Capability assessment method and device for group discussion participants, electronic equipment and storage medium

Abstract

The invention provides a capability assessment method and device for group discussion participants, electronic equipment and a storage medium, and relates to the technical field of artificial intelligence, and the method comprises the steps: obtaining text data of group discussion; inputting the text data into a pre-trained personal general ability assessment model and a pre-trained team leader assessment model, and outputting a personal general ability assessment result and a team leader assessment result; wherein the personal universal ability evaluation model is a large language model which is trained by performing supervision and fine tuning through a training data set comprising an information node, a decision node and a utility node and combining a low-rank self-adaptive method; the team leader ability evaluation model is used for evaluating the thinking excitation ability, conflict coordination ability and discussion field control ability of the participants based on an event sequence containing participant identifiers, speaking keywords and speaking timestamps. Through the method provided by the invention, the personal general ability can be effectively evaluated, and the team leader ability can be quantitatively evaluated.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a method, apparatus, electronic device, and storage medium for assessing the abilities of participants in a group discussion. Background Technology

[0002] In public administration and corporate decision-making, group discussion is a key link in pooling collective wisdom and forming scientific decisions.

[0003] However, current technologies for evaluating discussion processes primarily rely on manual observation or traditional text analysis, which has significant limitations. Manual evaluation methods are highly subjective and lack dimensionality, making it difficult to quantify participants' deep thinking abilities and dynamic interaction effectiveness. Traditional text analysis tools, on the other hand, often remain at the level of static word frequency statistics or superficial semantic analysis, failing to delve into the decision-making logic and systems thinking inherent in speeches, nor effectively capture and quantify participants' team leadership behaviors such as inspiring others, resolving conflicts, and guiding consensus. Furthermore, while general language models possess powerful language understanding capabilities, directly applying them to this complex evaluation scenario suffers from poor interpretability, insufficient differentiation of subtle individual ability differences, and a superficial understanding of dynamic interaction processes. This often results in superficial evaluation results that fail to provide accurate and actionable improvement criteria. Therefore, a method is urgently needed to address these issues. Summary of the Invention

[0004] This invention provides a method, apparatus, electronic device, and storage medium for assessing the abilities of participants in group discussions, in order to address the deficiencies in the prior art.

[0005] This invention provides a method for assessing the abilities of participants in group discussions, comprising the following steps: Obtain the text data of the group discussion; The text data is input into a pre-trained personal general ability assessment model and a team leadership ability assessment model, and the personal general ability assessment results and team leadership ability assessment results are output. The individual general ability assessment model is constructed based on decision analysis theory. It is fine-tuned under supervision using a training dataset containing information nodes, decision nodes, and utility nodes, and is trained using a large language model with a low-rank adaptive method. The team leadership ability assessment model is a dynamic interactive assessment model constructed based on a multidimensional Hawkes process. It is used to assess participants' ability to stimulate thinking, coordinate conflicts, and control discussions based on an event sequence containing participant identifiers, speaking keywords, and speaking timestamps.

[0006] According to a method for assessing the abilities of participants in a group discussion provided by the present invention, the training process of the individual general ability assessment model includes: Obtain a pre-constructed training dataset; wherein the training dataset includes: the text to be analyzed and the corresponding standard nodes; the standard nodes include: information nodes, decision nodes, and utility nodes; Based on the training dataset, the pre-trained large language model is fine-tuned using a low-rank adaptive supervised fine-tuning method to enable the model to identify and extract standard nodes from text, thereby obtaining the personal general ability assessment model. The low-rank adaptive method freezes the main parameters of the pre-trained large language model and injects trainable adapter parameters only into the attention layer to adapt the model to the standard node extraction task.

[0007] According to the present invention, a method for assessing the abilities of participants in a group discussion includes inputting the text data into a pre-trained personal general ability assessment model and outputting personal general ability assessment results, comprising: Based on a knowledge corpus related to the discussion topic, a reference answer influence graph is constructed using the aforementioned personal general ability assessment model; wherein, the reference answer influence graph includes: information nodes and directed relations, decision nodes and directed relations, and utility nodes and directed relations; Based on the content of the participants' statements in the group discussion, the nodes are extracted and semantically mapped through the personal general ability assessment model, and associated with the standard nodes and directed relationships of the reference answer influence graph to generate a personal influence graph for each participant. The evaluation is based on the node hierarchy distribution in the participant's personal influence graph, generating the personal general ability evaluation result.

[0008] According to a method for assessing the abilities of participants in a group discussion provided by the present invention, the process of constructing the knowledge corpus includes: The large language model is invoked to generate decision analysis topics and corresponding core keywords; Based on the decision analysis topic and the corresponding core keywords, real-time background information is obtained through enhanced retrieval generation. The decision analysis topic, its corresponding core keywords, and the real-time background information are integrated to generate the knowledge corpus.

[0009] According to a method for assessing the abilities of participants in a group discussion provided by the present invention, the individual general ability assessment includes at least: systems thinking ability assessment and solution feasibility assessment ability assessment; The evaluation, based on the node hierarchy distribution in the participant's personal influence graph, generates the individual general ability evaluation result, including: By analyzing the abstract hierarchical distribution of nodes in the participant's personal influence diagram, the number of nodes for macro-themes, abstract concepts, and specific information is counted, and the system's thinking ability is evaluated based on the product of the number of nodes for macro-themes, abstract concepts, and specific information. The feasibility assessment capability of the proposed solutions is evaluated by calling a large model intelligent agent to score the decision-making solutions proposed by the participants in multiple dimensions. The multiple dimensions include: awareness of resource constraints, risk identification and management, perception of execution threshold, and dynamic considerations.

[0010] According to a method for assessing the capabilities of participants in a group discussion provided by the present invention, the step of inputting the text data into a pre-trained team leadership capability assessment model and outputting team leadership capability assessment results includes: The text data of the group discussion is preprocessed into an event sequence; wherein each event in the event sequence includes a participant identifier, speaking keywords, and speaking timestamp; Based on the event sequence, the team leadership assessment model is used to evaluate participants' ability to stimulate thinking, coordinate conflicts, and control discussions.

[0011] According to a method for assessing the abilities of participants in a group discussion provided by the present invention, the assessment of participants' ability to stimulate thinking based on the event sequence and through the team leadership ability assessment model includes: Based on the event sequence, the self-motivation intensity parameter and the other-motivation intensity parameter of each participant in the group discussion are calculated using a multidimensional Hawkes process model. The thinking stimulation ability of each participant is evaluated based on the weighted sum of the self-motivation intensity parameter and the other-motivation intensity parameter of each participant.

