Interdisciplinary assignment determination method and system based on multi-agent simulation

Abstract

The application discloses a cross-disciplinary homework determination method and system based on multi-agent simulation, relates to the technical field of text processing, and comprises the following steps: constructing a structure bad homework; generating student agent portrait information, and dividing different behavior information groups according to preset behavior information and the student agent portrait information; simulating conversation processing through a teaching assistant agent and a student agent in a multi-agent simulation system according to the structure bad homework, and obtaining simulation conversation information; cutting the simulation conversation information into multiple collaboration events, classifying the multiple collaboration events, obtaining a classification result, and calculating a core collaboration index according to the simulation conversation information; and when the classification result and the core collaboration index meet preset conditions, determining a to-be-simulated cross-disciplinary group homework as a target cross-disciplinary group homework. According to the application, the judgment on the disciplinary homework according to the experience of relevant technical personnel can be avoided, the quality fluctuation of the homework can be reduced, and the accuracy of the target cross-disciplinary group homework is improved.

Description

Technical Field

[0001] This invention relates to the field of text processing technology, and specifically to a method and system for determining interdisciplinary tasks based on multi-agent simulation. Background Technology

[0002] Interdisciplinary collaboration has become a core training objective of higher education. Problem-based learning is the mainstream teaching model for cultivating this ability. Poorly structured problems are the core characteristics of high-quality interdisciplinary group assignments. Their openness fosters the collision of perspectives from different disciplines and encourages students to construct knowledge through negotiation and integration.

[0003] In related technologies, when constructing interdisciplinary group assignments, if done by technical personnel, the process relies heavily on their experience, leading to significant fluctuations in assignment quality across different disciplines and reducing the accuracy of interdisciplinary group assignments. When using large language model-assisted assignment tools to generate interdisciplinary group assignments, the tools directly generate the assignment text content, failing to facilitate collaboration between students from different majors, thus also reducing the accuracy of interdisciplinary group assignments. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for determining interdisciplinary tasks based on multi-agent simulation, so as to solve the technical problems existing in related technologies.

[0005] To achieve the above objectives, in a first aspect, the present invention provides a method for determining interdisciplinary tasks based on multi-agent simulation, comprising:

[0006] The interdisciplinary group assignments to be simulated are identified, which are learning tasks that students from different majors need to complete collaboratively. The learning tasks include teaching information for each major and its corresponding professional background description information, as well as assignments that each major needs to handle collaboratively.

[0007] Constructing poorly structured assignments based on different teaching information;

[0008] Student intelligent agent profile information is generated based on the professional background description information corresponding to each major. Based on the preset behavioral information and the student intelligent agent profile information, students corresponding to the same major are divided into different behavioral information groups. The preset behavioral information includes multiple behavioral information groups, and each behavioral information group corresponds to one behavioral information. The same behavioral information group includes students corresponding to all majors.

[0009] Based on the poorly structured task, the teaching assistant agent and student agent in the multi-agent simulation system perform simulated dialogue processing on each behavioral information group to obtain the simulated dialogue information of the behavioral information group. The student agent and teaching assistant agent are both obtained through the Big Prophet model.

[0010] The simulated dialogue information is segmented into multiple independent collaborative events, and these collaborative events are classified to obtain classification results. At the same time, core collaborative indicators are calculated based on the simulated dialogue information.

[0011] When the classification results and core collaboration indicators meet the preset conditions, the interdisciplinary group assignment to be simulated will be identified as the target interdisciplinary group assignment.

[0012] Optionally, the construction of poorly structured assignments based on different teaching information includes:

[0013] Semantic parsing is performed on each piece of teaching information to obtain semantic structure features, and teaching features are extracted from the semantic structure features. The teaching features include core theme information, application scenarios, user information, and boundary constraints.

[0014] Construct problem-based contextual information among different majors based on multiple teaching characteristics;

[0015] Extract multidisciplinary perspective information from the problem context information, identify interdisciplinary contradiction and conflict information in the multidisciplinary perspective information, decompose the interdisciplinary contradiction and conflict information into multiple conflict information, and match corresponding scaffold resources for each conflict information. Each conflict information includes subject area information and core contradiction information.

[0016] Multiple conflicting information and their corresponding support resources were identified as structurally flawed operations.

[0017] Optionally, generating student intelligent agent profile information based on the professional background description information corresponding to each major includes:

[0018] The professional background description information is standardized and mapped according to the ISCED-F subject standard database to obtain standardized subject information, and professional competency information is determined based on the standardized subject information.

[0019] Information on professional background descriptions is processed using an interdisciplinary writing competence assessment framework to obtain information on collaborative abilities.

[0020] The professional background information, professional ability information, and collaborative ability information in the professional background description information are identified as the student's intelligent agent profile.

[0021] Optionally, the determination of professional competence information based on standardized subject information includes:

[0022] Based on the target professional competency parameters and the preset correspondence in the standardized subject information, the sub-professional competency information is determined, where the preset correspondence is the correspondence between professional competency parameters and professional competency information;

[0023] In response to a user's modification of a target professional competency parameter, the system determines the professional competency parameter corresponding to the modification operation and determines the professional competency information based on the professional competency parameter and a preset correspondence.

[0024] Optionally, based on the poorly structured task, the teaching assistant agent and student agent in the multi-agent simulation system perform simulated dialogue processing on each behavioral information group to obtain the simulated dialogue information of the behavioral information group, including:

[0025] Determine the target behavioral information of the behavioral information group, and determine the target behavioral label information of the behavioral information group based on the target behavioral information and the behavioral mapping model, wherein the behavioral mapping model is a mapping model between the behavioral information group and the behavioral label information, and the target behavioral information is any one of the behavioral information in the behavioral information group;

[0026] Based on the poorly structured task, the simulation dialogue process of the agent in the behavior information group is controlled by the teaching assistant agent, and the sub-simulation dialogue corresponding to each student in the behavior information group is generated by the student agent based on the target behavior label information.

[0027] The sub-simulation dialogues corresponding to multiple students in the behavior information group are identified as simulation dialogue information.

[0028] Optionally, the simulated dialogue process based on poorly structured tasks, which controls the agents in the behavior information group through the teaching assistant agent, and generates sub-simulated dialogues for each student in the behavior information group based on target behavior label information through the student agent, includes:

[0029] The teaching assistant agent sequentially sets multiple conflicting information and their corresponding scaffolding resources in poorly structured assignments as the current discussion topic, and determines the speaking order of each student in the behavior information group based on the target behavior label information through a dynamic polling strategy.

[0030] Based on the speaking order, a sub-simulation dialogue is generated by the student agent for the current discussion topic, wherein the sub-simulation dialogue includes the dialogue of all students in the behavioral information group;

[0031] When the number of discussions on the current topic reaches the preset number of rounds or when a discussion result is obtained on the current topic, the simulation discussion stops and the sub-simulation dialogue corresponding to each student in the behavior information group is output.

