

[0007] Based on the discussion results of the human experts and the AI ​​experts, the discussion opinions of the human experts and the AI ​​experts on the discussion issue are determined respectively;

[0008] The decision-making process involves fusing the perspectives of the human experts and the AI ​​experts on the research question to obtain the problem decision for the group discussion.

[0009] According to a human-computer fusion group discussion method based on a large language model provided by the present invention, the step of fusing the discussion opinions of the human experts and the AI ​​experts on the discussion problem to obtain the problem decision of the group discussion includes:

[0010] Feature extraction is performed on the discussion viewpoints of the human experts and the AI ​​experts respectively to obtain the semantic features of the discussion viewpoints of the human experts and the AI ​​experts respectively;

[0011] Information is extracted from the discussion viewpoints of the human experts and the AI ​​experts to obtain key information about their respective discussion viewpoints;

[0012] Based on the semantic features and key information of the discussion viewpoints of the human experts and the AI ​​experts, the discussion viewpoints of the human experts and the AI ​​experts on the discussion problem are fused to obtain the problem decision of the discussion problem of the group discussion.

[0013] According to the present invention, a human-computer fusion group discussion method based on a large language model includes a decision fusion process for the discussion opinions of human experts and AI experts on the discussion problem, based on the semantic features and key information of their respective discussion opinions, to obtain a problem decision for the discussion problem corresponding to the group discussion. This process includes:

[0014] Based on the key information of the discussion viewpoints of the human experts and the AI ​​experts, candidate discussion viewpoints are selected from the discussion viewpoints of the human experts and the AI ​​experts.

[0015] Based on the semantic features of each candidate discussion viewpoint, semantic matching is performed on each candidate discussion viewpoint to obtain semantic matching results;

[0016] Based on the semantic matching results, the discussion opinions of the human experts and the AI ​​experts on the research question are fused to obtain the problem decision for the research question.

[0017] According to a human-computer fusion group discussion method based on a large language model provided by the present invention, the step of fusing the discussion opinions of the human experts and the AI ​​experts on the discussion problem based on the semantic matching result to obtain the problem decision of the discussion problem includes:

[0018] Based on the semantic matching results, the candidate discussion viewpoints are fused to obtain a fused discussion viewpoint;

[0019] Based on the integrated discussion viewpoints, a decision is generated for the discussion question, resulting in a problem decision for the discussion question.

[0020] According to the present invention, a human-computer integrated group discussion method based on a large language model is provided, wherein the method involves fusing the candidate discussion viewpoints based on the semantic matching results to obtain a fused discussion viewpoint, including:

[0021] Based on the semantic matching results, the candidate discussion viewpoints are merged to obtain the first category of discussion viewpoints;

[0022] The topic is extracted based on the discussion opinions of the human experts and the AI ​​experts to obtain the topic and topic intensity of the discussion opinions of the human experts and the AI ​​experts;

[0023] Based on the topic and intensity of the viewpoints, the viewpoints of the human experts and the AI ​​experts are merged to obtain the second type of viewpoints.

[0024] The first type of discussion viewpoints and the second type of discussion viewpoints are merged to obtain a merged discussion viewpoint.

[0025] According to the present invention, a human-computer integrated group discussion method based on a large language model is provided, wherein the AI ​​expert and the human expert work together in the group discussion to jointly discuss the discussion problem and obtain the discussion result;

[0026] The discussion results include at least one of the following: the record of speeches made during the group discussion, expert actions, the breakdown record of the discussion questions, and the opinion vote on the discussion questions.

[0027] This invention also provides a human-computer integrated group discussion device based on a large-scale language model, applied to a discussion system. The discussion system integrates artificial intelligence (AI) experts, which are constructed using a large-scale language model. The device includes:

[0028] The discussion monitoring unit is used to monitor the group discussion and obtain the discussion results of the human experts and the AI ​​experts participating in the group discussion on the discussion questions of the group discussion.

[0029] The opinion determination unit is used to determine the opinions of the human experts and the AI ​​experts on the discussion issue based on the discussion results of the human experts and the AI ​​experts, respectively.

[0030] The decision fusion unit is used to fuse the discussion opinions of the human experts and the AI ​​experts on the discussion problem to obtain the problem decision of the group discussion.

[0031] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the human-computer fusion group discussion method based on a large language model as described above.

[0032] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the human-computer fusion group discussion method based on a large language model as described above.

[0033] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the human-computer fusion group discussion method based on a large language model as described above.

[0034] The present invention provides a human-machine fusion group discussion method and apparatus based on a large-scale language model. It monitors group discussions, obtaining the discussion results of human and AI experts on the discussion problem. Based on these results, it determines the viewpoints of both experts on the problem. Finally, it fuses these viewpoints to arrive at a problem decision for the group discussion. This fully utilizes the model's computational power and human creativity, combining the powerful capabilities of a large-scale language model with human intelligence to jointly address complex problems in group discussions. This improves the efficiency and collaboration of the discussions, overcoming the limitations of traditional group discussions in fully leveraging machine computing power and human creativity. It enables group discussions to better adapt to the needs of solving complex problems and fostering innovative thinking, finding the most feasible solutions to complex problems, and demonstrating flexibility and practicality. Attached Figure Description

[0035] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0036] Figure 1 This is a flowchart illustrating the human-computer integrated group discussion method based on a large language model provided by the present invention.

