CN-121981863-A - Accompanying learning method and system based on multiple agents

Abstract

The invention relates to the technical field of intersection of artificial intelligence and education technology, and provides a multi-agent-based accompanying learning method and system, comprising the steps of acquiring learning behavior data for learning condition analysis; based on the project reaction theory, calculating the grasping probability of the students to the knowledge points, reasoning the learning emotion states of the students through the dialogue content of the learning interaction, updating the memory module of the digital learning partner agent according to the learning condition analysis result, the grasping probability of the knowledge points and the learning emotion states, further recommending matched practice problems, and generating a personalized learning path of the students through the constructed digital teacher agent. The invention realizes the dynamic generation of practice problem recommendation and personalized learning path based on the collaborative update of the digital learning partner agent and the digital teacher agent, thereby improving the accuracy, continuity and self-adaptability of learning support.

Inventors

• MA KUN
• CHEN JIAYIN
• Jing Yuanping

Assignees

• 济南大学

Dates

Publication Date

20260505

Application Date

20260408

Claims (10)

1. 1. The multi-agent-based syndrome learning method is characterized by comprising the following steps of: Acquiring learning behavior data, extracting a problem solving inference chain of a student aiming at a problem, and comparing the problem solving inference chain with a standard knowledge inference path to obtain a learning condition analysis result; based on knowledge points corresponding to the target questions and learning condition analysis results, calculating the grasping probability of students on the knowledge points based on the project reaction theory; reasoning the learning emotion state of the student based on the dialogue content of the learning interaction in the learning behavior data; Updating a memory module of the constructed digital learning partner agent according to the learning condition analysis result, the grasping probability of the knowledge points and the learning emotion state, and generating a practice problem recommendation result which is matched with the current learning state of the student based on the updated digital learning partner agent; based on the learning condition analysis result and the acquired learning task information, a personalized learning path of the student is generated through the built digital teacher agent.
2. 2. The multi-agent based companion learning method of claim 1 wherein the constructing of digital companion agents includes student image and multi-layer memory modules; Based on the analysis result of the learning condition, the mastering probability of the knowledge points and the learning emotion state data, comprehensively characterizing the current learning characteristics of the students to form a multidimensional characteristic set serving as a student portrait; The memory module comprises a fact memory module, a short-term memory module and a long-term memory module; the fact memory module is used for storing student learning records obtained based on learning behavior data analysis in real time; The short-term memory module is used for storing short-term student learning records obtained in a recent short term; the long-term memory module is used for storing the learning record of the student forming the long-term memory so as to simulate the stable ability structure and the knowledge mastering mode which are gradually formed by the student in the long-term learning process.
3. 3. The multi-agent-based syndrome learning method of claim 1, wherein the method for obtaining learning behavior data of students, extracting a problem solving inference chain of the students aiming at problems, comparing with a standard knowledge inference path, and obtaining learning condition analysis results of the students comprises the following steps: Extracting answer results of students aiming at the objective title in learning behavior data and performing intermediate calculation; Encoding the answer result and the intermediate calculation step to generate an intermediate calculation step vector, and processing the intermediate calculation step vector through embedding operation and a self-attention mechanism to obtain the dependency relationship and the step importance weight among the answer steps; based on the intermediate calculation step vector, the dependency relationship among the solving steps and the step importance weight, constructing a solving problem representation comprising a step sequence relationship and a key reasoning path; carrying out semantic analysis and causal inference on the solution representation by using a subject education big language model, and mapping the student solution process into a structural solution inference chain; verifying the rationality, consistency and integrity of each reasoning node in the structural question solving reasoning chain, comparing the rationality, consistency and integrity with the standard knowledge reasoning path, positioning the reasoning deviation nodes, and determining the reason of the question answering error according to the reasoning deviation nodes.
4. 4. The multi-agent based syndrome learning method of claim 1, wherein the method for calculating the probability of learning knowledge points by students based on the knowledge points corresponding to the target subjects and the learning situation analysis results and the project reaction theory comprises the steps of: Based on learning behavior data, extracting a student response result to the object title, and combining a distinguishing parameter and a difficulty parameter of the object title, and carrying out parameter estimation on the student response result based on an item reaction theoretical model to obtain a student capacity factor; and mapping the questions to corresponding knowledge points in the knowledge graph, and calculating the mastering probability of the students on each knowledge point according to the student capacity factors.
5. 5. The multi-agent based syndrome learning method of claim 4 wherein the method for calculating the probability of a student grasping each knowledge point comprises: vectorizing the intermediate calculation steps in the student capacity factors, the topics and the learning behavior data, and splicing to form student behavior feature vectors; And carrying out weighted calculation and nonlinear transformation on the student behavior feature vectors to obtain feature weights of all knowledge points, and mapping the feature weights into mastering probabilities of students on all knowledge points through a normalization function.
