Learning path recommendation system based on knowledge graph and multi-modal learning analysis
The learning path recommendation system, which utilizes knowledge graphs and multimodal learning analytics, addresses the lack of targeted and dynamic adaptability in online learning paths. It enables adaptive planning of personalized learning paths, thereby improving learning efficiency and effectiveness.
Patent Information
- Application Number
- CN202511821673.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-05
- Publication Date
- 2026-03-06
AI Technical Summary
Existing online learning platforms lack comprehensive analysis of learners' cognitive states and multimodal learning behaviors, resulting in recommended learning paths that lack specificity and dynamic adaptability, making it difficult to meet personalized learning needs.
A learning path recommendation system based on knowledge graphs and multimodal learning analysis is adopted. Through modules for multimodal learning data collection, knowledge graph construction, multimodal fusion analysis, and learning path generation, combined with reinforcement learning algorithms, personalized learning paths are generated and adjusted in real time to adapt to the dynamic changes of learners.
It significantly improves learning efficiency and effectiveness. Through multimodal data fusion and dynamic knowledge graph construction, it accurately diagnoses learners' cognitive status, dynamically adjusts recommended content, and achieves adaptive planning of personalized learning paths.
Smart Images
Figure CN121614671A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of smart education technology, and in particular to a learning path recommendation system based on knowledge graphs and multimodal learning analysis. Background Technology
[0002] In existing online learning platforms, learning resource recommendations are mostly based on single-dimensional user behavior data, such as click-through rates or test scores. This lacks comprehensive analysis of learners' cognitive states, knowledge structure relationships, and multimodal learning behaviors, resulting in recommended learning paths that often lack relevance and dynamic adaptability. Traditional recommendation systems struggle to capture the complex dependencies between knowledge points and cannot dynamically adjust to real-time changes in learners' cognitive states, leading to low learning efficiency and failing to meet personalized learning needs. Therefore, how to build a system that can integrate multi-source learning data, dynamically reflect learners' cognitive states, and generate personalized learning paths has become a pressing technical problem in the current online education field. Summary of the Invention
[0003] In view of this, to address the problems existing in the technical background, this invention proposes a learning path recommendation system based on knowledge graphs and multimodal learning analysis. Specifically, it includes the following:
[0004] A learning path recommendation system based on knowledge graphs and multimodal learning analysis includes a multimodal learning data acquisition module, a knowledge graph construction module, a multimodal fusion analysis module, a learning path generation module, and a feedback and update module. The output of the multimodal learning data acquisition module is connected to the input of the knowledge graph construction module and the multimodal fusion analysis module. The output of the knowledge graph construction module is connected to the multimodal fusion analysis module. The output of the multimodal fusion analysis module is connected to the learning path generation module. The output of the learning path generation module is connected to the feedback and update module. The output of the feedback and update module feeds back to the knowledge graph construction module and the multimodal fusion analysis module, respectively.
[0005] In one embodiment of the present invention, the multimodal learning data acquisition module includes a learning behavior acquisition unit, a cognitive state acquisition unit, and a knowledge mastery acquisition unit; the learning behavior acquisition unit is used to collect clickstream data, video viewing time, and homework submission behavior data from the student's online learning platform; the cognitive state acquisition unit is used to collect reaction time and error pattern data from the student's answer records; the knowledge mastery acquisition unit is used to collect test scores and practice accuracy data for each knowledge point; the data output by each acquisition unit is preprocessed and then input into the knowledge graph construction module and the multimodal fusion analysis module, respectively.
[0006] In one embodiment of the present invention, the knowledge graph construction module treats knowledge points as entities and labels each knowledge point with a mastery status and cognitive difficulty weight based on the student's test scores and answer behavior, thereby transforming the static knowledge structure into a dynamic evolution graph that reflects the individual's cognitive state.
[0007] In one embodiment of the present invention, the learning path generation module, based on the student's comprehensive cognitive feature vector output by the multimodal fusion analysis module, combines a reinforcement learning algorithm to make learning sequence decisions and generate an optimal learning path that conforms to the student's cognitive level and learning objectives.
[0008] In one embodiment of the present invention, the specific formula of the reinforcement learning algorithm is as follows:
[0009] Q(s,a)=Q(s,a)+α[r+γmaxQ(s',a')-Q(s,a)];
[0010] Where s represents the current knowledge point status, a represents the selected learning action, r is the immediate reward, α is the learning rate, and γ is the discount factor. By continuously evaluating the long-term value of different learning actions, the learning path decision is gradually optimized to ensure that the recommended path is both in line with the current cognitive level and can efficiently achieve the learning goal.