[0037] Figure 2This is the overall framework diagram of the group discussion process provided by the present invention;

[0038] Figure 3 This is a flowchart illustrating the decision fusion process provided by the present invention;

[0039] Figure 4 This is a schematic diagram of the human-computer integrated group discussion device based on a large language model provided by the present invention;

[0040] Figure 5 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0041] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0042] In the field of group discussions, traditional meeting formats may be limited by time and space, restricting the possibility of in-depth discussions and reducing participant enthusiasm. It is particularly noteworthy that current discussion software often has limited functionality, making it difficult to flexibly address the needs of complex problem discussions. Its lack of personalization and intelligence results in a series of limitations in information processing, decision-making time, and participant experience.

[0043] As large language models have developed, their enormous potential in the field of natural language processing has become increasingly recognized. However, despite their excellent performance in natural language processing, their application in group discussions remains relatively limited. Furthermore, traditional discussion software / tools often fail to fully utilize the capabilities of large language models, resulting in limitations in information processing, delays in decision-making, and a poor experience for participants.

[0044] To address this, this invention provides a human-computer integrated group discussion method based on a large-scale language model. The aim is to construct a human-computer integrated group discussion approach that fully utilizes the model's computational power and human creativity. It combines the powerful capabilities of a large-scale language model with human intelligence to jointly address complex problems in group discussions, thereby improving the efficiency and collaboration of the discussions. This overcomes the limitations of traditional group discussions in fully utilizing machine computational power and human creativity, enabling group discussions to better adapt to the needs of solving complex problems and fostering innovative thinking. It finds the most feasible solutions to complex problems, demonstrating flexibility and practicality.

[0045] Figure 1This is a flowchart illustrating the human-computer integrated group discussion method based on a large language model provided by the present invention, as shown below. Figure 1 As shown, this method is applied to a discussion system that integrates AI (Artificial Intelligence) experts. These AI experts are constructed using a large language model. The method includes:

[0046] Step 110: Monitor the group discussion and obtain the discussion results of the human experts and AI experts participating in the group discussion on the discussion questions.

[0047] Step 120: Based on the discussion results of human experts and AI experts, determine the discussion viewpoints of human experts and AI experts on the discussion issue respectively;

[0048] Step 130: Integrate the opinions of human experts and AI experts on the research question to obtain the problem decision for the group discussion.

[0049] Specifically, considering that traditional group discussions cannot efficiently and effectively utilize the massive amounts of data in group discussions and cannot fully leverage machine computing power and human creativity, this invention proposes a human-machine integrated group discussion method based on a large-scale language model. The large-scale language model processes the text in the group discussion and provides assistance and control to human experts at each stage of the group discussion.

[0050] Figure 2 This is a general framework diagram of the group discussion process provided by the present invention, such as... Figure 2 As shown, the entire discussion process is divided into three stages: the discussion preparation stage, the multi-round discussion stage, and the consensus assessment stage.

[0051] During the preparation phase of the seminar, administrators customize processes, invite human experts, upload materials, and set permissions. In the multi-round seminar phase, human experts can utilize various support tools for in-depth discussions. The seminar system provides efficient support using intelligent tools and algorithms, and summarizes seminar data in real time for optimization. Simultaneously, the seminar system incorporates AI experts as seminar participants, collaborating with human experts to combine the model's computational power with human creativity, achieving effective human-machine integration. In the consensus-building and evaluation phase, the seminar system collects consensus, integrates discussion viewpoints, and provides voting and scoring functions, while ensuring the transparency and traceability of the seminar process. During this process, the seminar system can leverage the powerful capabilities of AI experts to integrate and summarize various discussion viewpoints from the group, analyze and compare different viewpoints, identify their consensus points and core opinions, thereby promoting consensus-building.

[0052] Specifically, the preparation phase is the starting point of the entire seminar process. Its purpose is to fully prepare for the group seminar, with specific objectives including customizing and creating the seminar space, managing human experts, uploading and downloading necessary materials, and establishing the seminar questions, objectives, and expected outcomes through the facilitator's presentation and material presentation. This helps provide participating experts with a comprehensive understanding and a clear direction for the seminar. First, administrators are appointed for the entire group seminar. These administrators then determine the seminar questions, objectives, and expected outcomes based on the seminar materials, and customize the seminar's workflow, human experts, and their tasks. Next, the seminar space is created, human experts are invited to join, seminar materials are uploaded, a facilitator's seat is set, and operational permissions are assigned to participating human experts based on their tasks. Finally, the facilitator delivers a presentation on the seminar topic and displays relevant data.

[0053] During this phase, participating human experts can access relevant interfaces through the discussion system to view the content of the moderator's presentation and related data on the discussion issues, such as relevant research, statistical data, and case studies, in order to better understand the background and purpose of the discussion. The discussion system provides an interactive interface to allow human experts to interact with the moderator, clarifying doubts through questions and answers, and ensuring that the goals of the group discussion are accurately defined.