6. 6. The multi-agent-based syndrome learning method of claim 1, wherein the updating of the built memory module of the digital syndrome agent according to the learning situation analysis result, the grasping probability of the knowledge points and the learning emotion state comprises the steps of: Updating static portrait features based on learning condition analysis results, knowledge point mastering probability and learning emotion state data, and mapping the updated static portrait features into dynamic response behavior results of students in a target subject scene by adopting a large language model to serve as a new student learning record; Based on the obtained student learning record, updating a memory module of the digital learning partner agent: based on the obtained student learning record, updating a fact memory module, and performing reinforced counting on similar historical records; According to the updated fact memory module, storing a new student learning record into the short-term memory module so as to store continuous learning behaviors of students in a recently set time; And updating the long-term memory module according to the intensified counting result and the forgetting rule to form a stable knowledge mastering representation of the student.
7. 7. The multi-agent based syndrome learning method of claim 1, wherein generating a practice problem recommendation result adapted to a current learning state of a student based on the updated digital syndrome agent comprises the steps of: Determining knowledge points to be consolidated, lifted or corrected according to knowledge point mastering probability, forgetting score and learning emotion state of students, and taking the knowledge points to be consolidated, lifted or corrected as target knowledge points; Based on the prompt word template, inputting the target knowledge point, the knowledge point mastering probability, the time from the last time of training the knowledge point and the forgetting score into a large language model, and generating a personalized practice problem recommendation result.
8. 8. The multi-agent based syndrome learning method of claim 1, wherein the method for generating a personalized learning path of a student through a built digital teacher agent based on the acquired learning task information and learning condition analysis result comprises the steps of: the digital teacher agent builds a problem map based on the field problem library; The digital teacher agent builds a knowledge graph based on the domain knowledge; Based on the knowledge graph, the analysis result of the learning situation of the students and the current learning task information, carrying out structure enhancement representation and learning state matching on the candidate knowledge points, determining a next learning target, and generating a personalized learning path containing the target knowledge points and a question set extracted from the problem graph.
9. 9. The multi-agent based syndrome learning method of claim 8 wherein the method of performing structure enhanced representation and learning state matching on candidate knowledge points based on knowledge maps, student learning situation analysis results, and current learning task information, determining a next learning objective, and generating a personalized learning path containing objective knowledge points and a set of topics extracted from a problem map comprises the steps of: Acquiring a precursor knowledge point and a postcursor knowledge point of a target knowledge point in a current learning task based on the knowledge graph, and enhancing the target knowledge point to obtain an enhanced text description sequence of the target knowledge point; coding the enhanced text description sequence of the target knowledge points and carrying out multi-layer neighborhood aggregation to obtain knowledge point representation with enhanced relationship; Based on the analysis result of the learning condition of the student and the current knowledge point, dynamically modeling the knowledge state of the student to obtain the time step of the student And enhanced representation of the relationship with the current knowledge point Splicing the knowledge point embedded vector and the answer result embedded vector to form a unified learning behavior representation; Inputting a learning behavior sequence formed by learning behavior representations into a sequence modeling network based on a self-attention mechanism, and modeling historical learning behaviors to obtain a context perception characteristic representation at the current moment; Based on the context perception feature representation, carrying out matching calculation on candidate knowledge points, calculating the recommendation probability of each knowledge point as a next learning target, selecting the knowledge point with the top ranking as a recommendation result, screening a question set corresponding to the knowledge point in a layered problem map, and generating a personalized learning path conforming to the current knowledge state and capability level of students by combining learning resources.
10. 10. Multi-agent-based syndrome learning system, comprising: the learning situation analysis module is configured to acquire learning behavior data, extract a problem solving inference chain of a student aiming at a problem, and compare the problem solving inference chain with a standard knowledge inference path to obtain a learning situation analysis result; The mastering probability recognition module is configured to calculate the mastering probability of the student on the knowledge points based on the knowledge points corresponding to the target topics and the learning condition analysis result and based on the project reaction theory; The emotion state reasoning module is configured to reason the learning emotion state of the student based on dialogue contents of learning interaction in the learning behavior data; The emotion state module is configured to update the memory module of the constructed digital learning partner agent according to the learning condition analysis result, the grasping probability of the knowledge points and the learning emotion state, and generate a practice problem recommendation result which is matched with the current learning state of the student based on the updated digital learning partner agent; And the personalized learning path generation module is configured to generate a personalized learning path of the student through the built digital teacher agent based on the learning condition analysis result and the acquired learning task information.