[0012] According to a method for assessing the capabilities of participants in a group discussion provided by the present invention, the conflict coordination capability includes: conflict resolution capability, openness capability, and collaborative capability; The assessment of participants' conflict resolution capabilities based on the event sequence using the team leadership assessment model includes: A topic distribution vector is generated based on the speaking keywords in the event sequence, and a score is calculated to indicate the dominance of each participant's viewpoint. Using the topic distribution vector as the independent variable and the opinion dominance score of each participant as the dependent variable, a ridge regression model is trained to obtain the influence coefficient. Based on the time-varying influence coefficient and the dominance of the viewpoint, calculate the conflict resolution ability score for each participant; and, The number of open-ended actions by each participant was counted. Each participant's openness ability is assessed based on the total number of openness behaviors they exhibit; and, Support and negation relationships are extracted from the text data, and a participant interaction network graph is constructed based on the support and negation relationships; The collaborative ability of each participant is assessed based on the net support of each participant in the interaction network graph.

[0013] According to the present invention, a method for assessing the abilities of participants in a group discussion includes the ability to control the discussion, which includes: focusing ability, time management ability, and convergence ability. The assessment of participants' discussion control abilities based on the event sequence using the team leadership assessment model includes: Extract the core discussion topic and calculate the semantic distance between each participant's statement and the core topic; The focusing ability of each participant is assessed based on the average semantic distance between each participant's statement and the core topic; and, The number of time-control actions of each participant was counted. Each participant's time management ability is assessed based on the total number of time control behaviors exhibited by the participants; and, Summarize the consensus points discussed, and calculate the change in semantic distance between the discussion content and the consensus points before and after each participant speaks. The convergence capability of each participant is evaluated based on the cumulative reduction in semantic distance resulting from all of their statements.

[0014] The present invention also provides a device for assessing the abilities of participants in a group discussion, comprising the following modules: The acquisition module is used to acquire text data from group discussions; The evaluation module is used to input the text data into a pre-trained personal general ability evaluation model and a team leadership ability evaluation model, and output personal general ability evaluation results and team leadership ability evaluation results. The individual general ability assessment model is constructed based on decision analysis theory. It is fine-tuned under supervision using a training dataset containing information nodes, decision nodes, and utility nodes, and is trained using a large language model with a low-rank adaptive method. The team leadership ability assessment model is a dynamic interactive assessment model constructed based on a multidimensional Hawkes process. It is used to assess participants' ability to stimulate thinking, coordinate conflicts, and control discussions based on an event sequence containing participant identifiers, speaking keywords, and speaking timestamps.

[0015] 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 computer program to implement the competence assessment method for group discussion participants as described above.

[0016] 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 competence assessment method for group discussion participants as described above.

[0017] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the competence assessment method for group discussion participants as described above.

[0018] This invention provides a method, apparatus, electronic device, and storage medium for assessing the abilities of participants in group discussions. It acquires text data from group discussions; inputs the text data into pre-trained individual general ability assessment models and team leadership ability assessment models; and outputs individual general ability assessment results and team leadership ability assessment results. The individual general ability assessment model is constructed based on decision analysis theory, using a training dataset containing information nodes, decision nodes, and utility nodes for supervised fine-tuning, and combined with a low-rank adaptive method to train a large language model. The team leadership ability assessment model is a dynamic interactive assessment model based on a multidimensional Hawkes process, used to assess participants' thinking stimulation ability, conflict coordination ability, and discussion control ability based on event sequences containing participant identifiers, speaking keywords, and speaking timestamps. Therefore, this invention utilizes a supervised fine-tuned large language model to parse participant speech, effectively assessing their individual general abilities. Simultaneously, by applying a multidimensional Hawkes process to dynamically model the interaction event sequences between participants, it quantitatively assesses their team leadership abilities. This not only overcomes the subjectivity and bias of manual assessment but also breaks through the bottleneck of traditional text analysis's insufficient capture of dynamic interactive processes. Attached Figure Description

[0019] 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.

[0020] Figure 1 This is a flowchart illustrating the ability assessment method for group discussion participants provided by the present invention.

[0021] Figure 2This is a schematic diagram of the ability assessment device for group discussion participants provided by the present invention.

[0022] Figure 3 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0023] 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.

[0024] The following is combined Figures 1-3 This invention describes a method, device, electronic equipment, and storage medium for assessing the abilities of participants in a group discussion.

[0025] It's important to note that group discussion is a common and effective decision-making method in public administration and within enterprises, particularly for dealing with complex and far-reaching decisions. Ideally, group discussions can pool collective wisdom, stimulate innovative thinking, and integrate diverse value preferences, thereby improving the quality of decisions. However, the group discussion process is also prone to problems such as groupthink, production blocking, and discussion polarization, which can undermine the scientific rigor and effectiveness of decisions. Therefore, how to objectively, comprehensively, and deeply analyze the group discussion process and accurately assess the individual capabilities of participants and the effectiveness of team interactions has become a key challenge in organizational management.

[0026] Currently, assessment methods for group discussions have significant limitations in the following aspects: 1. Inherent defects of traditional manual evaluation methods: These methods rely heavily on human observation and expert experience, and have many pain points: strong subjectivity, vague evaluation criteria, and results are easily affected by factors such as the expert's personal preferences, experience limitations, and even distraction during the session; single dimension, often focusing too much on outcome-oriented and explicit indicators such as "who speaks the most," while neglecting key interactive processes such as stimulating viewpoints and resolving conflicts; empty feedback, with evaluation reports mostly remaining at the level of qualitative descriptions such as "proactive" and "needs improvement," lacking quantifiable and actionable improvement suggestions.

[0027] 2. The limitations of assessment tools: Some assessment tools on the market also have shortcomings. For example, psychological scales are based on static dimensions and cannot capture the real-time strategy adjustments made by participants in the discussion according to the changes in the situation. Traditional text analysis methods usually regard group performance as a simple sum of individual contributions, ignoring the dynamic game and mutual influence among participants, and cannot distinguish between "superficial harmony" and "real cooperation".

[0028] 3. Bottlenecks in the application of the General Large Language Model (LLM): Although the LLM is powerful in language processing, its limitations are also very obvious when it is directly used for talent assessment: it lacks interpretability, and its conclusion generation process is like a "black box," making it difficult to trace the basis; it lacks discrimination, and in scenarios with a large number of participants and dense opinions, it is difficult to effectively identify subtle differences in abilities between individuals; it has a shallow understanding of the scenario, and can only complete surface semantic analysis, unable to deeply understand the complex value trade-offs and dynamic games behind the discussion, and is prone to outputting "correct but useless statements"; its value ceiling is obvious, and its function is easily substitutable.

[0029] Based on this, the present invention provides a method for assessing the abilities of participants in group discussions, in order to solve at least one of the above-mentioned problems.

[0030] Figure 1 This is a flowchart illustrating the ability assessment method for group discussion participants provided by the present invention, as shown below. Figure 1 As shown, the method includes the following: Step 100: Obtain the text data of the group discussion.

[0031] Specifically, text data of group discussions refers to transcripts of the discussion, either transcribed from audio or directly obtained, and must include speaker identifiers and timestamps.