[0054] It is worth noting that after the preparation for the seminar is completed, when entering the multi-round seminar phase, managers can break down the seminar questions into multiple discussion topics, and conduct discussions on each topic. This process is called a sub-seminar, such as... Figure 2 The discussion is divided into sub-discussions 1 through n. During these multiple rounds, human experts can freely share their personal perspectives, experiences, and ideas, and respond to other members, creating a dynamic discussion atmosphere. This encourages diverse thinking and information exchange among participants, fully leveraging collective wisdom. While human experts engage in in-depth discussions, the system utilizes group thinking control tools to stimulate collective thinking and monitor, identify, and eliminate negative thinking in real time, promoting the emergence of innovative viewpoints and ensuring a consistently positive and collaborative atmosphere, thereby improving discussion efficiency. Crucially, AI experts built on large-scale language models will participate in the discussions, exploring research problems alongside human experts, engaging in intellectual exchange, answering their questions, and proposing constructive viewpoints and solutions, leading the group discussion to a more comprehensive and profound level. To support this in-depth discussion, human experts can freely access relevant data from the group discussion. The system provides interfaces to various discussion support tools, such as literature review, data extrapolation, and simulation, enabling experts to analyze problems more comprehensively and efficiently.

[0055] Furthermore, during multiple rounds of discussions, the system also records all the experts' speeches, viewpoints, and opinions, and aggregates this information into a rich data source, which is then categorized and stored in the discussion database. This provides diverse data and reference information for subsequent discussions, and offers human experts a wealth of interactive tools, making the discussions no longer a one-way information transmission, but a process of multi-participation and open communication, which helps to uncover more innovative thinking and comprehensive problem-solving.

[0056] Specifically, during the multi-round discussion phases, human experts can utilize various discussion support tools, such as audio and video, whiteboard operation, data uploading and sharing, and anonymous speaking, to support problem-solving discussions. When human experts engage in in-depth discussions, the discussion system uses intelligent tools and algorithms to collaborate with them for efficient discussions. For example, group thinking control tools identify and adjust undesirable thinking patterns (rigid thinking, dependent thinking, etc.) among human experts; topic evolution tools help human experts extract and abstract topics in real time; an intelligent question-answering engine provides intelligent question-answering services; data inference tools perform fast and accurate data calculations according to the requirements of human experts; and literature retrieval and recommendation tools provide efficient and personalized information support for discussion experts. During the multi-round discussion, expert discussion data (speech records, expert actions, etc.) is stored in the database in real time, providing data for the underlying algorithms of the discussion system to help improve system performance. During the discussion, the moderator can further break down the topics of sub-discussions based on the discussion progress and establish breakout sessions for further discussion. The discussion system will summarize the data of the discussion process in real time (voting results, problem breakdown results, expert speeches, shared whiteboard operations, etc.). The moderator can confirm the end of the discussion or start the next round of discussion based on the discussion situation. Through multiple rounds of discussion and repeated iterations, the discussion will become more and more perfect.

[0057] Finally, in the consensus-building and evaluation phase, the discussion system will gather the viewpoints of both human and AI experts, aiming to reach a consensus. During this process, through consensus collection and viewpoint integration, the discussion system can clearly present differing opinions. To further determine the consensus and evaluate the quality of the discussion results, the system provides a voting and scoring function. Human experts can vote and score different opinions anonymously or non-anonymously based on their own opinions and understanding. The discussion system can then perform statistical analysis based on the voting results to identify widely accepted viewpoints, thereby helping the moderator and human experts better understand the degree and direction of the consensus.

[0058] In addition, to ensure the transparency and traceability of the discussion process, the discussion system provides a viewpoint tracing function, allowing human experts to view all speaking records and decision-making paths at any time during the discussion process, understand the source and evolution of each viewpoint, and comprehend the formation of consensus.

[0059] Based on this, in this embodiment of the invention, in order to realize human-machine integrated group discussion, the discussion system can monitor the group discussion process after the discussion begins, so as to obtain the speeches, opinions and suggestions of each expert participating in the group discussion, thereby forming two categories of discussion results: the discussion results of the human experts participating in the group discussion, which includes the speech records, operations, opinions, etc. of each human expert participating in the group discussion, and the discussion results of the AI ​​experts, which also include the AI ​​experts' leading speeches in the group discussion, their responses to the questions of human experts, and their opinions and suggestions in the discussion.

[0060] It should be noted here that in the group discussion, AI experts and human experts collaborate to jointly explore the discussion questions and obtain discussion results. The discussion results include the experts' speaking records and actions in the group discussion, the moderator's breakdown of the discussion questions, and the voting results of human experts on the opinions and suggestions formed by the group experts on the discussion questions. That is, the discussion results include any one or more of the following: speaking records in the group discussion, expert actions, breakdown records of the discussion questions, and opinion votes on the discussion questions.

[0061] The results of the discussion can be obtained through multiple channels. For example, the experts' speeches can be obtained from the text communication recorded in the interactive interface, the experts' operations and their corresponding times can be obtained through system logs and running files, the experts' votes and scores on different opinions can be obtained through questionnaires and voting, and the text records and experts' operations can be obtained through the shared whiteboard.