[0011] In one embodiment of the present invention, the learning path generation module is further configured to dynamically adjust subsequent recommended content based on real-time feedback data from students during the learning process, thereby achieving adaptive learning path planning; the dynamic adjustment includes changing the order of knowledge points, replacing learning resources, and adjusting the difficulty level.
[0012] The above technical solution has the following beneficial effects:
[0013] This invention, through deep fusion of multimodal data and construction of a dynamic knowledge graph, can not only accurately diagnose learners' cognitive states but also dynamically adjust recommended content based on real-time feedback, thereby significantly improving learning efficiency and effectiveness. The multimodal learning data acquisition module comprehensively collects learners' behavioral data, cognitive state data, and knowledge mastery data, providing a rich data foundation for subsequent analysis. The knowledge graph construction module structures knowledge points and their relationships and, combined with learners' mastery status, constructs a dynamic cognitive graph, accurately reflecting learners' knowledge structure deficiencies and cognitive difficulty distribution. The multimodal fusion analysis module deeply integrates data from different sources and modalities, extracting learners' comprehensive cognitive characteristics to provide precise input for path generation. The learning path generation module, based on reinforcement learning algorithms, generates and dynamically optimizes learning paths by combining learners' real-time states and learning objectives. The feedback and update module updates the knowledge graph and multimodal analysis model in real time based on learners' performance during the path learning process, forming a closed-loop optimization mechanism. Attached Figure Description
[0014] Figure 1 This is a schematic diagram of the structure of a learning path recommendation system based on knowledge graph and multimodal learning analysis according to the present invention. Detailed Implementation
[0015] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0016] like Figure 1 As shown, a learning path recommendation system based on knowledge graphs and multimodal learning analysis includes a multimodal learning data acquisition module, a knowledge graph construction module, a multimodal fusion analysis module, a learning path generation module, and a feedback and update module. The output of the multimodal learning data acquisition module is connected to the input of the knowledge graph construction module and the multimodal fusion analysis module. The output of the knowledge graph construction module is connected to the multimodal fusion analysis module. The output of the multimodal fusion analysis module is connected to the learning path generation module. The output of the learning path generation module is connected to the feedback and update module. The output of the feedback and update module feeds back to the knowledge graph construction module and the multimodal fusion analysis module, respectively.
[0017] In this embodiment of the invention, the modules are sequentially connected to form a closed loop. The multimodal learning data acquisition module is responsible for collecting students' learning behavior data, cognitive state data, and knowledge mastery data from the online learning platform. The learning behavior acquisition unit records students' clickstream, video viewing time, and assignment submission behavior. The cognitive state acquisition unit analyzes students' response time and error patterns. The knowledge mastery acquisition unit collects students' test scores and practice accuracy rates on various knowledge points. After cleaning, noise reduction, and standardization preprocessing, this data is input into the knowledge graph construction module and the multimodal fusion analysis module, respectively, providing the system with comprehensive and multi-faceted learner status information. The knowledge graph construction module uses knowledge points as entities and labels each knowledge point with mastery status and cognitive difficulty weights based on students' test scores and answering behavior. This transforms the static knowledge structure into a dynamic evolution graph that reflects an individual's cognitive state. The graph not only reveals the logical relationships between knowledge points but also reflects the learner's mastery level and cognitive load on different knowledge points in real time.
[0018] Example 2, based on Example 1, describes a multimodal learning data acquisition module comprising a learning behavior acquisition unit, a cognitive state acquisition unit, and a knowledge mastery acquisition unit. The learning behavior acquisition unit collects clickstream data, video viewing time, and homework submission data from the student's online learning platform. The cognitive state acquisition unit collects reaction time and error pattern data from student answer records. The knowledge mastery acquisition unit collects test scores and practice accuracy data for each knowledge point. The data output from each acquisition unit is preprocessed and then input to the knowledge graph construction module and the multimodal fusion analysis module, respectively. The multimodal fusion analysis module receives raw multimodal data from the data acquisition module and dynamic graph information provided by the knowledge graph module. Through feature extraction and fusion techniques, it integrates behavioral data, cognitive data, and knowledge mastery data into a unified student comprehensive cognitive feature vector. This vector comprehensively depicts the learner's current cognitive state, learning habits, and knowledge structure characteristics. Its working principle involves using a multimodal neural network or attention mechanism to align and weightedly fuse data from different sources, thereby extracting more discriminative feature representations and providing accurate input to the learning path generation module.