[0032] Step 200: Input the text data into the pre-trained personal general ability assessment model and team leadership ability assessment model, and output the personal general ability assessment results and team leadership ability assessment results; The individual general ability assessment model is constructed based on decision analysis theory. It is fine-tuned under supervision using a training dataset containing information nodes, decision nodes, and utility nodes, and is trained using a large language model with a low-rank adaptive method. The team leadership ability assessment model is a dynamic interactive assessment model constructed based on a multidimensional Hawkes process. It is used to assess participants' ability to stimulate thinking, coordinate conflicts, and control discussions based on an event sequence containing participant identifiers, speaking keywords, and speaking timestamps.

[0033] Table 1

[0034] Table 1 shows the list of ability dimensions. As shown in Table 1, in the personal general ability assessment section, this embodiment assesses three personal general abilities: thinking ability, decision-making ability, and innovation ability, as well as nine sub-abilities under them. The method of using a large language model combined with decision analysis theory is used to analyze the participants' speech content and compare it with the structure defined in the influence diagram. Specifically, general personal abilities include: thinking ability, which is further subdivided into logical thinking ability, abstract inductive ability, and systems thinking ability. This is assessed by identifying logical reasoning patterns in speeches, analyzing the abstraction process of extracting patterns from specific cases, and identifying a systematic analytical framework of "macro-meso-micro" levels; decision-making ability, which is further subdivided into goal identification and decomposition ability, information collection and integration ability, and the ability to predict the impact of solutions. This is assessed by identifying the decomposition relationship between means and ends, counting the number of key information nodes mentioned by participants, and the number of times they predict the possible consequences of decision-making solutions; and innovation ability, which is further subdivided into creativity, associative ability, and feasibility assessment ability. This is assessed by calculating the number of novel viewpoints in speeches and the semantic span between viewpoints, identifying the correlation patterns of multiple pieces of information or solutions being affected by the same factor, and using a large model to measure the feasibility of solutions from multiple dimensions.

[0035] Referring again to Table 1, in the team leadership ability assessment section, this embodiment evaluates three team leadership abilities—thinking stimulation ability, conflict coordination ability, and discussion control ability—and their seven sub-abilities, using a dynamic process model (rather than static graph structure analysis) to capture real-time interactions among participants. Specifically, team leadership abilities include: thinking stimulation ability, which is assessed by modeling the emergence of new viewpoints in the discussion as a multidimensional Hawkes process and calculating the motivational intensity parameters of each participant towards others and themselves; conflict coordination ability, which is subdivided into openness, conflict resolution ability, and collaboration, by identifying the distribution of discussion topics and using regression analysis to calculate the positive or negative impact of participants' speeches on the degree of conflict on the field, counting the number of positive openness behaviors such as "accepting others' viewpoints," and counting the frequency of viewpoints being accepted or rejected by others; and discussion control ability, which is subdivided into focus, time management ability, and convergence, by calculating the distance between participants' speeches and the core discussion topic, as well as the change in the distance between the discussion content and the consensus point before and after their speeches, identifying and counting the number of specific control behaviors for assessment.

[0036] The Personal General Competency Assessment Model is a specialized, customized large language model built upon decision analysis theory. This theory deconstructs complex decision-making problems into elements such as objectives, information, solutions, and utility. Based on this, the model undergoes supervised fine-tuning using a training dataset containing information nodes, decision nodes, and utility nodes. For example, a sample in the training data might contain the text "Implementing a four-day work week can improve employee well-being, but the impact on customer service needs to be assessed," labeled with the corresponding information nodes "four-day work week" and "customer service," the decision node "whether to implement," and the utility nodes "improvement of employee well-being" and "risk of service response delay." Low-rank adaptive methods play a crucial role in this process. By freezing most of the parameters of the large language model and injecting only a small number of trainable low-rank adapter matrices into the attention layer of the Transformer architecture, it enables the general model to accurately adapt to the specialized task of node extraction at extremely low cost.

[0037] The team leadership competence assessment model is based on the Multidimensional Hawkes process, a classic event sequence modeling method that quantifies the incentive effect of historical events on the probability of future events. In this embodiment, the model preprocesses the discussion text into an event sequence. Each event includes a participant identifier (e.g., "Employee A"), speaking keywords (e.g., "shift system"), and speaking timestamp (e.g., "00:03:15"). By calculating the intensity of a participant's speech on their own and others' subsequent speeches, the model achieves a quantitative assessment of thinking stimulation ability (e.g., A's speech triggered continuous responses from B and C), conflict coordination ability (e.g., D's speech effectively eased the disagreement between E and F), and discussion control ability (e.g., G repeatedly guided the topic back to the meeting agenda).

[0038] Through the aforementioned dual-model collaborative architecture, this application achieves separate modeling and joint evaluation of the individual cognitive abilities and team interaction abilities of participants in group discussions.

[0039] In one embodiment, consider a group discussion on whether a four-day work week should be implemented. The system first acquires the text data of all participants' statements. In the individual general competence assessment, the model analyzes participant A's statements, identifying key points such as "improving employee well-being," "potentially reducing short-term corporate productivity," and "needing phased pilot implementation," thus assessing her strong systems thinking and decision-making abilities. Simultaneously, in the team leadership competence assessment, the Multidimensional Hawkes process model analyzes the event sequence: when participant B raises the point that "a four-day work week might lead to insufficient customer service coverage," the model captures that this event significantly stimulated participant C to add in subsequent statements that "a shift system could solve this," and also stimulated participant D to counter that "digital customer service can fill manpower gaps." By calculating the "stimulation intensity" of B's ​​statements on C and D, the model quantifies B's outstanding ability to stimulate thought. Finally, the system synthesizes the two assessment reports, comprehensively presenting each member's performance in cognitive depth and team influence.

[0040] The above describes the steps of the method for assessing the abilities of participants in group discussions provided by this invention. As can be seen from the above description, the method for assessing the abilities of participants in group discussions provided by this invention involves acquiring text data of the group discussion; inputting the text data into a pre-trained personal general ability assessment model and a team leadership ability assessment model; and outputting personal general ability assessment results and team leadership ability assessment results. The personal general ability assessment model is constructed based on decision analysis theory, and is a large language model trained using a training dataset containing information nodes, decision nodes, and utility nodes under supervised fine-tuning, combined with a low-rank adaptive method. The team leadership ability assessment model is a dynamic interactive assessment model constructed based on a multidimensional Hawkes process, used to assess participants' thinking stimulation ability, conflict coordination ability, and discussion control ability based on an event sequence containing participant identifiers, speaking keywords, and speaking timestamps. Therefore, this invention utilizes a supervised fine-tuned large language model to analyze participants' speeches, which can effectively assess their individual general abilities. At the same time, it applies a multidimensional Hawkes process to dynamically model the sequence of interactive events among participants, thereby quantitatively assessing their team leadership abilities. This not only overcomes the subjectivity and one-sidedness of manual assessment, but also breaks through the bottleneck of traditional text analysis's inadequate capture of dynamic interactive processes.

[0041] Based on the above embodiments, in this embodiment, the training process of the personal general ability assessment model includes: Step S210: Obtain a pre-constructed training dataset; wherein the training dataset includes: the text to be analyzed and the corresponding standard nodes; the standard nodes include: information nodes, decision nodes and utility nodes.