[0062] After obtaining the discussion results, in this embodiment of the invention, these results can be further analyzed and processed to extract discussion viewpoints representing human opinions and AI expert opinions from the massive and disorganized discussion data. That is, the discussion system can use multi-dimensional methods, such as text extraction, information summarization, key information extraction, viewpoint summarization, relevance fusion, and irrelevant filtering, to extract and summarize two representative categories of viewpoints from the discussion data containing massive amounts of discussion information: the discussion viewpoints of human experts and AI experts on the discussion issue.

[0063] It should be noted here that each category of discussion viewpoints does not contain only one viewpoint. Because the discussion process of highly complex issues is usually complicated and lengthy, the information generated, as well as the opinions and suggestions therein, is very tedious and complex. Therefore, the viewpoints extracted are usually not just one type. The process of extracting viewpoints mentioned above is not to merge different opinions, but to extract viewpoints from a massive amount of opinions and suggestions to obtain a variety of representative discussion viewpoints.

[0064] Then, the decision-making process can be integrated based on the discussion views of human experts and AI experts to obtain the problem decision of the group discussion. This process is the consensus-building and evaluation process. The discussion system can make decisions and evaluate the discussion problem based on the discussion views of human experts and AI experts to determine the final decision, that is, the problem decision of the discussion problem. This problem decision can be the optimal solution to the discussion problem, the best interpretation of the discussion problem, or a feasible implementation plan.

[0065] Specifically, the discussion system first integrates the viewpoints of human experts and AI experts on the discussion issue. That is, through intelligent algorithms, it compares and analyzes different viewpoints, identifies similarities, and extracts common core viewpoints, thus forming a more representative and comprehensive perspective. Then, based on this comprehensive perspective, a decision is generated, leading to an understanding of the discussion issue, solutions, and implementation paths, thereby arriving at a problem-solving decision. By intelligently analyzing experts' statements and behaviors, consensus points and key viewpoints are identified, thereby facilitating the achievement of consensus.

[0066] This invention provides a human-machine fusion group discussion method based on a large-scale language model. It monitors the group discussion, obtaining the discussion results of both human and AI experts on the discussion question. Based on these results, it determines the viewpoints of both experts on the discussion question. Finally, it fuses these viewpoints to arrive at a decision on the corresponding discussion question. This method fully utilizes the model's computational power and human creativity, combining the powerful capabilities of a large-scale language model with human intelligence to jointly address complex problems in group discussions. This improves the efficiency and collaboration of the discussions, overcoming the limitations of traditional group discussions in fully utilizing machine computing power and human creativity. It enables group discussions to better adapt to the needs of solving complex problems and fostering innovative thinking, finding the most feasible solutions to complex problems, and demonstrating flexibility and practicality.

[0067] Based on the above embodiments, step 130 includes:

[0068] Feature extraction is performed on the discussion viewpoints of human experts and AI experts to obtain the semantic features of their respective discussion viewpoints.

[0069] Information was extracted from the discussion viewpoints of both human experts and AI experts to obtain key information about their respective viewpoints.

[0070] Based on the semantic features and key information of the discussion viewpoints of human experts and AI experts, the decision-making fusion of the discussion viewpoints of human experts and AI experts on the discussion problem is carried out to obtain the problem decision of the group discussion.

[0071] Specifically, step 130 involves fusing the viewpoints of human experts and AI experts on the discussion issue to obtain the problem decision for the group discussion. This process may include:

[0072] First, feature extraction can be performed on the discussion viewpoints of both human experts and AI experts to extract the semantic features of both viewpoints. That is, based on the discussion viewpoints of human experts and AI experts on the discussion issue, semantic feature extraction can be performed to extract the core semantic information that can represent the corresponding viewpoints from the discussion viewpoints of human experts and AI experts, and encode it to form the features of the corresponding viewpoints, thereby obtaining the semantic features of the discussion viewpoints of human experts and AI experts on the discussion issue.

[0073] Simultaneously, information can be extracted from the discussion viewpoints of both human and AI experts to extract key information from both perspectives. In other words, based on the discussion viewpoints of human and AI experts on the research issue, key information can be extracted to reveal the core information reflecting the corresponding viewpoints in both perspectives, thereby obtaining the key information from both perspectives.

[0074] Subsequently, based on the key information and semantic features of the discussion viewpoints of human experts and AI experts, decision fusion can be performed. Specifically, by utilizing the key information and semantic features of the discussion viewpoints, the discussion viewpoints of human experts and AI experts are first fused to obtain representative and comprehensive viewpoints. Then, based on these comprehensive viewpoints, decision generation is performed to generate an understanding of the discussion problem, solutions, and implementation paths, thereby obtaining a problem decision for the discussion problem.

[0075] Based on the above embodiments, and based on the semantic features and key information of the discussion viewpoints of human experts and AI experts, a decision fusion is performed on the discussion viewpoints of human experts and AI experts on the discussion issue to obtain the problem decision for the group discussion, including:

[0076] Based on key information from the discussions of both human and AI experts, candidate discussion viewpoints were selected from their respective perspectives.