[0032] Optionally, the core collaboration metrics include convergence, integration, adversarial relationship, and cross-group event correlation, wherein the convergence is expressed by the following formula:

[0033] ;

[0034] ;

[0035] in, To measure the convergence of the simulated dialogue information, For the weight function, This represents the number of discussions in the simulated dialogue information. For target behavior label information;

[0036] The integration degree and the adversarial degree are expressed by the following formula:

[0037] ;

[0038] ;

[0039] ;

[0040] ;

[0041] in, For the first The cumulative integration degree of simulated dialogue information. For the first Cumulative adversarial intensity in simulated dialogue information. For smoothing coefficients, For the first The cumulative integration degree of information in round-robin simulation dialogue. This is the basic weight vector corresponding to each behavior label in the target behavior label information. For the first Cumulative adversarial intensity in simulated dialogue information. As the integration weight in the simulated dialogue information, The adversarial weights are used in the simulated dialogue information.

[0042] Optionally, the method further includes:

[0043] When the classification results and core collaboration indicators do not meet the preset conditions, collaboration failure information is generated.

[0044] In response to a user's input operation regarding collaboration failure information, the system determines the target data information corresponding to the input operation and sends the target data information to the server. The server then generates target optimization suggestions based on the target data information and a preset suggestion correspondence. Based on the target optimization suggestions, a new interdisciplinary group assignment to be simulated is generated, along with new classification results and new core collaboration indicators. This process continues until the new classification result is at least one of a positive event or a neutral event, and the new core collaboration indicators meet preset conditions. At this point, the new interdisciplinary group assignment to be simulated is identified as the target interdisciplinary group assignment. The preset suggestion correspondence is the correspondence between data information and optimization suggestions.

[0045] Secondly, the present invention also provides a cross-disciplinary task determination system based on multi-agent simulation, including a first determination module, a poorly structured task construction module, a student profile generation module, a multi-agent collaborative simulation module, a simulation result analysis module, and a second determination module;

[0046] The first determining module is used to determine the interdisciplinary group assignment to be simulated, wherein the interdisciplinary group assignment to be simulated is a learning task that students from different majors need to complete collaboratively, and the learning task includes the teaching information of each major and its corresponding professional background description information and the assignment that each major needs to process collaboratively;

[0047] The poorly structured assignment construction module is used to construct poorly structured assignments based on different teaching information;

[0048] The student profile generation module is used to generate student intelligent agent profile information based on the professional background description information corresponding to each major, and to divide students corresponding to the same major into different behavioral information groups according to the preset behavioral information and the student intelligent agent profile information. The preset behavioral information includes multiple behavioral information groups, and each behavioral information group corresponds to one behavioral information. The same behavioral information group includes students corresponding to all majors.

[0049] The multi-agent collaborative simulation module is used to perform simulated dialogue processing on each behavioral information group according to the poorly structured task through the teaching assistant agent and student agent in the multi-agent simulation system to obtain the simulated dialogue information of the behavioral information group. The student agent and teaching assistant agent are both obtained through the Big Prophet model.

[0050] The simulation result analysis module is used to segment the simulation dialogue information into multiple independent collaborative events, classify the multiple collaborative events according to a preset correspondence, obtain classification results, and calculate core collaborative indicators based on the simulation dialogue information.

[0051] The second determining module is used to determine the interdisciplinary group assignment to be simulated as the target interdisciplinary group assignment when the classification results and core collaboration indicators meet the preset conditions.

[0052] The above technical solution constructs poorly structured assignments based on different teaching information, generates student agent profiles based on the professional background descriptions corresponding to each major, and divides students of the same major into different behavioral information groups according to preset behavioral information and student agent profiles. Based on the poorly structured assignments, the teaching assistant agent and student agent in the multi-agent simulation system conduct simulated dialogue processing on each behavioral information group to obtain simulated dialogue information for the behavioral information group. The simulated dialogue information is classified to obtain classification results, and core collaboration indicators are calculated based on the simulated dialogue information. When the classification results and core collaboration indicators meet preset conditions, the interdisciplinary group assignment to be simulated is determined as the target interdisciplinary group assignment. By simulating dialogues between students using the Big Prophet model and determining the target interdisciplinary group assignment based on the simulated dialogue information, the collision and integration of different disciplinary perspectives and the collaboration process can be simulated. This avoids judging subject assignments based on the experience of relevant technical personnel, reduces assignment quality fluctuations, and improves the accuracy of the target interdisciplinary group assignment.

[0053] Other features and advantages of the present invention will be described in detail in the following detailed description section. Attached Figure Description

[0054] Figure 1 This is a schematic diagram illustrating a cross-disciplinary task determination method based on multi-agent simulation according to an exemplary embodiment of the present invention.

[0055] Figure 2 This is a schematic diagram illustrating the simulation process of a multi-agent simulation system according to an exemplary embodiment of the present invention.

[0056] Figure 3 This is a flowchart illustrating a cross-disciplinary task determination method based on multi-agent simulation according to an exemplary embodiment of the present invention.

[0057] Figure 4 This is a schematic diagram illustrating a platform page according to an exemplary embodiment of the present invention. Detailed Implementation

[0058] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention, so as to provide a better understanding of the concept of the present invention, the technical problem solved, the technical features constituting the technical solution, and the technical effects brought about.

[0059] With the development of large language model technology, academia and industry have begun to explore AI-assisted homework design, mainly in three technical directions: LLM (Large Language Model) content generation tools: These tools automatically generate exercises, homework content, and evaluation criteria through prompting engineering, multi-step pipelines, and domain-specific fine-tuning. Representative solutions include PFQS (Process-Focused Collaboration Quality Score)'s "plan-then-generate" strategy and TwinStar (Twin-Star Simulation Framework)'s "generate-evaluate" dual-engine architecture. Multi-agent simulation in educational scenarios: These tools use LLM-driven agents to simulate learner behavior for teaching exercises and task evaluation. Representative solutions include TutorUp (Tutor Training Simulation System)'s simulated student teaching training system, SimClass (Simulation-Classroom Framework)'s classroom behavior simulation, and VizQStudio (Visual-Quantitative Simulation System). Studio (Visualization-Quantitative Analysis Workbench) provides simulation verification of exercise difficulty; General Domain Multi-Agent Collaboration System: For complex tasks, role-based intelligent agent teams are built to simulate real workflows. Representative solutions include MetaGPT (open source multi-agent software engineering framework) software development intelligent agent framework and ChatDev (chat-driven software development multi-agent framework) software development full-process simulation.

[0060] In the first related technology, interdisciplinary group assignments are mainly generated through LLM assignments. A large language model automatically generates the group assignment text based on the teaching topic, supporting multiple rounds of revisions and adjustments by teachers. The core process is: input assignment topic and requirements → large model generates assignment framework and content → teacher manually modifies and improves → outputs the final solution.

[0061] In the second related technology, interdisciplinary group assignments are mainly generated through student simulation teaching tools. By simulating groups of students at different levels, this assists teachers in training teaching skills or assessing the difficulty of exercises. Its core process is: configuring student ability profiles → starting the simulation → outputting simulation behavior statistics → teachers adjusting teaching content based on the results.