[0042] Step S220: Based on the training dataset, fine-tune the pre-trained large language model using a low-rank adaptive supervised fine-tuning method to enable the model to identify and extract standard nodes from text, thereby obtaining the personal general ability assessment model; wherein, the low-rank adaptive method freezes the main parameters of the pre-trained large language model and injects trainable adapter parameters only into the attention layer to adapt the model to the standard node extraction task.

[0043] It's important to note that Supervised Fine-Tuning (SFT) refers to training a pre-trained large language model on a specific dataset containing input-output paradigms, with the aim of teaching the model how to perform a specific task. Training an entire large language model typically requires enormous computational resources, while Low-Rank Adaptive (LoRA) is an efficient fine-tuning technique. Its core method involves freezing most of the parameters of the pre-trained model and adding only a small number of new trainable parameters (adaptors) to key parts of the model (such as the attention layer). By training only these adapters, the model can acquire new and specific skills with minimal computational cost.

[0044] This embodiment uses SFT-LoRA technology to train a large model specifically for extracting decision nodes. First, a training dataset is manually constructed based on corpora from fields such as public health and emergency decision-making. Each dataset contains the text to be analyzed and a list of information nodes, decision nodes, and utility nodes in a standard format from which the model is expected to extract. Then, LoRA technology is used to train the model on a base large model (deepseek-llm-7b-base), teaching the model how to accurately identify and extract these three types of nodes from the text according to instructions. This LoRA adapter is used for node extraction from reference answers and participant statements.

[0045] The ability assessment method for group discussion participants provided in this embodiment constructs a dedicated training dataset containing three types of standard nodes: information, decision, and utility. This provides the model with structured and quantifiable learning objectives, ensuring a close integration of ability assessment and decision analysis theory. At the same time, it employs supervised fine-tuning technology based on low-rank adaptation. While preserving the general knowledge and high-order reasoning capabilities of the pre-trained large language model, it efficiently and accurately adapts it to the specific task of node extraction with extremely low computational cost.

[0046] Based on the above embodiments, in this embodiment, step 200 inputs the text data into a pre-trained personal general ability assessment model and outputs the personal general ability assessment result, including: Step 210: Based on a knowledge corpus related to the discussion topic, construct a reference answer influence graph using the personal general ability assessment model; wherein, the reference answer influence graph includes: information nodes and directed relations, decision nodes and directed relations, and utility nodes and directed relations.

[0047] The construction process of the knowledge corpus includes: Step S211: Call the large language model to generate decision analysis topics and corresponding core keywords.

[0048] Step S212: Based on the decision analysis topic and the corresponding core keywords, obtain real-time background information through enhanced retrieval.

[0049] Step S213: Integrate the decision analysis topic, the corresponding core keywords, and the real-time background information to generate the knowledge corpus.

[0050] It should be noted that Retrieval Enhanced Generation (RAG) is a technical framework that combines a large language model with external information retrieval. By obtaining relevant information from real-time updated knowledge bases (such as the Internet), it provides accurate and dynamic context for the language model generation process, thereby significantly improving the accuracy and timeliness of the generated content.

[0051] This embodiment uses RAG to generate high-quality assessment questions closely integrated with real-world contexts. The process is as follows: First, a query statement is constructed based on the topics and keywords automatically generated by the large model. An external search engine (such as Google Search) is called through an interface to obtain the latest news, challenges, and developments related to the topic. Then, the title and content summary of each retrieved result are integrated to form a background information text containing real-time dynamics. Finally, this real-time background information is used as context and injected into the prompt template along with the original topics and keywords to guide the deepseek-reasoner to synthesize the final decision analysis question.

[0052] Specifically, this embodiment generates high-quality decision analysis questions using RAG. The specific operation process is as follows: First, the large language model is invoked to generate specific, decision analysis-suitable themes (such as "historical district renovation") and several core keywords (such as "cultural relic protection," "resident needs," "commercial development," "tourism development," and "infrastructure") around a broad field (such as "government management"). Then, based on the generated themes and keywords, a web search is automatically performed through the RAG framework to capture the latest news, challenges, and development trends related to the theme, and this information is integrated into dynamic and realistic background materials. Finally, the themes, keywords, and background materials are injected into specially designed prompts, which guide the large model to generate open-ended decision problems with high discriminative power across the "3+3" capability dimensions defined in Table 1 of this embodiment.

[0053] To ensure the effectiveness of the generated questions, this embodiment automates the verification of their discriminative power. This embodiment assigns different roles to the large model, generating two significantly different answers to the same question: a high-quality answer simulating the tone of an "industry expert skilled in decision analysis" and a low-quality answer simulating the tone of an "ordinary college graduate lacking work experience." Simultaneously, a simplified "reference answer" is generated for the question, covering the information, decisions, and utility required to solve the problem. This embodiment uses the following two methods to score the two simulated answers: calculating how many nodes each answer covers in the reference answer set (a high-quality answer should cover more and more core nodes); and calling the automatic evaluation system developed according to this embodiment to directly score the two answers on the "3+3" capability dimensions. When the "expert" answer significantly outperforms the "graduate" answer in most aspects of node coverage and capability scores, the question is considered to have sufficient discriminative power and is included in the knowledge corpus.

[0054] Step 220: Based on the content of the participants' speeches in the group discussion, nodes are extracted and semantically mapped using the personal general ability assessment model, and associated with the standard nodes and directed relationships of the reference answer influence graph to generate a personal influence graph for each participant.

[0055] It should be noted that this embodiment assesses three general personal abilities—thinking ability, decision-making ability, and innovation ability—and their nine sub-abilities. It uses a large language model combined with decision analysis theory to analyze the participants' statements and compare them with the structure defined in the influence diagram.

[0056] This embodiment first constructs a comprehensive and structured knowledge corpus on decision-making problems as a benchmark for assessing individual general abilities. It then utilizes the SFT-LoRA model to automatically extract information nodes, decision nodes, and utility nodes related to the decision-making problem from the corpus. Multiple large-scale model agents are used to perform quality screening and deduplication on these nodes, classifying them according to different dimensions, such as known / unknown information, final / means / goals, and macro-themes / abstract concepts / concrete information. Based on the categorized and labeled node list, this embodiment uses another set of agents to batch identify three types of directed edges that may exist between nodes: causal relationships, information support relationships, and evidence update relationships, constructing a complete reference answer influence graph.

[0057] Next, this embodiment constructs a participant influence graph. First, the speech of each participant is preprocessed to improve text quality. Then, the SFT-LoRA model is called again to extract all nodes mentioned in the speech. The semantic similarity between these nodes and the reference answer nodes is calculated using the SentenceTransformer model, and they are mapped to the normalized nodes of the reference answer. Finally, it is determined which directed edges in the reference answer influence graph are explicitly or implicitly mentioned in the participant's speech, and the mentioned directed edges are used to construct the directed edge set for that participant.

[0058] Step 230: Based on the node hierarchy distribution in the participant's personal influence graph, perform an evaluation to generate the personal general ability evaluation result.