[0019] Example 3, based on Example 1, the knowledge graph construction module in this example treats knowledge points as entities and labels the mastery status and cognitive difficulty weight of each knowledge point according to the student's test scores and answer behavior, transforming the static knowledge structure into a dynamic evolution graph that reflects the individual's cognitive state.
[0020] In this embodiment, the learning path generation module, based on the student's comprehensive cognitive feature vector output by the multimodal fusion analysis module, combines a reinforcement learning algorithm to make learning sequence decisions and generate an optimal learning path that matches the student's cognitive level and learning objectives.
[0021] The specific formula for the reinforcement learning algorithm is as follows:
[0022] Q(s,a)=Q(s,a)+α[r+γmaxQ(s',a')-Q(s,a)];
[0023] Where s represents the current knowledge point status, a represents the selected learning action, r is the immediate reward, α is the learning rate, and γ is the discount factor. By continuously evaluating the long-term value of different learning actions, the learning path decision is gradually optimized to ensure that the recommended path is both in line with the current cognitive level and can efficiently achieve the learning goal.
[0024] The feedback and update module receives the output of the learning path generation module and real-time performance data of students during the learning process. By analyzing the effectiveness of the path recommendation and the learner's progress, it generates feedback signals and transmits them to the knowledge graph construction module and the multimodal fusion analysis module, respectively. The knowledge graph module updates the mastery status and cognitive difficulty weight of knowledge points based on the feedback, while the multimodal fusion analysis module adjusts the parameters of the feature fusion model using the new data, thus forming a continuously optimized closed-loop system. Its working principle is that through continuous iterative updates, the system can adapt to the dynamic changes of learners, maintain the accuracy and timeliness of the recommended path, and ultimately achieve continuous improvement in learning outcomes.
[0025] The basic principles and main features of the present invention have been described above. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are only illustrative of the principles of the present invention. Various changes and modifications can be made to the present invention without departing from the spirit and scope of the present invention. All such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the invention is defined by the appended claims and their equivalents.
Claims
1. A learning path recommendation system based on knowledge graph and multi-modal learning analysis, characterized in that, The system comprises a multi-modal learning data acquisition module, a knowledge graph construction module, a multi-modal fusion analysis module, a learning path generation module, and a feedback and updating module.
2. The learning path recommendation system based on knowledge graph and multi-modal learning analysis according to claim 1, characterized in that, The multi-modal learning data acquisition module comprises a learning behavior acquisition unit, a cognitive state acquisition unit, and a knowledge mastery acquisition unit. The cognitive state acquisition unit is configured to acquire reaction time and error mode data in student answer records. The knowledge mastery acquisition unit is configured to acquire test scores and exercise accuracy data of students on each knowledge point. 3.The learning path recommendation system based on knowledge graph and multi-modal learning analysis of claim 1, wherein, The knowledge graph construction module takes knowledge points as entities, labels mastery status and cognitive difficulty weight for each knowledge point according to test scores and answering behaviors of students, and converts static knowledge structure into a dynamic evolution graph reflecting individual cognitive status. 4.The learning path recommendation system based on knowledge graph and multi-modal learning analysis of claim 1, wherein, The learning path generation module generates an optimal learning path meeting the cognitive level and learning goal of students based on the student comprehensive cognitive feature vector output by the multi-modal fusion analysis module and in combination with a reinforcement learning algorithm. 5.The learning path recommendation system based on knowledge graph and multi-modal learning analysis according to claim 4, characterized in that, The reinforcement learning algorithm of the reinforcement learning is specifically as follows: Q(s,a)=Q(s,a)+α[r+γmaxQ(s',a')-Q(s,a)]; where s represents the current knowledge point state, a represents the selected learning action, r is the immediate reward, α is the learning rate, and γ is the discount factor. By continuously evaluating the long-term value of different learning actions, the learning path decision is gradually optimized to ensure that the recommended path meets the current cognitive level and efficiently achieves the learning goal. 6.The learning path recommendation system based on knowledge graph and multi-modal learning analysis of claim 4, wherein, The learning path generation module is further configured to dynamically adjust subsequent recommended content according to real-time feedback data of students in the path learning process, thereby realizing adaptive learning path planning. The dynamic adjustment includes knowledge point order change, learning resource replacement, and difficulty level adjustment.