[0062] However, the inventors discovered that when generating interdisciplinary group assignments using the methods in the first related technology, these tools, focusing only on content output and not simulating the assignment implementation process, cannot predict the collaborative effect of the assignment in the classroom. Teachers cannot use the tools to know whether the designed assignment will lead to an imbalance in participation from students of different majors, whether it will generate valuable disciplinary conflicts, or whether the supporting resources can effectively support students in solving problems. Ultimately, they can only rely on experience to judge, resulting in large fluctuations in assignment quality and delayed problem exposure, making it impossible to correct before deployment. When generating interdisciplinary group assignments using the methods in the second related technology, existing simulation tools are mainly geared towards teaching and training or exercise assessment in a single discipline and do not model the core characteristics of poorly structured interdisciplinary assignments. Therefore, they cannot reproduce the collaborative process driven by disciplinary conflicts. They cannot simulate the collision and integration of different disciplinary perspectives, nor can they analyze the correlation between conflict depth, resource usage, and collaboration quality. They also lack supporting assignment optimization and version management functions and cannot directly support the entire process design of interdisciplinary group assignments. Furthermore, neither method has established a closed-loop design system that allows for the quantification of structural defects, the preview of collaborative processes, and the traceability of design iterations. Teachers are unable to accurately calibrate the openness of assignments through tools, nor can they data-drivenly evaluate and optimize the interdisciplinary teaching value of assignments.

[0063] In view of this, the present invention provides a method and system for determining interdisciplinary tasks based on multi-agent simulation, in order to solve the technical problems existing in the above-mentioned related technologies.

[0064] like Figure 1 As shown, Figure 1 This is a schematic diagram illustrating a cross-disciplinary task determination method based on multi-agent simulation according to an exemplary embodiment of the present invention, with reference to... Figure 1 The method includes:

[0065] S101: Determine the interdisciplinary group assignment to be simulated, wherein the interdisciplinary group assignment to be simulated is a learning task that students from different majors need to complete collaboratively, and the learning task includes the teaching information of each major and its corresponding professional background description information and the assignment that each major needs to process collaboratively.

[0066] S102: Constructing poorly structured assignments based on different teaching information;

[0067] S103: Generate student agent profile information based on the professional background description information corresponding to each major, and divide students corresponding to the same major into different behavior information groups according to the preset behavior information and student agent profile information. The preset behavior information includes multiple behavior information groups, and each behavior information group corresponds to one behavior information. The same behavior information group includes students corresponding to all majors.

[0068] S104: Based on the poorly structured operation, the teaching assistant agent and the student agent in the multi-agent simulation system perform simulated dialogue processing on each behavioral information group to obtain the simulated dialogue information of the behavioral information group. The student agent and the teaching assistant agent are both obtained through the big oracle model.

[0069] S105: Segment the simulated dialogue information into multiple independent collaborative events, classify the multiple collaborative events to obtain the classification results, and calculate the core collaborative indicators based on the simulated dialogue information.

[0070] S106: When the classification results and core collaboration indicators meet the preset conditions, the interdisciplinary group assignment to be simulated will be identified as the target interdisciplinary group assignment.

[0071] The above technical solution constructs poorly structured assignments based on different teaching information, generates student agent profiles based on the professional background descriptions corresponding to each major, and divides students of the same major into different behavioral information groups according to preset behavioral information and student agent profiles. Based on the poorly structured assignments, the teaching assistant agent and student agent in the multi-agent simulation system conduct simulated dialogue processing on each behavioral information group to obtain simulated dialogue information for the behavioral information group. The simulated dialogue information is classified to obtain classification results, and core collaboration indicators are calculated based on the simulated dialogue information. When the classification results and core collaboration indicators meet preset conditions, the interdisciplinary group assignment to be simulated is determined as the target interdisciplinary group assignment. By simulating dialogues between students using the Big Prophet model and determining the target interdisciplinary group assignment based on the simulated dialogue information, the collision and integration of different disciplinary perspectives and the collaboration process can be simulated. This avoids judging subject assignments based on the experience of relevant technical personnel, reduces assignment quality fluctuations, and improves the accuracy of the target interdisciplinary group assignment.

[0072] To enable those skilled in the art to better understand the interdisciplinary task determination method based on multi-agent simulation provided by this invention, the above steps are illustrated in detail below.

[0073] For example, the simulated interdisciplinary group assignment can be a learning task requiring collaboration from students with different disciplinary backgrounds. Its core objective is to cultivate interdisciplinary collaboration skills through the collision of disciplinary perspectives and knowledge integration. Therefore, this simulated interdisciplinary group assignment is geared towards students from different majors and requires collaborative work from students of different disciplines. Consequently, this simulated interdisciplinary group assignment can include teaching information for each major, a description of the corresponding professional background, and the collaborative tasks that students from each major need to complete. The teaching information can include teaching materials for the major, and the professional background description can include an introduction to the major and information on the direction of problem-solving.

[0074] In this embodiment of the invention, the interdisciplinary group assignment to be simulated can be a pre-set assignment of assignment tasks to students of different majors before the teacher assigns relevant learning tasks to students of different majors.

[0075] For example, after determining the interdisciplinary group assignment to be simulated, poorly structured assignments can be constructed based on the teaching information in the interdisciplinary group assignments. Among them, poorly structured assignments can be problems that lack clear initial and target states, do not have a unique optimal solution, and have multiple solution paths, and are the core task carriers of interdisciplinary collaborative learning.

[0076] Among possible approaches, the construction of poorly structured assignments based on different teaching information includes:

[0077] Semantic parsing is performed on each piece of teaching information to obtain semantic structure features, and teaching features are extracted from the semantic structure features. The teaching features include core theme information, application scenarios, user information, and boundary constraints.

[0078] Construct problem-based contextual information among different majors based on multiple teaching characteristics;

[0079] Extract multidisciplinary perspective information from the problem context information, identify interdisciplinary contradiction and conflict information in the multidisciplinary perspective information, decompose the interdisciplinary contradiction and conflict information into multiple conflict information, and match corresponding scaffold resources for each conflict information. Each conflict information includes subject area information and core contradiction information.

[0080] Multiple conflicting information and their corresponding support resources were identified as structurally flawed operations.

[0081] It should be understood that semantic parsing processing is a process of semantic understanding and information extraction for free text dialogue logs generated by multi-agent simulation. In this embodiment of the invention, for the teaching information corresponding to each major, semantic parsing processing is performed on the teaching information to obtain semantic structure features, which can be structured labeled data information. Then, teaching features can be extracted from the semantic structure features, which can include core theme information, application scenarios, user information, and boundary constraints. Among them, core theme information can be the task theme, objectives, task requirements, and interaction specifications that the interdisciplinary group needs to complete collaboratively; application scenarios can be the environment and scenario settings in which the simulation takes place, which can include the group organization form, interaction mode, task background, and environmental conditions; user information can be all attribute information of each virtual student subject participating in the collaboration, which can include professional discipline background, professional ability parameters, consensus integration-collaboration-communication three-dimensional collaborative ability levels, and individual characteristics standardized by ISCED-F; boundary constraints can be the hard restrictions imposed by the system on the agents. By controlling the basic openness of the assignment through the leniency of the constraints, it is ensured that the task has no fixed standard answer and requires multidisciplinary perspectives to complete collaboratively, which meets the core characteristics of poorly structured assignments.