[0059] It should be noted that, as shown in Table 1, the general personal ability assessment includes at least: systems thinking ability assessment and solution feasibility assessment ability assessment.

[0060] Step 230 evaluates the participants based on the node hierarchy distribution in their personal influence graph, generating the personal general ability assessment result, including: Step 231: By analyzing the abstract hierarchical distribution of nodes in the participant's personal influence graph, the number of nodes for macro-themes, abstract concepts, and specific information is counted, and the system's thinking ability is evaluated based on the product of the number of nodes for macro-themes, the number of nodes for abstract concepts, and the number of nodes for specific information.

[0061] Step 232: By calling the large model intelligent agent to score the decision-making schemes proposed by the participants in multiple dimensions, the feasibility assessment capability of the schemes is evaluated; wherein, the multiple dimensions include: resource constraint awareness, risk identification and management, execution threshold perception and dynamic consideration.

[0062] This embodiment assesses participants' systems thinking ability by analyzing the abstract hierarchy of nodes in their influence graph. It counts the number of nodes mentioned by participants that have been categorized into Level 1 (macro-level themes), Level 2 (abstract concepts), and Level 3 (specific information) in the reference answer, and uses their product to measure systems thinking ability. This assessment method ensures that participants only receive high scores when they can effectively connect macro-level themes, abstract concepts, and specific information.

[0063] This embodiment evaluates the feasibility assessment capability of each decision-making scheme proposed by participants by calling a large model to score each scheme from multiple dimensions. First, it identifies all proposed decision-making schemes from the participant's influence graph. Then, it uses a scoring agent to independently score each scheme from four dimensions: resource constraint awareness, risk identification and management, execution threshold perception, and dynamic considerations. The participant's score is the sum of the average scores of all proposed schemes across these four dimensions.

[0064] The method for assessing the abilities of participants in group discussions provided in this embodiment establishes an objective and quantifiable general ability assessment system by constructing a structured reference answer influence diagram as an assessment benchmark and generating individual influence diagrams of participants based on this for comparative analysis.

[0065] Based on the above embodiments, in this embodiment, step 200 inputs the text data into a pre-trained team leadership ability assessment model and outputs the team leadership ability assessment result, including: Step 240: Preprocess the text data of the group discussion into an event sequence; wherein each event in the event sequence includes a participant identifier, speaking keywords, and speaking timestamp.

[0066] Step 250: Based on the event sequence, assess the participants' ability to stimulate thinking, coordinate conflicts, and control discussions using the team leadership assessment model.

[0067] It should be noted that the Multidimensional Hawkes process is a statistical model used to analyze time-point event sequences. Its core characteristic lies in its ability to capture the stimulating effect between events, that is, the occurrence of one event may increase the probability of other future events. When an event increases the probability of the occurrence of events of the same type, it is called self-stimulation; when it increases the probability of the occurrence of events of different types, it is called mutual stimulation. The strength of the stimulating effect usually decays over time.

[0068] This embodiment applies a multidimensional Hawkes process to model the team discussion process as a sequence of events, where an event is defined as a participant uttering a keyword at a certain moment. The stimuli intensity between events depends on the speaker relationship (self / others), the semantic distance of the keywords, and the topic relevance of the keywords (whether they belong to the same topic cluster, which is obtained through Louvain community discovery algorithm clustering). This embodiment uses the maximum likelihood estimation (MLE) method to estimate the stimuli intensity, seeking parameters that best explain the actual speaking sequence. These parameters include self-motivation intensity (the ability of a speaker to motivate subsequent speakers) and other-motivation intensity (the ability to be motivated by the speeches of others). The optimization algorithm uses the Powell algorithm. Self-motivation intensity and other-motivation intensity are the scoring criteria for thinking stimulation ability.

[0069] This embodiment utilizes a topic model to represent unstructured discussion content as probability distribution vectors across several key topics. Then, through regression analysis, it establishes a statistical relationship between topic distribution and the dominance of each participant's viewpoint. In this way, it data-driven insights reveal each participant's preferences for different topics and quantifies the impact of their statements on group conflict.

[0070] This embodiment directly utilizes the semantic understanding capabilities of a large language model to generate topics (rather than a traditional statistical topic model). It designs prompts to enable the large model to automatically extract topics and represent the spoken content as probability distribution vectors on these topics. Furthermore, it uses the large model to assess the semantic similarity between the current discussion content and the core positions of each participant, generating a real-time, quantitative score of opinion dominance. This embodiment employs ridge regression to calculate the influence coefficient of each topic on the opinion dominance of each participant; its regularization form is suitable for situations where there is correlation between independent variables (correlation exists between different topics).

[0071] The method for assessing the capabilities of participants in group discussions provided in this embodiment preprocesses the group discussion text data into a structured event sequence, and then uses a team leadership capability assessment model to comprehensively evaluate the ability to stimulate thinking, coordinate conflicts, and control the discussion, thereby achieving a multi-dimensional dynamic quantitative analysis of team leadership behavior.

[0072] Based on the above embodiments, in this embodiment, step 250, based on the event sequence, assesses the participants' ability to stimulate thinking through the team leadership assessment model, including: Based on the event sequence, the self-motivation intensity parameter and the other-motivation intensity parameter of each participant in the group discussion are calculated using a multidimensional Hawkes process model. The thinking stimulation ability of each participant is evaluated based on the weighted sum of the self-motivation intensity parameter and the other-motivation intensity parameter of each participant.

[0073] Specifically, this embodiment utilizes a multidimensional Hawkes process to quantitatively assess participants' ability to stimulate thought by modeling the event sequence of "a participant uttering a keyword at a certain moment" during team discussions. The specific operational process is as follows: First, the discussion text is processed into an event sequence based on participants, keywords, and time. A knowledge graph containing topic cluster information is constructed using the Louvain community detection algorithm, based on keyword co-occurrence and semantic relationships (obtained through a pre-trained word vector model). Then, a Hawkes process model, where the stimuli intensity depends on this knowledge graph (speaker relationships, keyword semantic distance, and keyword topic relevance), is applied to fit the event sequence. Finally, maximum likelihood estimation and the Powell algorithm are used to solve for the two core parameters: self-motivation intensity and the intensity of motivation from others for each participant. A participant's score for stimulating thought is the sum of their normalized self-motivation intensity and the intensity of motivation from others.

[0074] Step 250, based on the event sequence, assesses the participants' conflict resolution capabilities using the team leadership assessment model, including: A topic distribution vector is generated based on the speaking keywords in the event sequence, and a score is calculated to indicate the dominance of each participant's viewpoint. Using the topic distribution vector as the independent variable and the opinion dominance score of each participant as the dependent variable, a ridge regression model is trained to obtain the influence coefficient. Based on the time-varying influence coefficient and the dominance of the viewpoint, calculate the conflict resolution ability score for each participant; and, The number of open-ended actions by each participant was counted. Each participant's openness ability is assessed based on the total number of openness behaviors they exhibit; and, Support and negation relationships are extracted from the text data, and a participant interaction network graph is constructed based on the support and negation relationships; The collaborative ability of each participant is assessed based on the net support of each participant in the interaction network graph.