[0077] Based on the semantic features of each candidate discussion viewpoint, semantic matching is performed on each candidate discussion viewpoint to obtain semantic matching results;

[0078] Based on the semantic matching results, the opinions of human experts and AI experts on the research issue are fused to obtain the problem decision for the research issue.

[0079] Specifically, the process of fusing the discussion viewpoints of human experts and AI experts on the discussion issue based on the semantic features and key information of their respective viewpoints to obtain the problem decision for the group discussion may include:

[0080] First, we can filter the key information from the discussion viewpoints of both human and AI experts to select highly accepted and representative viewpoints, i.e., candidate discussion viewpoints. Specifically, we can calculate multiple aspects such as information repetition, semantic similarity, text similarity, and logical relevance based on the key information of each discussion viewpoint to select viewpoints with high repetition, high semantic similarity, high text similarity, and high logical relevance from all discussion viewpoints. Such viewpoints can be considered as representative candidate discussion viewpoints that are widely recognized by experts.

[0081] Next, the semantic features of each selected candidate discussion viewpoint can be determined, and semantic matching can be performed on each candidate discussion viewpoint based on these semantic features to obtain semantic matching results. That is, based on the semantic features of each candidate discussion viewpoint, semantic matching can be performed on each candidate discussion viewpoint to determine which viewpoints are semantically similar or identical, thereby obtaining semantic matching results.

[0082] Here, the semantic matching result can be a candidate discussion point of view with semantic matching, or a candidate discussion point of view with semantic matching and its semantic matching degree, or a candidate discussion point of view with a high matching degree after being filtered by the matching degree threshold.

[0083] Subsequently, this semantic matching result can be used to fuse the discussion viewpoints of human experts and AI experts on the research issue, thereby obtaining the problem decision. That is, based on the semantic matching results representing the semantic matching situation between candidate discussion viewpoints, such as which candidate discussion viewpoints are semantically matched and the degree of semantic matching, the discussion viewpoints of various human experts and AI experts can be fused to obtain a representative and comprehensive viewpoint. Then, a decision can be generated based on this comprehensive viewpoint to generate an understanding of the research issue, solutions, and implementation paths, thereby obtaining the problem decision.

[0084] Based on the above embodiments, and based on the semantic matching results, the discussion opinions of human experts and AI experts on the research issue are fused to obtain the problem decision for the research issue, including:

[0085] Based on the semantic matching results, the viewpoints of each candidate discussion are merged to obtain the merged discussion viewpoint;

[0086] Based on the integrated discussion of viewpoints, decision-making is generated for the discussion issues, resulting in problem decisions for the discussion issues.

[0087] Specifically, based on semantic matching results, the process of fusing the opinions of human experts and AI experts on the discussion issue to obtain the issue decision can include the following steps:

[0088] First, the semantic matching results obtained from semantic matching can be used to fuse viewpoints with high semantic similarity to extract comprehensive and representative viewpoints, thus obtaining a fused discussion viewpoint. That is, the viewpoints can be fused based on the semantic matching results between candidate discussion viewpoints. For example, the candidate discussion viewpoints of each group of semantic matching can be sorted in order of similarity from large to small or small to large, and the viewpoints of the top few groups of candidate discussion viewpoints can be fused within the group. Alternatively, all candidate discussion viewpoints of semantic matching can be fused, or candidate discussion viewpoints with semantic matching degree greater than the matching degree threshold can be fused to obtain a fused viewpoint. Then, the final fused discussion viewpoint can be obtained directly based on this fused viewpoint, or based on this fused viewpoint combined with viewpoints of each candidate fused viewpoint that have not been fused.

[0089] Subsequently, based on this fusion of discussion viewpoints, a decision can be generated for the discussion problem, resulting in a problem decision. This means that based on the comprehensive and representative fusion of discussion viewpoints, a decision can be generated for the discussion problem during the group discussion process. This involves generating an understanding, solutions, and implementation paths for the discussion problem through a comprehensive perspective, thus obtaining the problem decision. Specifically, the discussion system can use a decision fusion engine to perform decision fusion to obtain the problem decision. That is, the decision fusion engine calls its solution generation tools or utilizes the solution generation capabilities of a large language model. Based on the fusion of discussion viewpoints, and combining the candidate discussion viewpoints and their semantic matching results, it generates and optimizes decision schemes to obtain the decision scheme for the current discussion round. Furthermore, this decision needs to be iteratively optimized. The discussion system and experts can evaluate and provide feedback on the current decision scheme, inputting it into the group discussion for multiple rounds of iterative optimization until the final decision, i.e., the problem decision, is obtained.

[0090] Based on the above embodiments, and based on the semantic matching results, the candidate discussion viewpoints are fused to obtain fused discussion viewpoints, including:

[0091] Based on the semantic matching results, the candidate discussion viewpoints are merged to obtain the first category of discussion viewpoints;

[0092] The topic is extracted based on the discussion opinions of human experts and AI experts, and the topic and topic intensity of the discussion opinions of human experts and AI experts are obtained.