[0082] After processing the teaching information corresponding to all majors, multiple teaching features can be obtained. Based on these features, problem context information between different majors can be constructed. This problem context information could be: "You are commissioned by an organization to design a remote medical digital platform for the rural elderly population. However, the local network infrastructure is weak, and there is a lack of comprehensive legislation specifically for medical data protection. You need to design a solution and develop a mobile health application that supports voice interaction, which must comply with legal and ethical standards." After obtaining the problem context information, multidisciplinary perspectives can be extracted. Then, potential interdisciplinary conflicts can be automatically identified and broken down into multiple independent, editable conflict information pieces. Each conflict information piece is labeled with the relevant discipline and core conflict. Conflict information is the core adjustment unit for poorly structured tasks: the more conflict information there is and the higher the conflict intensity, the higher the degree of structural malformation in the simulated interdisciplinary group task; conversely, the lower the degree of structural malformation, the less conflict information there is.

[0083] Next, corresponding scaffolding resources are matched to each conflict, supporting students in problem-solving without unduly restricting their exploration space. These scaffolding resources are supplementary learning materials that support students in solving problems and resolving conflicts, guiding them forward without limiting their exploration. First, a web search interface is used to retrieve learning resources such as academic materials, industry standards, case studies, and research reports related to the core theme of the conflict. The search results are then filtered for applicability, and explanations are generated to justify resource selection, explaining why each resource can support the resolution of the corresponding conflict. Resources are bound to conflict information one-to-one. When conflict information is adjusted, the system automatically assesses the suitability of existing resources and provides replacement suggestions for resources with insufficient suitability. Teachers can manually add, delete, and edit resource content, adjusting the openness of resources to achieve fine-grained control over the scaffolding support. Finally, multiple conflict pieces of information and their corresponding scaffolding resources are identified as poorly structured assignments.

[0084] In another possible approach, conflict information can be displayed on the platform's page, allowing teachers to adjust it. When teachers see conflict information on the platform that needs modification, they can add, delete, or modify the content or intensity of the conflict. The background description and constraints in the problem context information can be automatically adjusted synchronously to ensure the overall logical consistency of the simulated interdisciplinary group assignment. This avoids the conflict from being disconnected from the context and maintains the two-way logical connection between conflict information and problem context information.

[0085] For example, after obtaining the professional background description information corresponding to each major, corresponding student intelligent agent profile information can be generated for each professional background description information. The student intelligent agent profile information can be a structured object composed of four core attributes: student name, major, grade level, and collaborative ability description. Preset behavioral information can include multiple behavioral information groups. In this embodiment of the invention, the multiple behavioral information groups can include a deficient group, a qualified group, and an excellent group. Each behavioral information group corresponds to relevant behavioral information. For example, the behavioral information corresponding to the deficient group may be that the analysis of disciplinary insights is relatively superficial, extreme positions are prone to occur in collaboration, there are obvious terminology barriers in cross-disciplinary communication, and it is difficult to proactively reach a consensus. The behavioral information corresponding to the qualified group may be that they can analyze the advantages and disadvantages of various disciplinary insights, have a collaborative and compromising awareness, and can try to find consensus, but it is difficult to independently complete knowledge integration under complex conflicts. The behavioral information corresponding to the excellent group may be that they can critically analyze the perspectives of various disciplines, proactively integrate different viewpoints through negotiation, adapt to non-professional audiences during communication, and efficiently promote consensus.

[0086] In one possible manner, generating student intelligent agent profile information based on the professional background description information corresponding to each major includes:

[0087] The professional background description information is standardized and mapped according to the ISCED-F subject standard database to obtain standardized subject information, and professional competency information is determined based on the standardized subject information.

[0088] Information on professional background descriptions is processed using an interdisciplinary writing competence assessment framework to obtain information on collaborative abilities.

[0089] The professional background information, professional ability information, and collaborative ability information in the professional background description information are identified as the student's intelligent agent profile.

[0090] It should be understood that the ISCED-F subject standard library (International Standard Classification of Education Fields of Education and Training) can be used to support the automatic standardization of subject names and professional competency descriptions; it also has a built-in interdisciplinary collaboration competency assessment scale, which can generate student group profiles at three competency levels, and supports manual parameter adjustment by teachers. First, the professional competency dimension is initialized. After obtaining the professional background description information, name standardization mapping can be performed based on the ISCED-F subject standard library to ensure the accuracy and consistency of subject definitions, thereby obtaining standardized subject information. Then, professional competency information can be determined based on this standardized subject information.

[0091] Then, the abilities of students corresponding to different majors can be set to a high level to control the professional ability variable and focus on the impact of collaboration ability on the assignment effect. At the same time, teachers can manually lower the professional ability parameters of corresponding students. Next, the interdisciplinary collaboration ability dimension is stratified. Based on the interdisciplinary collaboration ability assessment framework, the professional background description information is processed to obtain collaboration ability information. The interdisciplinary collaboration ability assessment framework can be a stratified scale containing three primary indicators: interdisciplinary knowledge integration ability, communication and expression ability, and teamwork awareness. Each indicator is set with three levels of "excellent, qualified, and insufficient" for operable behavioral descriptions, and customized expressions are made in combination with the professional context. Finally, the student agent is divided into three groups of behavioral information at three ability levels, and the interdisciplinary collaboration ability assessment framework can be obtained through existing technologies.

[0092] After obtaining information on professional skills and collaborative abilities, the professional background information, professional skills information, and collaborative ability information in the professional background description information can be identified as the student's intelligent agent profile.

[0093] In possible ways, the determination of professional competence information based on standardized subject information includes:

[0094] Based on the target professional competency parameters and the preset correspondence in the standardized subject information, the sub-professional competency information is determined, where the preset correspondence is the correspondence between professional competency parameters and professional competency information;

[0095] In response to a user's modification of a target professional competency parameter, the system determines the professional competency parameter corresponding to the modification operation and determines the professional competency information based on the professional competency parameter and a preset correspondence.

[0096] It should be understood that when determining professional competency information, it is also possible to obtain professional competency information by having users modify professional competency parameters, where users can be teachers. A preset correspondence can be established between professional competency information and professional competency parameters. The professional competency information can be a three-tiered textual description of professional competency based on ISCED-F standardized subject coding and customized according to the relevant professional context. The professional competency parameters can be the selection weights for the three levels of competency descriptions. First, the target professional competency parameters can be extracted from the standardized subject information. After obtaining the target professional competency parameters, sub-professional competency information can be determined based on the target professional competency parameters and the preset correspondence. When users see that some information in the sub-professional competency information needs adjustment, they can modify the parameters in the target professional competency parameters, and then match the professional competency information based on the modified professional competency parameters. This can improve the stratification of interdisciplinary collaboration capabilities, thereby improving the accuracy of target interdisciplinary group assignments.