[0075] Specifically, this embodiment assesses participants’ conflict coordination ability through three sub-abilities: openness, conflict resolution ability, and collaboration.

[0076] First, in this embodiment, a large model agent is invoked to analyze the complete speech of each participant and count the number of times they perform four types of open behavior: "actively listen", "acknowledge others' views", "analyze the causes of conflict" and "actively promote discussion". The openness is measured by the sum of the number of occurrences of the four types of behavior.

[0077] Secondly, this embodiment evaluates conflict resolution capability through regression analysis based on semantic distance and topic models. The specific operation process is as follows: First, a large model is invoked to automatically identify several core topics from the complete discussion text and extract summaries of the core viewpoints and positions of each participant. Then, the discussion text is divided into multiple segments, obtaining the probability distribution of each segment on each topic (i.e., topic vector) and the score of each participant's viewpoint dominance in this segment. A ridge regression model is trained for each participant, with the topic vector as the independent variable and the viewpoint dominance as the dependent variable, to calculate the influence coefficient of each topic on the viewpoint dominance of each participant. Finally, the negative value of the sum of the viewpoint dominance of all participants at any given time is defined as the on-site conflict level at that time. The on-site conflict level before and after each participant's speech is calculated separately. The final score of a participant's conflict resolution capability is the cumulative reduction in the on-site conflict level brought about by all their speeches.

[0078] Finally, this embodiment calls a large model agent to extract all direct "support" and "negation" interactions from the discussion text, constructing a directed network graph. Nodes in the graph represent participants, support relationships are positive edges (weight +1), and negation relationships are negative edges (weight -1). A participant's collaborative score is calculated by dividing the sum of the weights of all their incoming edges (i.e., the number of times they were supported minus the number of times they were negated) by the total number of interactions in the network, reflecting the net degree to which their viewpoint was accepted by the team.

[0079] Step 250, based on the event sequence, assesses the participants' discussion control ability using the team leadership assessment model, including: Extract the core discussion topic and calculate the semantic distance between each participant's statement and the core topic; The focusing ability of each participant is assessed based on the average semantic distance between each participant's statement and the core topic; and, The number of time-control actions of each participant was counted. Each participant's time management ability is assessed based on the total number of time control behaviors exhibited by the participants; and, Summarize the consensus points discussed, and calculate the change in semantic distance between the discussion content and the consensus points before and after each participant speaks. The convergence capability of each participant is evaluated based on the cumulative reduction in semantic distance resulting from all of their statements.

[0080] Specifically, this embodiment assesses participants' ability to control the discussion through three sub-abilities: focus, time management, and convergence.

[0081] First, this embodiment calls a large model agent to extract the core topic from the discussion content, and uses the SentenceTransformer to calculate the semantic distance between the core topic and each participant's speech. The average semantic distance between all the participants' speeches and the core topic is used to measure the focusing ability.

[0082] Secondly, in this embodiment, a large model agent is invoked to analyze all the speech content of each participant and count the total number of times they mention time-related intentions (reasonable time arrangement, reminders, effective topic control, etc.). The total number of times is used to measure the ability to manage time.

[0083] Finally, this embodiment utilizes a large model agent to summarize the final decision (i.e., the consensus point) of the entire discussion. For each participant's speech, the semantic distance between the discussion content before and after the speech and the consensus point is calculated. The difference between the two distances is the convergence contribution of that speech. The convergence capability is measured by the sum of the semantic distance reductions brought about by all the participants' speeches.

[0084] The method for assessing the capabilities of participants in group discussions provided in this embodiment transforms group discussions into a structured sequence of events. It uses a multidimensional Hawkes process to precisely quantify the participants' thought-stimulation effect, dynamically tracks the conflict resolution process through a topic regression model, and objectively evaluates discussion control effectiveness using semantic analysis. This multimodal approach not only overcomes the shortcomings of traditional assessment methods in characterizing the interaction process but also reconstructs the true picture of leadership behavior from multiple dimensions, including temporal evolution, topic progression, and relationship networks. Ultimately, it achieves an interpretable and quantifiable systematic assessment of core leadership capabilities such as thought stimulation, conflict resolution, and discussion control, providing unprecedented depth of insight for organizational talent selection.

[0085] The following describes the ability assessment device for group discussion participants provided by the present invention. The ability assessment device for group discussion participants described below can be referred to in correspondence with the ability assessment method for group discussion participants described above.

[0086] Figure 2 This is a schematic diagram of the ability assessment device for group discussion participants provided by the present invention, as shown below. Figure 2 As shown, the ability assessment device for group discussion participants provided by the present invention includes: Module 201 is used to acquire text data from group discussions; Evaluation module 202 is used to input the text data into a pre-trained personal general ability evaluation model and a team leadership ability evaluation model, and output personal general ability evaluation results and team leadership ability evaluation results; The individual general ability assessment model is constructed based on decision analysis theory. It is fine-tuned under supervision using a training dataset containing information nodes, decision nodes, and utility nodes, and is trained using a large language model with a low-rank adaptive method. The team leadership ability assessment model is a dynamic interactive assessment model constructed based on a multidimensional Hawkes process. It is used to assess participants' ability to stimulate thinking, coordinate conflicts, and control discussions based on an event sequence containing participant identifiers, speaking keywords, and speaking timestamps.

[0087] The present invention provides a device for assessing the abilities of participants in group discussions. This device acquires text data from group discussions and inputs this text data into pre-trained individual general ability assessment models and team leadership ability assessment models, outputting individual general ability assessment results and team leadership ability assessment results. The individual general ability assessment model is constructed based on decision analysis theory, using a training dataset containing information nodes, decision nodes, and utility nodes for supervised fine-tuning, and trained using a low-rank adaptive method to create a large language model. The team leadership ability assessment model is a dynamic interactive assessment model based on a multidimensional Hawkes process, used to assess participants' thinking stimulation ability, conflict coordination ability, and discussion control ability based on event sequences containing participant identifiers, speaking keywords, and speaking timestamps. Therefore, the present invention utilizes a supervised fine-tuned large language model to parse participant speech, effectively assessing their individual general abilities. Simultaneously, by applying a multidimensional Hawkes process to dynamically model the interactive event sequences between participants, it quantitatively assesses their team leadership abilities. This not only overcomes the subjectivity and bias of manual assessment but also breaks through the bottleneck of traditional text analysis's insufficient capture of dynamic interactive processes.

[0088] Based on the above embodiments, in this embodiment, the device further includes a training module, specifically used for: Obtain a pre-constructed training dataset; wherein the training dataset includes: the text to be analyzed and the corresponding standard nodes; the standard nodes include: information nodes, decision nodes, and utility nodes; Based on the training dataset, the pre-trained large language model is fine-tuned using a low-rank adaptive supervised fine-tuning method to enable the model to identify and extract standard nodes from text, thereby obtaining the personal general ability assessment model. The low-rank adaptive method freezes the main parameters of the pre-trained large language model and injects trainable adapter parameters only into the attention layer to adapt the model to the standard node extraction task.

