Abstract: Aiming at the deficiencies of the existing learning path recommendation methods in personalized adaptability,dynamic adaptability and multi-objective optimization,a collaborative recommendation model based on hypergraph neural network and knowledge tracking is proposed.By constructing an undirected graph of learning resources to encode the association relationship of resources,generating the embedding vectors of learning resources,and combining with the hypergraph neural network to aggregate the historical behavior data of learners,the interaction features between learners and learning resources are captured.It designs a multi-objective optimization strategy,utilizes the non-dominated sorting genetic algorithm(NSGA-II) to generate the Pareto frontier solution set,synchronously optimizes the accuracy,difficulty adaptability,effectiveness and diversity of learning resources of the learning path,and combines weight distribution and comprehensive utility function to improve the quality of the learning path.Experiments show that the proposed method is tested on the MOOCCube and MOOPer datasets.Among them,the HR@5 and MRR@5 of the MOOPer data reach 93.9% and 90.7% respectively,achieving precise recommendation of learning paths and verifying the effectiveness of the model in the modeling of learners’ historical interactions and the integration of course structure constraints.

Key words: Learning path recommendation, Hypergraph neural network, Knowledge tracking, Multi-objective optimization, Weight allocation