[0097] For example, multi-agent simulation can be a technology that simulates real group behavior through multiple agents with autonomous decision-making capabilities. In this invention, it is used to reproduce the interdisciplinary collaboration process of students with different ability levels. After obtaining poorly structured assignments and behavioral information groups, simulated dialogues can be conducted between students in the behavioral information groups based on the poorly structured assignments using a multi-agent simulation system. Specifically, simulated dialogue processing can be performed on each behavioral information group by teaching assistant agents and student agents in the multi-agent simulation system to obtain simulated dialogue information. The teaching assistant agent can be a virtual teaching assistant role in the simulation environment, mainly responsible for process control, order maintenance, appropriate guidance, conflict mediation, and task reminders. The student agent can be a virtual student individual simulating a real interdisciplinary course, and is the core subject of the group collaboration simulation. Both the teaching assistant agent and the student agent are obtained through a large language model. Furthermore, in subsequent steps, the reasonableness of the allocation of interdisciplinary group assignments to be simulated can be judged based on the simulated dialogue information.

[0098] In one possible manner, the step of processing simulated dialogue for each behavioral information group based on the poorly structured task through a teaching assistant agent and a student agent in a multi-agent simulation system to obtain simulated dialogue information for the behavioral information group includes:

[0099] Determine the target behavioral information of the behavioral information group, and determine the target behavioral label information of the behavioral information group based on the target behavioral information and the behavioral mapping model, wherein the behavioral mapping model is a mapping model between the behavioral information group and the behavioral label information, and the target behavioral information is any one of the behavioral information in the behavioral information group;

[0100] Based on the poorly structured task, the simulation dialogue process of the agent in the behavior information group is controlled by the teaching assistant agent, and the sub-simulation dialogue corresponding to each student in the behavior information group is generated by the student agent based on the target behavior label information.

[0101] The sub-simulation dialogues corresponding to multiple students in the behavior information group are identified as simulation dialogue information.

[0102] It should be understood that the behavior mapping model can be a mapping model between behavior information groups and behavior label information. Specifically, when the behavior information group is an excellent group, the corresponding behavior label information can be that coordination behavior and solution behavior have the highest trigger probability, which can actively guide the direction of discussion, while opposition behavior has a low trigger probability, and opposition is usually accompanied by reasoning. When the behavior information group is a qualified group, the corresponding behavior label information can be that coordination behavior has a medium trigger probability, reasoning behavior is triggered first when facing complex conflicts, and solution behavior can be triggered as the discussion progresses. When the behavior information group is a poor group, the corresponding behavior label information can be that opposition behavior has the highest trigger probability, coordination behavior is rarely triggered actively, unconditional agreement behavior is easily triggered in simple conflicts, and reasoning and solution behavior need to be triggered by resource allocation in complex conflicts.

[0103] Behavioral labeling information consists of behavioral labels and corresponding quantitative information. Behavioral labels include five categories: coordination, reasoning, agreement, disagreement, and solutions. Specifically, when the behavioral label is coordination, the corresponding quantitative information can be statements that involve negotiating the current conflict, adjusting the pace of discussion, advocating for perspective-taking, and promoting consensus. When the behavioral label is reasoning, the corresponding quantitative information can be statements that cite supporting resources to back up one's own viewpoint, demonstrate one's position, refute dissent, and conduct logical deduction. When the behavioral label is agreement, the corresponding quantitative information can be statements that express approval, agreement, or support for others' viewpoints or solutions. When the behavioral label is disagreement, the corresponding quantitative information can be statements that express disagreement, questioning, or rejection of others' viewpoints or solutions. When the behavioral label is solutions, the corresponding quantitative information can be statements that propose integrated or compromising solutions to attempt to resolve the current conflict.

[0104] Therefore, we can first determine which type of target behavior information the behavior information group belongs to, and then determine the target behavior label information based on the target behavior information and the behavior mapping model. Afterwards, the teaching assistant agent can control the simulated dialogue flow of the agents in the behavior information group, and the student agent can generate sub-simulated dialogues corresponding to each student in the behavior information group. All the sub-simulated dialogues corresponding to all students are then identified as the simulated dialogue information.

[0105] In one possible manner, the simulated dialogue process based on poorly structured tasks, whereby the teaching assistant agent controls the agents in the behavior information group, and the student agent generates sub-simulated dialogues corresponding to each student in the behavior information group based on target behavior label information, includes:

[0106] The teaching assistant agent sequentially sets multiple conflicting information and their corresponding scaffolding resources in poorly structured assignments as the current discussion topic, and determines the speaking order of each student in the behavior information group based on the target behavior label information through a dynamic polling strategy.

[0107] Based on the speaking order, a sub-simulation dialogue is generated by the student agent for the current discussion topic, wherein the sub-simulation dialogue includes the dialogue of all students in the behavioral information group;

[0108] When the number of discussions on the current topic reaches the preset number of rounds or when a discussion result is obtained on the current topic, the simulation discussion stops and the sub-simulation dialogue corresponding to each student in the behavior information group is output.

[0109] It should be understood that, such as Figure 2 As shown, the teaching assistant agent is responsible for global scheduling. During simulation, it can perform parallel simulation of multiple behavioral information groups or simulate a single behavioral information group. In this embodiment of the invention, taking the simulation dialogue of a single behavioral information group as an example, the teaching assistant agent can fully demonstrate the interaction logic of the three stages: free discussion, solution presentation, and conflict switching.

[0110] In the specific processing, multiple conflict information points and their corresponding scaffolding resources from poorly structured assignments are extracted from the teaching assistant agent. These conflict information points are then sequentially set as the current discussion topics to simulate discussions for each conflict information point. Subsequently, based on the target behavior label information, the speaking order of each student in the behavior information group can be determined in the teaching assistant agent using a dynamic polling strategy, thereby avoiding a single subject dominating the discussion.

[0111] Based on the speaking order, the student agent can generate a sub-simulation dialogue for the current discussion topic and simultaneously update the speaking content to the global dialogue context. During the discussion, when the discussion on the current topic reaches a preset number of rounds or when a clear solution is found, the simulation discussion can be stopped and the sub-simulation dialogue can be output.

[0112] In another possible approach, when a discussion outcome is reached, the teaching assistant agent determines that the discussion has entered the solution presentation phase. All students are then invited to express their opinions on the current discussion topic. If all students in the behavior information group agree, indicating a unified discussion, the conflict information corresponding to the current discussion topic is considered resolved. The teaching assistant agent automatically switches the discussion topic to the next conflict information and re-enters the discussion phase. If any student in the behavior information group objects, the discussion continues. The simulation discussion can stop and output a sub-simulation dialogue when the discussion on the current topic reaches a preset number of rounds or a clear solution is reached. During the simulation using both the teaching assistant agent and the student agent, the entire simulation synchronously records information such as the speaker, their major, target behavior tag, speech content, whether scaffold resources were used, and the resource number used, outputting complete simulation dialogue information.