[0089] Based on the above embodiments, in this embodiment, the evaluation module 202 is specifically used for: Based on a knowledge corpus related to the discussion topic, a reference answer influence graph is constructed using the aforementioned personal general ability assessment model; wherein, the reference answer influence graph includes: information nodes and directed relations, decision nodes and directed relations, and utility nodes and directed relations; Based on the content of the participants' statements in the group discussion, the nodes are extracted and semantically mapped through the personal general ability assessment model, and associated with the standard nodes and directed relationships of the reference answer influence graph to generate a personal influence graph for each participant. The evaluation is based on the node hierarchy distribution in the participant's personal influence graph, generating the personal general ability evaluation result.

[0090] Based on the above embodiments, in this embodiment, the device further includes a construction module, specifically used for: The large language model is invoked to generate decision analysis topics and corresponding core keywords; Based on the decision analysis topic and the corresponding core keywords, real-time background information is obtained through enhanced retrieval generation. The decision analysis topic, its corresponding core keywords, and the real-time background information are integrated to generate the knowledge corpus.

[0091] Based on the above embodiments, in this embodiment, the personal general ability assessment includes at least: systems thinking ability assessment and solution feasibility assessment ability assessment; The evaluation module 202 is specifically used for: By analyzing the abstract hierarchical distribution of nodes in the participant's personal influence diagram, the number of nodes for macro-themes, abstract concepts, and specific information is counted, and the system's thinking ability is evaluated based on the product of the number of nodes for macro-themes, abstract concepts, and specific information. The feasibility assessment capability of the proposed solutions is evaluated by calling a large model intelligent agent to score the decision-making solutions proposed by the participants in multiple dimensions. The multiple dimensions include: awareness of resource constraints, risk identification and management, perception of execution threshold, and dynamic considerations.

[0092] Based on the above embodiments, in this embodiment, the evaluation module 202 is further configured to: The text data of the group discussion is preprocessed into an event sequence; wherein each event in the event sequence includes a participant identifier, speaking keywords, and speaking timestamp; Based on the event sequence, the team leadership assessment model is used to evaluate participants' ability to stimulate thinking, coordinate conflicts, and control discussions.

[0093] Based on the above embodiments, in this embodiment, the evaluation module 202 is specifically used for: Based on the event sequence, the self-motivation intensity parameter and the other-motivation intensity parameter of each participant in the group discussion are calculated using a multidimensional Hawkes process model. The thinking stimulation ability of each participant is evaluated based on the weighted sum of the self-motivation intensity parameter and the other-motivation intensity parameter of each participant.

[0094] Based on the above embodiments, in this embodiment, the conflict coordination capability includes: conflict resolution capability, openness capability, and collaborative capability; The evaluation module 202 is specifically used for: A topic distribution vector is generated based on the speaking keywords in the event sequence, and a score is calculated to indicate the dominance of each participant's viewpoint. Using the topic distribution vector as the independent variable and the opinion dominance score of each participant as the dependent variable, a ridge regression model is trained to obtain the influence coefficient. Based on the time-varying influence coefficient and the dominance of the viewpoint, calculate the conflict resolution ability score for each participant; and, The number of open-ended actions by each participant was counted. Each participant's openness ability is assessed based on the total number of openness behaviors they exhibit; and, Support and negation relationships are extracted from the text data, and a participant interaction network graph is constructed based on the support and negation relationships; The collaborative ability of each participant is assessed based on the net support of each participant in the interaction network graph.

[0095] Based on the above embodiments, in this embodiment, the field control capability discussed includes: focusing capability, timing capability, and convergence capability; The evaluation module 202 is specifically used for: Extract the core discussion topic and calculate the semantic distance between each participant's statement and the core topic; The focusing ability of each participant is assessed based on the average semantic distance between each participant's statement and the core topic; and, The number of time-control actions of each participant was counted. Each participant's time management ability is assessed based on the total number of time control behaviors exhibited by the participants; and, Summarize the consensus points discussed, and calculate the change in semantic distance between the discussion content and the consensus points before and after each participant speaks. The convergence capability of each participant is evaluated based on the cumulative reduction in semantic distance resulting from all of their statements.

[0096] Figure 3 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 3 As shown, the electronic device can be a robot or other electronic device, and may include: a processor 310, a communications interface 320, a memory 330, and a communication bus 340. The processor 310, communications interface 320, and memory 330 communicate with each other via the communication bus 340. The processor 310 can call logical instructions from the memory 330 to execute a method for assessing the capabilities of participants in a group discussion, including: Obtain the text data of the group discussion; The text data is input into a pre-trained personal general ability assessment model and a team leadership ability assessment model, and the personal general ability assessment results and team leadership ability assessment results are output. The individual general ability assessment model is constructed based on decision analysis theory. It is fine-tuned under supervision using a training dataset containing information nodes, decision nodes, and utility nodes, and is trained using a large language model with a low-rank adaptive method. The team leadership ability assessment model is a dynamic interactive assessment model constructed based on a multidimensional Hawkes process. It is used to assess participants' ability to stimulate thinking, coordinate conflicts, and control discussions based on an event sequence containing participant identifiers, speaking keywords, and speaking timestamps.

[0097] Furthermore, the logical instructions in the aforementioned memory 330 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.

[0098] On the other hand, the present invention also provides a computer program product, the computer program product comprising a computer program that can be stored on a non-transitory computer-readable storage medium, wherein when the computer program is executed by a processor, the computer is capable of executing the group discussion participant competence assessment method provided by the above methods, including: Obtain the text data of the group discussion; The text data is input into a pre-trained personal general ability assessment model and a team leadership ability assessment model, and the personal general ability assessment results and team leadership ability assessment results are output. The individual general ability assessment model is constructed based on decision analysis theory. It is fine-tuned under supervision using a training dataset containing information nodes, decision nodes, and utility nodes, and is trained using a large language model with a low-rank adaptive method. The team leadership ability assessment model is a dynamic interactive assessment model constructed based on a multidimensional Hawkes process. It is used to assess participants' ability to stimulate thinking, coordinate conflicts, and control discussions based on an event sequence containing participant identifiers, speaking keywords, and speaking timestamps.

[0099] In another aspect, 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 a method for assessing the competence of participants in a group discussion provided by the methods described above, including: Obtain the text data of the group discussion; The text data is input into a pre-trained personal general ability assessment model and a team leadership ability assessment model, and the personal general ability assessment results and team leadership ability assessment results are output. The individual general ability assessment model is constructed based on decision analysis theory. It is fine-tuned under supervision using a training dataset containing information nodes, decision nodes, and utility nodes, and is trained using a large language model with a low-rank adaptive method. The team leadership ability assessment model is a dynamic interactive assessment model constructed based on a multidimensional Hawkes process. It is used to assess participants' ability to stimulate thinking, coordinate conflicts, and control discussions based on an event sequence containing participant identifiers, speaking keywords, and speaking timestamps.

[0100] 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.

[0101] 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.