[0093] Based on the topic and intensity of the viewpoints, the viewpoints of human experts and AI experts are merged to obtain the second category of discussion viewpoints;

[0094] The first and second types of discussion viewpoints are integrated to obtain a merged discussion viewpoint.

[0095] Specifically, based on the semantic matching results, the process of fusing the candidate discussion viewpoints to obtain the fused discussion viewpoint includes the following steps:

[0096] First, based on the semantic matching results mentioned above, the candidate discussion viewpoints can be merged to obtain the first type of discussion viewpoints. That is, the semantic matching results obtained from semantic matching can be used to merge viewpoints with high semantic similarity to obtain a comprehensive and representative viewpoint after merging. Specifically, the viewpoints can be merged based on the semantic matching situation between the candidate discussion viewpoints as represented by the semantic matching results. For example, the candidate discussion viewpoints of each group of semantic matching can be sorted in order of similarity from large to small or from small to large, and the viewpoints within the first few groups of candidate discussion viewpoints can be merged. Alternatively, all candidate discussion viewpoints of semantic matching can be merged, or candidate discussion viewpoints with semantic matching degree greater than the matching degree threshold can be merged to obtain the merged viewpoints, i.e., the first type of discussion viewpoints.

[0097] Simultaneously, the discussion viewpoints of human experts and AI experts can be extracted to obtain the viewpoint topics and topic strengths of human experts and AI experts. That is, topic mining and summary extraction can be performed separately based on the discussion viewpoints of human experts and AI experts to obtain the topic information and summary information of the discussion viewpoints of human experts and AI experts. That is, the viewpoint topics of human experts and AI experts. Furthermore, similarity calculations and correlation calculations between topics and discussion issues can be performed based on the viewpoint topics of each discussion viewpoint to determine the topic strength of each viewpoint topic.

[0098] Next, based on the viewpoints and their intensity obtained in the previous step, the viewpoints of human experts and AI experts can be merged to obtain the second type of viewpoints. That is, based on the viewpoints and their intensity, viewpoints with high content repetition, high similarity between topics, close relevance to the discussion question, and similar logic can be merged to obtain the merged viewpoints. At the same time, viewpoints with extremely low relevance to the discussion question can be discarded. Based on the retained viewpoints and the merged viewpoints, the second type of viewpoints can be determined.

[0099] Then, the first and second types of discussion viewpoints can be merged to obtain the fused discussion viewpoint. That is, the two types of target discussion viewpoints obtained from the discussion viewpoints of human experts and AI experts through different processing methods can be merged to extract the creative, helpful, comprehensive and representative viewpoints from both, and then merge them to obtain the final fused discussion viewpoint.

[0100] Based on the above embodiments, the overall process of decision fusion may include:

[0101] Figure 3 This is a flowchart illustrating the decision fusion process provided by the present invention, as shown below. Figure 3 As shown, during the consensus-building and evaluation phases, the discussion system can utilize a large-scale language model to integrate and summarize various suggestions and opinions from experts. Through intelligent algorithms, it compares and analyzes different opinions, identifies consensus points and core viewpoints, thereby promoting the formation and achievement of consensus. Its implementation process mainly involves the following steps: viewpoint fusion, decision generation, and iterative optimization.

[0102] During the viewpoint fusion process, the discussion system utilizes topic models to extract topics from the discussion viewpoints of human experts and AI experts to obtain viewpoint topics and determine topic strength. Viewpoints are then merged based on topic and topic strength. Simultaneously, information extraction and feature extraction are performed on the discussion viewpoints of both human and AI experts. The semantic computing capabilities of the large language model are used to conduct in-depth analysis of each discussion viewpoint, extracting key information and semantic features for viewpoint screening and semantic matching. Viewpoints are then merged based on the semantic matching results. Afterward, the merged viewpoints from the above two types can be further fused to obtain the fused discussion viewpoints.

[0103] In the decision generation process, the decision fusion engine of the discussion system calls upon its solution generation tools or utilizes the solution generation capabilities of a large language model. Based on the fusion of discussion viewpoints, and combined with the candidate discussion viewpoints and their semantic matching results, it generates and optimizes decision solutions to obtain the decision solution for the current discussion round.

[0104] During the iterative optimization phase, the discussion system and experts can evaluate and provide feedback on the current decision-making options, and input them into the group discussion for multiple rounds of iterative optimization until a final decision is reached, namely the problem decision of the discussion issue.

[0105] Throughout the decision-making process, through the calculation and deduction of a large-scale language model, the discussion system can not only quantitatively and qualitatively assess the degree of consensus but also provide objective references for decision-making solutions. If disputes arise or sufficient consensus is not achieved after the assessment, the discussion system can input feedback information back into the in-depth discussion stage to further deepen the discussion. This ensures that the formation of the problem decision fully respects and considers the opinions of each expert participating in the group discussion, achieving the greatest possible consensus while maintaining openness and allowing for the exploration of more opinions.

[0106] The human-computer fusion group discussion device based on a large language model provided by the present invention is described below. The human-computer fusion group discussion device based on a large language model described below and the human-computer fusion group discussion method based on a large language model described above can be referred to and corresponded to each other.