[0113] For example, such as Figure 3 As shown, after obtaining the simulated dialogue information, the continuous dialogue flow in the simulated dialogue information can be divided into multiple independent collaborative events. These collaborative events are then classified according to their occurrences. The Tavily API is used to search for scaffold resources. For example, if a collaborative event achieves a solution quickly through coordination and communication with minimal reasoning and argumentation, it is classified as a positive event. If a collaborative event achieves a solution through repeated reasoning and resource mobilization, it is classified as a neutral event. If a collaborative event fails to reach a valid consensus, resulting in a collaborative deadlock, and the point of disagreement is ultimately skipped or shelved, it is classified as a negative event. Simultaneously, core collaboration indicators can be calculated based on the simulated dialogue information. These core collaboration indicators include convergence, integration, antagonism, and cross-group event correlation. Integration and antagonism are two-dimensional core indicators used to measure the quality of interdisciplinary collaboration. Integration reflects the degree of knowledge fusion and consensus achievement, while antagonism reflects the intensity of conflict between disciplinary viewpoints.

[0114] Convergence, used to measure the degree of focus of discussion within a single event. It iterates through the target behavior label information of all dialogues within the event, accumulating scores according to preset weights. This is expressed by the following formula:

[0115] ;

[0116] ;

[0117] in, To measure the convergence of the simulated dialogue information, For the weight function, This represents the number of discussions in the simulated dialogue information. The target behavior is labeled with information; a positive convergence value indicates that the discussion is approaching consensus, while a negative value indicates that the discussion is diverging, and the absolute value reflects the degree of focus or divergence.

[0118] In calculating integration and antagonism, the Exponential Moving Average (EMA) algorithm can be used for smoothing, avoiding drastic fluctuations in values ​​caused by a single round of speaking. This can be expressed by the following formula:

[0119] ;

[0120] ;

[0121] ;

[0122] ;

[0123] in, For the first The cumulative integration degree of information in round-robin simulation dialogue. For the first Cumulative adversarial intensity in simulated dialogue information. For smoothing coefficients, For the first The cumulative integration degree of information in round-robin simulation dialogue. This is the basic weight vector corresponding to each behavior label in the target behavior label information. For the first Cumulative adversarial intensity in simulated dialogue information. As the integration weight in the simulated dialogue information, The adversarial weights are used in the simulated dialogue information.

[0124] Cross-group event relevance measures the semantic relevance of discussions on the same conflict between different capability groups. It can be achieved by fusing three basic metrics: string similarity, term frequency-inverse document frequency cosine similarity, and keyword matching bonus. When the basic similarity exceeds a threshold, a large language model is additionally invoked for deep semantic scoring, and the final relevance score is obtained through segmented weighted summation. A higher score indicates that the discussions of the conflicting information by different groups are more similar; a lower score indicates greater disagreement among the groups.

[0125] For example, after obtaining the classification results and core collaboration indicators, the target interdisciplinary group assignment can be determined based on these results. Specifically, when the classification results and core collaboration indicators meet preset conditions, the interdisciplinary group assignment to be simulated can be determined as the target interdisciplinary group assignment. These preset conditions include: convergence of at least 30, integration of at least 4.5, and adversarial degree of at least 5.5; the proportion of positive and neutral events in the event classification results being at least 75% of all collaborative events, and negative events accounting for no more than 25%; the "propose solution" tag appearing at least once during the dialogue, and the combined proportion of "negotiation" and "agreement" tags being at least as high as the combined proportion of "reasoning" and "opposition" tags.

[0126] In some possible ways, the method further includes:

[0127] When the classification results and core collaboration indicators do not meet the preset conditions, collaboration failure information is generated.

[0128] In response to a user's input operation regarding collaboration failure information, the system determines the target data information corresponding to the input operation and sends the target data information to the server. The server then generates target optimization suggestions based on the target data information and a preset suggestion correspondence. Based on the target optimization suggestions, a new interdisciplinary group assignment to be simulated is generated, and new classification results and new core collaboration indicators are generated based on the new interdisciplinary group assignment to be simulated. When the new classification results and core collaboration indicators meet preset conditions, the new interdisciplinary group assignment to be simulated is identified as the target interdisciplinary group assignment. The preset suggestion correspondence is the correspondence between data information and optimization suggestions.

[0129] It should be understood that when the classification results and core collaboration indicators do not meet the preset conditions, it can indicate that the interdisciplinary group work to be simulated is mismatched, and changes can be made accordingly.

[0130] Specifically, when the classification results and core collaboration indicators fail to meet preset conditions, it can indicate a negative event in the classification results. Collaboration failure information can be generated using three dimensions: participation imbalance, conflict intensity, and resource support. The participation imbalance dimension primarily calculates the percentage of students from different majors speaking. If the percentage of a single major exceeds a threshold, it is judged as superficial consensus or discussion deviation caused by participation imbalance. The conflict intensity dimension primarily calculates the average antagonism and integration degree of the corresponding conflict. If the antagonism degree consistently exceeds the threshold and the integration degree is below the threshold, it is judged as a collaboration deadlock caused by excessive conflict intensity. The resource support dimension primarily tracks the resource usage and timing under the corresponding conflict. If the resource usage is low and the percentage of negative events is high, it is judged as discussion stagnation caused by insufficient scaffolding resources.

[0131] After obtaining collaboration failure information, it can be displayed on the relevant platform page, allowing users to analyze it. Specifically, after observing collaboration failure information, users can analyze it and match corresponding target data information to adjust the interdisciplinary group assignment to be simulated. Specifically, the final comprehensive results of collaboration failure information can include deadlocks caused by excessive conflict intensity, stagnation of discussion due to insufficient resource support, and superficial consensus caused by imbalanced participation.

[0132] like Figure 4 As shown, when collaboration failure information is displayed on the platform page, users can input the target information they want to modify based on the information displayed in the simulation view and feedback view on the platform page. The target information can then be sent to the server, which generates optimization suggestions based on the target information and a preset suggestion mapping. The preset optimization suggestion mapping is as follows: when the final comprehensive result of the collaboration failure information is a stalemate caused by excessive conflict intensity, the optimization suggestion is to adjust the conflict description, lower the entry threshold, split secondary conflicts, or supplement background constraints, while providing the revised complete conflict text; when the final comprehensive result of the collaboration failure information is a discussion stagnation caused by insufficient resource support, the optimization suggestion is to call a professional search interface to retrieve matching scaffolding resources, mark the applicable scenarios and expected functions of the resources, and add them to the resource list; when the final comprehensive result of the collaboration failure information is a superficial consensus caused by unbalanced participation, the optimization suggestion is to adjust the conflict design, add task anchors for weaker majors, force students from different majors to participate deeply, and provide specific scenario modification solutions.

[0133] Then, based on the target optimization suggestions, new interdisciplinary group assignments to be simulated can be generated, and the above steps can be repeated to process the new interdisciplinary group assignments to be simulated, so as to obtain new classification results and new core collaboration indicators. When the new classification results and core collaboration indicators meet the preset conditions, the new interdisciplinary group assignments to be simulated are identified as the target interdisciplinary group assignments.

[0134] In the specific implementation process, the "Design of a Smart Home Health Monitoring System for Elderly Users" was set as an interdisciplinary group assignment to be simulated, and the specific implementation process and beneficial effects of the present invention were explained in detail.

[0135] Teachers upload teaching information such as smart home technology documents and elderly health care guidelines, describing the group as consisting of students from three majors: computer science, humanities, and law. First, student agent profiles are generated for the three behavioral information groups. Then, problem scenarios are constructed based on the three majors, identifying core conflicting information such as "technical feasibility vs. user acceptance," "data accuracy vs. privacy ethics," and "feature richness vs. ease of use." Corresponding scaffolding resources, including technical standards, elderly psychology research, and medical ethics guidelines, are matched to generate an initial V1.0 version of the assignment.