[0102] 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 method for assessing the abilities of participants in group discussions, characterized in that, include: Obtain the text data of the group discussion; The text data is input into a pre-trained personal general ability assessment model and a team leadership ability assessment model, and the personal general ability assessment results and team leadership ability assessment results are output. The individual general ability assessment model is constructed based on decision analysis theory. It is fine-tuned under supervision using a training dataset containing information nodes, decision nodes, and utility nodes, and is trained using a large language model with a low-rank adaptive method. The team leadership ability assessment model is a dynamic interactive assessment model constructed based on a multidimensional Hawkes process. It is used to assess participants' ability to stimulate thinking, coordinate conflicts, and control discussions based on an event sequence containing participant identifiers, speaking keywords, and speaking timestamps.

2. The method for assessing the abilities of participants in group discussions according to claim 1, characterized in that, The training process of the personal general ability assessment model includes: Obtain a pre-constructed training dataset; wherein the training dataset includes: the text to be analyzed and the corresponding standard nodes; the standard nodes include: information nodes, decision nodes, and utility nodes; Based on the training dataset, the pre-trained large language model is fine-tuned using a low-rank adaptive supervised fine-tuning method to enable the model to identify and extract standard nodes from text, thereby obtaining the personal general ability assessment model. The low-rank adaptive method freezes the main parameters of the pre-trained large language model and injects trainable adapter parameters only into the attention layer to adapt the model to the standard node extraction task.

3. The method for assessing the abilities of participants in group discussions according to claim 2, characterized in that, The step of inputting the text data into a pre-trained personal general ability assessment model and outputting personal general ability assessment results includes: Based on a knowledge corpus related to the discussion topic, a reference answer influence graph is constructed using the aforementioned personal general ability assessment model; wherein, the reference answer influence graph includes: information nodes and directed relations, decision nodes and directed relations, and utility nodes and directed relations; Based on the content of the participants' statements in the group discussion, the nodes are extracted and semantically mapped through the personal general ability assessment model, and associated with the standard nodes and directed relationships of the reference answer influence graph to generate a personal influence graph for each participant. The evaluation is based on the node hierarchy distribution in the participant's personal influence graph, generating the personal general ability evaluation result.

4. The method for assessing the abilities of participants in group discussions according to claim 3, characterized in that, The construction process of the knowledge corpus includes: The large language model is invoked to generate decision analysis topics and corresponding core keywords; Based on the decision analysis topic and the corresponding core keywords, real-time background information is obtained through enhanced retrieval generation. The decision analysis topic, its corresponding core keywords, and the real-time background information are integrated to generate the knowledge corpus.

5. The method for assessing the abilities of participants in group discussions according to claim 3, characterized in that, The general personal ability assessment includes at least: systems thinking ability assessment and solution feasibility assessment ability assessment; The evaluation, based on the node hierarchy distribution in the participant's personal influence graph, generates the individual general ability evaluation result, including: By analyzing the abstract hierarchical distribution of nodes in the participant's personal influence diagram, the number of nodes for macro-themes, abstract concepts, and specific information is counted, and the system's thinking ability is evaluated based on the product of the number of nodes for macro-themes, abstract concepts, and specific information. The feasibility assessment capability of the proposed solutions is evaluated by calling a large model intelligent agent to score the decision-making solutions proposed by the participants in multiple dimensions. The multiple dimensions include: awareness of resource constraints, risk identification and management, perception of execution threshold, and dynamic considerations.

6. The method for assessing the abilities of participants in group discussions according to claim 1, characterized in that, The step of inputting the text data into a pre-trained team leadership ability assessment model and outputting team leadership ability assessment results includes: The text data of the group discussion is preprocessed into an event sequence; wherein each event in the event sequence includes a participant identifier, speaking keywords, and speaking timestamp; Based on the event sequence, the team leadership assessment model is used to evaluate participants' ability to stimulate thinking, coordinate conflicts, and control discussions.

7. The method for assessing the abilities of participants in group discussions according to claim 6, characterized in that, The assessment of participants' ability to stimulate thinking, based on the event sequence and using the team leadership assessment model, includes: Based on the event sequence, the self-motivation intensity parameter and the other-motivation intensity parameter of each participant in the group discussion are calculated using a multidimensional Hawkes process model. The thinking stimulation ability of each participant is evaluated based on the weighted sum of the self-motivation intensity parameter and the other-motivation intensity parameter of each participant.

8. The method for assessing the abilities of participants in group discussions according to claim 7, characterized in that, The conflict coordination capability includes: conflict resolution capability, openness capability, and collaborative capability; The assessment of participants' conflict resolution capabilities based on the event sequence using the team leadership assessment model includes: A topic distribution vector is generated based on the speaking keywords in the event sequence, and a score is calculated to indicate the dominance of each participant's viewpoint. Using the topic distribution vector as the independent variable and the opinion dominance score of each participant as the dependent variable, a ridge regression model is trained to obtain the influence coefficient. Based on the time-varying influence coefficient and the dominance of the viewpoint, calculate the conflict resolution ability score for each participant; and, The number of open-ended actions by each participant was counted. Each participant's openness ability is assessed based on the total number of openness behaviors they exhibit; and, Support and negation relationships are extracted from the text data, and a participant interaction network graph is constructed based on the support and negation relationships; The collaborative ability of each participant is assessed based on the net support of each participant in the interaction network graph.

9. The method for assessing the abilities of participants in group discussions according to claim 7, characterized in that, The discussion of control over the situation includes: the ability to focus, the ability to manage time, and the ability to converge. The assessment of participants' discussion control abilities based on the event sequence using the team leadership assessment model includes: Extract the core discussion topic and calculate the semantic distance between each participant's statement and the core topic; The focusing ability of each participant is assessed based on the average semantic distance between each participant's statement and the core topic; and, The number of time-control actions of each participant was counted. Each participant's time management ability is assessed based on the total number of time control behaviors exhibited by the participants; and, Summarize the consensus points discussed, and calculate the change in semantic distance between the discussion content and the consensus points before and after each participant speaks. The convergence capability of each participant is evaluated based on the cumulative reduction in semantic distance resulting from all of their statements.

10. A device for assessing the abilities of participants in a group discussion, characterized in that, include: The acquisition module is used to acquire text data from group discussions; The evaluation module is used to input the text data into a pre-trained personal general ability evaluation model and a team leadership ability evaluation model, and output personal general ability evaluation results and team leadership ability evaluation results. The individual general ability assessment model is constructed based on decision analysis theory. It is fine-tuned under supervision using a training dataset containing information nodes, decision nodes, and utility nodes, and is trained using a large language model with a low-rank adaptive method. The team leadership ability assessment model is a dynamic interactive assessment model constructed based on a multidimensional Hawkes process. It is used to assess participants' ability to stimulate thinking, coordinate conflicts, and control discussions based on an event sequence containing participant identifiers, speaking keywords, and speaking timestamps.

11. 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 computer program, it implements the competence assessment method for group discussion participants as described in any one of claims 1 to 9.

12. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the competence assessment method for group discussion participants as described in any one of claims 1 to 9.

Images