[0107] Figure 4 This is a schematic diagram of the human-computer integrated group discussion device based on a large language model provided by the present invention, as shown below. Figure 4 As shown, the device is used in a seminar system that integrates artificial intelligence (AI) experts. These AI experts are constructed using a large language model. The device includes:

[0108] The discussion monitoring unit 410 is used to monitor the group discussion and obtain the discussion results of the human experts and the AI ​​experts participating in the group discussion on the discussion questions of the group discussion.

[0109] The opinion determination unit 420 is used to determine the discussion opinions of the human experts and the AI ​​experts on the discussion issue based on the discussion results of the human experts and the AI ​​experts, respectively.

[0110] The decision fusion unit 430 is used to fuse the discussion opinions of the human experts and the AI ​​experts on the discussion problem to obtain the problem decision of the group discussion.

[0111] This invention provides a human-machine fusion group discussion device based on a large-scale language model. It monitors group discussions, obtaining the discussion results of both human and AI experts on the discussion problem. Based on these results, it determines the viewpoints of both experts on the problem. Finally, it fuses these viewpoints to arrive at a problem decision for the group discussion. This device fully utilizes the model's computational power and human creativity, combining the powerful capabilities of a large-scale language model with human intelligence to address complex problems in group discussions. This improves the efficiency and collaboration of discussions, overcoming the limitations of traditional group discussions in fully leveraging machine computing power and human creativity. It enables group discussions to better adapt to the needs of solving complex problems and fostering innovative thinking, finding the most feasible solutions to complex issues, and demonstrating flexibility and practicality.

[0112] Based on the above embodiments, the decision fusion unit 430 is used for:

[0113] Feature extraction is performed on the discussion viewpoints of the human experts and the AI ​​experts respectively to obtain the semantic features of the discussion viewpoints of the human experts and the AI ​​experts respectively;

[0114] Information is extracted from the discussion viewpoints of the human experts and the AI ​​experts to obtain key information about their respective discussion viewpoints;

[0115] Based on the semantic features and key information of the discussion viewpoints of the human experts and the AI ​​experts, the discussion viewpoints of the human experts and the AI ​​experts on the discussion problem are fused to obtain the problem decision of the discussion problem of the group discussion.

[0116] Based on the above embodiments, the decision fusion unit 430 is used for:

[0117] Based on the key information of the discussion viewpoints of the human experts and the AI ​​experts, candidate discussion viewpoints are selected from the discussion viewpoints of the human experts and the AI ​​experts.

[0118] Based on the semantic features of each candidate discussion viewpoint, semantic matching is performed on each candidate discussion viewpoint to obtain semantic matching results;

[0119] Based on the semantic matching results, the discussion opinions of the human experts and the AI ​​experts on the research question are fused to obtain the problem decision for the research question.

[0120] Based on the above embodiments, the decision fusion unit 430 is used for:

[0121] Based on the semantic matching results, the candidate discussion viewpoints are fused to obtain a fused discussion viewpoint;

[0122] Based on the integrated discussion viewpoints, a decision is generated for the discussion question, resulting in a problem decision for the discussion question.

[0123] Based on the above embodiments, the decision fusion unit 430 is used for:

[0124] Based on the semantic matching results, the candidate discussion viewpoints are merged to obtain the first category of discussion viewpoints;

[0125] The topic is extracted based on the discussion opinions of the human experts and the AI ​​experts to obtain the topic and topic intensity of the discussion opinions of the human experts and the AI ​​experts;

[0126] Based on the topic and intensity of the viewpoints, the viewpoints of the human experts and the AI ​​experts are merged to obtain the second type of viewpoints.

[0127] The first type of discussion viewpoints and the second type of discussion viewpoints are merged to obtain a merged discussion viewpoint.

[0128] Based on the above embodiments, in the group discussion, the AI ​​experts and the human experts work together to jointly discuss the discussion questions and obtain discussion results;

[0129] The discussion results include at least one of the following: the record of speeches made during the group discussion, expert actions, the breakdown record of the discussion questions, and the opinion vote on the discussion questions.

[0130] Figure 5 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 5As shown, the electronic device may include a processor 510, a communications interface 520, a memory 530, and a communication bus 540, wherein the processor 510, communications interface 520, and memory 530 communicate with each other via the communication bus 540. The processor 510 can call logical instructions in the memory 530 to execute a human-machine fusion group discussion method based on a large language model. This method is applied to a discussion system that integrates artificial intelligence (AI) experts, which are constructed using a large language model. The method includes: monitoring the group discussion to obtain the discussion results of the human experts and the AI ​​experts participating in the group discussion on the discussion question; determining the discussion opinions of the human experts and the AI ​​experts on the discussion question based on their respective discussion results; and performing decision fusion on the discussion opinions of the human experts and the AI ​​experts on the discussion question to obtain the problem decision for the group discussion.