[0136] The teacher initiated the simulation, running three groups of student agents through a multi-agent simulation system. After the simulation, the results showed that the excellent and qualified groups successfully resolved all conflicts; however, the subpar group experienced a negative event in the "data accuracy vs. privacy ethics" conflict; resource statistics showed that ethical resources were accessed very infrequently under this conflict. When the teacher clicked on this negative event, the dialogue view automatically located the corresponding discussion segment, confirming that the students' lack of knowledge about the ethical assessment framework caused the discussion to remain at a superficial level, failing to develop an effective solution.

[0137] The teacher selects the conflict information and scaffolding resources, inputs the target data information, selects the materials, and clicks "optimization recommendation." The system diagnoses the cause and recommends the resource "popular science on the ethics of privacy in elderly medical data," suggesting that the conflict description be adjusted to a more specific issue closer to the application scenario to lower the discussion threshold. After the teacher accepts the suggestion, the system generates assignment version V1.1. The teacher restarts the simulation verification, and the results show that the negative events under this conflict in the insufficient group disappear and turn into neutral events, achieving the ideal state of "conflict but integration." The teacher confirms that this poorly structured assignment version is suitable for students of different abilities, exports the final assignment plan, and obtains the target interdisciplinary group assignment.

[0138] This example demonstrates that the present invention can expose problems in assignment design through simulation in advance and provide targeted optimization solutions, helping teachers to accurately calibrate the degree of structural defects in assignments. While ensuring interdisciplinary depth, it avoids students getting stuck in collaborative deadlock, significantly improving the design quality and teaching adaptability of interdisciplinary group assignments.

[0139] Based on the same concept, this invention also discloses a cross-disciplinary task determination system based on multi-agent simulation, including a first determination module, a poorly structured task construction module, a student profile generation module, a multi-agent collaborative simulation module, a simulation result analysis module, and a second determination module;

[0140] The first determining module is used to determine the interdisciplinary group assignment to be simulated, wherein the interdisciplinary group assignment to be simulated is a learning task that students from different majors need to complete collaboratively, and the learning task includes the teaching information of each major and its corresponding professional background description information and the assignment that each major needs to process collaboratively;

[0141] The poorly structured assignment construction module is used to construct poorly structured assignments based on different teaching information;

[0142] The student profile generation module is used to generate student intelligent agent profile information based on the professional background description information corresponding to each major, and to divide students corresponding to the same major into different behavioral information groups according to the preset behavioral information and the student intelligent agent profile information. The preset behavioral information includes multiple behavioral information groups, and each behavioral information group corresponds to one behavioral information. The same behavioral information group includes students corresponding to all majors.

[0143] The multi-agent collaborative simulation module is used to perform simulated dialogue processing on each behavioral information group according to the poorly structured task through the teaching assistant agent and student agent in the multi-agent simulation system to obtain the simulated dialogue information of the behavioral information group. The student agent and teaching assistant agent are both obtained through the Big Prophet model.

[0144] The simulation result analysis module is used to segment the simulation dialogue information into multiple independent collaborative events, classify the multiple collaborative events according to a preset correspondence, obtain classification results, and calculate core collaborative indicators based on the simulation dialogue information.

[0145] The second determining module is used to determine the interdisciplinary group assignment to be simulated as the target interdisciplinary group assignment when the classification results and core collaboration indicators meet the preset conditions.

[0146] The above technical solution abstracts the fuzzy interdisciplinary group assignments to be simulated into independently editable disciplinary conflict information. A linkage modification mechanism is established around this conflict information, linking the problem context, disciplinary conflicts, and scaffolding resources. By adjusting the number of conflict points, the intensity of disputes, and boundary constraints, the degree of structural defects in the assignments can be finely and quantitatively calibrated, solving the problem of experience-dependent openness and lack of quantitative control in traditional designs. A dual-dimensional student agent profile information of "professional ability + interdisciplinary collaboration ability" is constructed, establishing five behavioral labels: coordination, reasoning, agreement, opposition, and solutions, as well as a probability distribution mapping model between behavioral labels and collaboration ability levels. A centralized architecture with centralized scheduling of teaching assistant agents supports parallel simulation of three student teams with different ability levels, fully reproducing the entire process of interdisciplinary collaboration driven by disciplinary conflicts, and enabling pre-deployment rehearsals of assignment effectiveness. Based on collaborative event segmentation, three core collaboration indicators are constructed: convergence, integration and antagonism, and cross-group event correlation. An exponential moving average algorithm is used to smooth the integration and antagonism values ​​round by round. Combined with event type classification and multi-dimensional statistical data, a structured and computable representation of the interdisciplinary collaboration process and quality is achieved, providing a quantitative basis for task optimization. Based on negative events and anomaly indicators from simulation output, the root cause diagnosis of collaboration failure is automatically completed from three dimensions: participation balance, conflict intensity, and resource support. Correspondingly, three directly reusable optimization schemes are generated: conflict text adjustment, scaffold resource supplementation, and participation structure optimization. This forms a traceable closed-loop iterative system of design, simulation, analysis, and optimization, enabling continuous optimization of data-driven task schemes.

[0147] The above technical solutions can overcome the limitations of interdisciplinary assignment design relying on experience and post-hoc revisions. By using multi-agent simulation to evaluate assignment effectiveness upfront, teachers can accurately calibrate the degree of structural deficiencies before deployment, effectively balancing the openness and completeness of assignments, and reducing the probability of collaboration failures and imbalanced participation in actual teaching. A structured design framework centered on disciplinary conflict has been constructed, transforming the originally vague "structurally flawed design requirements" into editable and evaluable conflict and resource units. This lowers the design threshold for interdisciplinary group assignments while ensuring the interdisciplinary depth and pedagogical value of the assignments.

[0148] 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 or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for determining interdisciplinary tasks based on multi-agent simulation, characterized in that, include: The interdisciplinary group assignments to be simulated are identified, which are learning tasks that students from different majors need to complete collaboratively. The learning tasks include teaching information for each major and its corresponding professional background description information, as well as assignments that each major needs to handle collaboratively. Constructing poorly structured assignments based on different teaching information; Student intelligent agent profile information is generated based on the professional background description information corresponding to each major. Based on the preset behavioral information and the student intelligent agent profile information, students corresponding to the same major are divided into different behavioral information groups. The preset behavioral information includes multiple behavioral information groups, and each behavioral information group corresponds to one behavioral information. The same behavioral information group includes students corresponding to all majors. Based on the poorly structured task, the teaching assistant agent and student agent in the multi-agent simulation system perform simulated dialogue processing on each behavioral information group to obtain the simulated dialogue information of the behavioral information group. The student agent and teaching assistant agent are both obtained through the Big Prophet model. The simulated dialogue information is segmented into multiple independent collaborative events, and these collaborative events are classified to obtain classification results. At the same time, core collaborative indicators are calculated based on the simulated dialogue information. When the classification results and core collaboration indicators meet the preset conditions, the interdisciplinary group assignment to be simulated will be identified as the target interdisciplinary group assignment.