Description

Accompanying learning method and system based on multiple agents Technical Field The invention relates to the technical field of cross correlation of artificial intelligence and education technology, in particular to a multi-agent-based accompanying learning method and system. Background The statements in this section merely provide background information related to the present disclosure and may not necessarily constitute prior art. With the development of large language models and intelligent agent technologies, educational intelligent agents are gradually applied to educational scenes such as teaching assistance, learning support and process evaluation. Compared with the traditional learning system mainly comprising course content transmission, learning management and result assessment, the learner needs continuous accompany, dynamic guiding and accurate feedback based on individual capability difference in the actual learning process. Especially in the whole process of pre-class preparation, in-class learning and post-class consolidation, the existing system is difficult to continuously adapt around the cognitive state, learning progress and emotion change of students, and the development requirements of personalized teaching and accompanying learning are difficult to meet. The conventional educational agent or accompanying learning system still has the problems that on one hand, the data of the learning process of students are not fully mined and utilized, knowledge mastering conditions, learning difficulties and state changes of the students are coarsely recognized, the actual learning states of the students are difficult to accurately represent in the accompanying learning process, so that the pertinence of learning support and teaching intervention is insufficient, on the other hand, the conventional learning path generation or recommendation mode is dependent on preset rules, fixed flows or static association relations, continuous response capability to the learning state changes of the students is lacking, and the personalized learning path is difficult to dynamically adjust according to the learning state changes of the students, so that the suitability of learning content recommendation and learning rhythm arrangement is insufficient. Disclosure of Invention In order to solve the problems, the invention provides a multi-agent-based accompanying learning method and system, which are based on collaborative updating of digital learning accompanying agents and digital teacher agents, and realize the recommendation of practice problems and the dynamic generation of personalized learning paths, thereby improving the accuracy, continuity and self-adaptability of learning support. In order to achieve the above purpose, the present invention adopts the following technical scheme: the first aspect of the invention provides a multi-agent-based syndrome learning method, comprising the following steps: Acquiring learning behavior data, extracting a problem solving inference chain of a student aiming at a problem, and comparing the problem solving inference chain with a standard knowledge inference path to obtain a learning condition analysis result; based on knowledge points corresponding to the target questions and learning condition analysis results, calculating the grasping probability of students on the knowledge points based on the project reaction theory; reasoning the learning emotion state of the student based on the dialogue content of the learning interaction in the learning behavior data; Updating a memory module of the constructed digital learning partner agent according to the learning condition analysis result, the grasping probability of the knowledge points and the learning emotion state, and generating a practice problem recommendation result which is matched with the current learning state of the student based on the updated digital learning partner agent; based on the learning condition analysis result and the acquired learning task information, a personalized learning path of the student is generated through the built digital teacher agent. Further technical proposal, a digital learning partner agent is constructed, which comprises a student image and a multi-layer memory module; Based on the analysis result of the learning condition, the mastering probability of the knowledge points and the learning emotion state data, comprehensively characterizing the current learning characteristics of the students to form a multidimensional characteristic set serving as a student portrait; The memory module comprises a fact memory module, a short-term memory module and a long-term memory module; the fact memory module is used for storing student learning records obtained based on learning behavior data analysis in real time; The short-term memory module is used for storing short-term student learning records obtained in a recent short term; the long-term memory module is used for storing the learning