[0131] Furthermore, the logical instructions in the aforementioned memory 530 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, essentially, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0132] On the other hand, the present invention also provides a computer program product, the computer program product comprising a computer program stored on a non-transitory computer-readable storage medium, the computer program comprising program instructions, and when the program instructions are executed by a computer, the computer is able to execute the human-computer fusion group discussion method based on a large language model provided by the above methods. This method is applied to a discussion system, the discussion system integrating artificial intelligence (AI) experts, the AI ​​experts being constructed through a large language model. The method includes monitoring the group discussion to obtain the discussion results of the human experts and the AI ​​experts participating in the group discussion on the discussion question; determining the discussion opinions of the human experts and the AI ​​experts on the discussion question based on their respective discussion results; and performing decision fusion on the discussion opinions of the human experts and the AI ​​experts on the discussion question to obtain the problem decision for the group discussion.

[0133] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, is implemented to perform the human-computer fusion group discussion method based on a large language model provided by the methods described above. This method is applied to a discussion system integrating artificial intelligence (AI) experts, which are constructed using a large language model. The method includes: monitoring the group discussion to obtain the discussion results of the human experts and the AI ​​experts participating in the group discussion on the discussion question; determining the discussion opinions of the human experts and the AI ​​experts on the discussion question based on their respective discussion results; and performing decision fusion on the discussion opinions of the human experts and the AI ​​experts on the discussion question to obtain a problem decision for the group discussion.

[0134] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0135] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0136] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A human-computer integrated group discussion method based on a large-scale language model, characterized in that, The method, applied to a seminar system integrating artificial intelligence (AI) experts constructed using a large language model, includes: The group discussion is monitored to obtain the discussion results of the human experts and the AI ​​experts participating in the group discussion on the discussion questions of the group discussion; Based on the discussion results of the human experts and the AI ​​experts, the discussion opinions of the human experts and the AI ​​experts on the discussion issue are determined respectively; Feature extraction is performed on the discussion viewpoints of the human experts and the AI ​​experts respectively to obtain the semantic features of the discussion viewpoints of the human experts and the AI ​​experts respectively; Information is extracted from the discussion viewpoints of the human experts and the AI ​​experts to obtain key information about their respective discussion viewpoints; Based on the key information of the discussion viewpoints of the human experts and the AI ​​experts, candidate discussion viewpoints are selected from the discussion viewpoints of the human experts and the AI ​​experts. Based on the semantic features of each candidate discussion viewpoint, semantic matching is performed on each candidate discussion viewpoint to obtain semantic matching results; Based on the semantic matching results, the candidate discussion viewpoints are merged to obtain the first category of discussion viewpoints; The topic is extracted based on the discussion opinions of the human experts and the AI ​​experts to obtain the topic and topic intensity of the discussion opinions of the human experts and the AI ​​experts; Based on the topic and intensity of the viewpoints, the viewpoints of the human experts and the AI ​​experts are merged to obtain the second type of viewpoints. The first type of discussion viewpoints and the second type of discussion viewpoints are integrated to obtain a merged discussion viewpoint; Based on the integrated discussion viewpoints, a decision is generated for the discussion question, resulting in a problem decision for the discussion question.

2. The human-computer integrated group discussion method based on a large-scale language model according to claim 1, characterized in that, In the group discussion, the AI ​​experts and human experts work together to discuss the research questions and obtain research results. The discussion results include at least one of the following: the record of speeches made during the group discussion, expert actions, the breakdown record of the discussion questions, and the opinion vote on the discussion questions.

3. A human-computer integrated group discussion device based on a large-scale language model, characterized in that, The device is applied to a seminar system that integrates artificial intelligence (AI) experts, which are constructed using large-scale language models. The discussion monitoring unit is used to monitor the group discussion and obtain the discussion results of the human experts and the AI ​​experts participating in the group discussion on the discussion questions of the group discussion. The opinion determination unit is used to determine the opinions of the human experts and the AI ​​experts on the discussion issue based on the discussion results of the human experts and the AI ​​experts, respectively. The decision fusion unit is used to extract features from the discussion viewpoints of the human experts and the AI ​​experts to obtain semantic features of their respective viewpoints; extract information from the discussion viewpoints of the human experts and the AI ​​experts to obtain key information of their respective viewpoints; based on the key information of their respective viewpoints, select candidate discussion viewpoints from the discussion viewpoints of the human experts and the AI ​​experts; and perform semantic matching on the candidate discussion viewpoints based on their semantic features to obtain semantic matching results. Based on the semantic matching results, the candidate discussion viewpoints are merged to obtain a first type of discussion viewpoint; topic extraction is performed based on the discussion viewpoints of the human experts and the AI ​​experts to obtain the viewpoint topics and topic strengths of the discussion viewpoints of the human experts and the AI ​​experts; based on the viewpoint topics and topic strengths, the discussion viewpoints of the human experts and the AI ​​experts are merged to obtain a second type of discussion viewpoint; the first type of discussion viewpoint and the second type of discussion viewpoint are fused to obtain a fused discussion viewpoint; based on the fused discussion viewpoint, a decision is generated for the discussion question to obtain the question decision for the discussion question.

4. An electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the human-computer fusion group discussion method based on a large language model as described in claim 1 or 2.

5. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the human-computer fusion group discussion method based on a large language model as described in claim 1 or 2.

6. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the human-computer fusion group discussion method based on a large language model as described in claim 1 or 2.

Images