2. The interdisciplinary task determination method based on multi-agent simulation according to claim 1, characterized in that, The construction of poorly structured assignments based on different teaching information includes: Semantic parsing is performed on each piece of teaching information to obtain semantic structure features, and teaching features are extracted from the semantic structure features. The teaching features include core theme information, application scenarios, user information, and boundary constraints. Construct problem-based contextual information among different majors based on multiple teaching characteristics; Extract multidisciplinary perspective information from the problem context information, identify interdisciplinary contradiction and conflict information in the multidisciplinary perspective information, decompose the interdisciplinary contradiction and conflict information into multiple conflict information, and match corresponding scaffold resources for each conflict information. Each conflict information includes subject area information and core contradiction information. Multiple conflicting information and their corresponding support resources were identified as structurally flawed operations.

3. The interdisciplinary task determination method based on multi-agent simulation according to claim 1, characterized in that, The process of generating student intelligent agent profile information based on the professional background description information corresponding to each major includes: The professional background description information is standardized and mapped according to the ISCED-F subject standard database to obtain standardized subject information, and professional competency information is determined based on the standardized subject information. Information on professional background descriptions is processed using an interdisciplinary writing competence assessment framework to obtain information on collaborative abilities. The professional background information, professional ability information, and collaborative ability information in the professional background description information are identified as the student's intelligent agent profile.

4. The method for determining interdisciplinary tasks based on multi-agent simulation according to claim 3, characterized in that, The determination of professional competence information based on standardized subject information includes: Based on the target professional competency parameters and the preset correspondence in the standardized subject information, the sub-professional competency information is determined, where the preset correspondence is the correspondence between professional competency parameters and professional competency information; In response to a user's modification of a target professional competency parameter, the system determines the professional competency parameter corresponding to the modification operation and determines the professional competency information based on the professional competency parameter and a preset correspondence.

5. The method for determining interdisciplinary tasks based on multi-agent simulation according to claim 1, characterized in that, According to the poorly structured task, the teaching assistant agent and student agent in the multi-agent simulation system perform simulated dialogue processing on each behavioral information group to obtain the simulated dialogue information of the behavioral information group, including: Determine the target behavioral information of the behavioral information group, and determine the target behavioral label information of the behavioral information group based on the target behavioral information and the behavioral mapping model, wherein the behavioral mapping model is a mapping model between the behavioral information group and the behavioral label information, and the target behavioral information is any one of the behavioral information in the behavioral information group; Based on the poorly structured task, the simulation dialogue process of the agent in the behavior information group is controlled by the teaching assistant agent, and the sub-simulation dialogue corresponding to each student in the behavior information group is generated by the student agent based on the target behavior label information. The sub-simulation dialogues corresponding to multiple students in the behavior information group are identified as simulation dialogue information.

6. The method for determining interdisciplinary tasks based on multi-agent simulation according to claim 5, characterized in that, The simulated dialogue process based on poorly structured tasks, which controls the agents in the behavior information group through the teaching assistant agent, generates sub-simulated dialogues for each student in the behavior information group based on target behavior label information through the student agent, including: The teaching assistant agent sequentially sets multiple conflicting information and their corresponding scaffolding resources in poorly structured assignments as the current discussion topic, and determines the speaking order of each student in the behavior information group based on the target behavior label information through a dynamic polling strategy. Based on the speaking order, a sub-simulation dialogue is generated by the student agent for the current discussion topic, wherein the sub-simulation dialogue includes the dialogue of all students in the behavioral information group; When the number of discussions on the current topic reaches the preset number of rounds or when a discussion result is obtained on the current topic, the simulation discussion stops and the sub-simulation dialogue corresponding to each student in the behavior information group is output.

7. The method for determining interdisciplinary tasks based on multi-agent simulation according to claim 1, characterized in that, The core collaboration metrics include convergence, integration, adversarial relationship, and cross-group event correlation. The convergence is expressed by the following formula: ； ； in, To measure the convergence of the simulated dialogue information, For the weight function, This represents the number of discussions in the simulated dialogue information. For target behavior label information; The integration degree and the adversarial degree are expressed by the following formula: ； ； ； ； in, For the first The cumulative integration degree of information in round-robin simulation dialogue. For the first Cumulative adversarial intensity in simulated dialogue information. For smoothing coefficients, For the first The cumulative integration degree of information in round-robin simulation dialogue. This is the basic weight vector corresponding to each behavior label in the target behavior label information. For the first Cumulative adversarial intensity in simulated dialogue information. As the integration weight in the simulated dialogue information, The adversarial weights are used in the simulated dialogue information.

8. The method for determining interdisciplinary tasks based on multi-agent simulation according to claim 1, characterized in that, The method further includes: When the classification results and core collaboration indicators do not meet the preset conditions, collaboration failure information is generated. In response to a user's input operation regarding collaboration failure information, the system determines the target data information corresponding to the input operation and sends the target data information to the server. The server then generates target optimization suggestions based on the target data information and a preset suggestion correspondence. Based on the target optimization suggestions, a new interdisciplinary group assignment to be simulated is generated, along with new classification results and new core collaboration indicators. This process continues until the new classification result is at least one of a positive event or a neutral event, and the new core collaboration indicators meet preset conditions. At this point, the new interdisciplinary group assignment to be simulated is identified as the target interdisciplinary group assignment. The preset suggestion correspondence is the correspondence between data information and optimization suggestions.

9. A cross-disciplinary task determination system based on multi-agent simulation, characterized in that, It includes a first determination module, a poorly structured task construction module, a student profile generation module, a multi-agent collaborative simulation module, a simulation result analysis module, and a second determination module; The first determining module is used to determine the interdisciplinary group assignment to be simulated, wherein the interdisciplinary group assignment to be simulated is a learning task that students from different majors need to complete collaboratively, and the learning task includes the teaching information of each major and its corresponding professional background description information and the assignment that each major needs to process collaboratively; The poorly structured assignment construction module is used to construct poorly structured assignments based on different teaching information; The student profile generation module is used to generate student intelligent agent profile information based on the professional background description information corresponding to each major, and to divide students corresponding to the same major into different behavioral information groups according to the preset behavioral information and the student intelligent agent profile information. The preset behavioral information includes multiple behavioral information groups, and each behavioral information group corresponds to one behavioral information. The same behavioral information group includes students corresponding to all majors. The multi-agent collaborative simulation module is used to perform simulated dialogue processing on each behavioral information group according to the poorly structured task through the teaching assistant agent and student agent in the multi-agent simulation system to obtain the simulated dialogue information of the behavioral information group. The student agent and teaching assistant agent are both obtained through the Big Prophet model. The simulation result analysis module is used to segment the simulation dialogue information into multiple independent collaborative events, classify the multiple collaborative events according to a preset correspondence, obtain classification results, and calculate core collaborative indicators based on the simulation dialogue information. The second determining module is used to determine the interdisciplinary group assignment to be simulated as the target interdisciplinary group assignment when the classification results and core collaboration indicators meet the preset conditions.

Images