Learning Path Generation and Cyber Attack Risk Assessment of Online Education Platform Using Transfer Learning and Transformer Optimization Technology

Abstract

Scarcity of new user data and neglect of the correlation between knowledge points are the main problems in learning path generation of online education platforms. To improve the quality of learning path generation on online education platforms and deal with possible network attack risks, transfer learning is used to reduce the dependence on new data, and the logical relationship between knowledge points is captured by improving Transformer to identify abnormal situations in user behavior. MOOCCube is used as the source task, and the BERT4Rec (Sequential Recommendation with Bidirectional Encoder Representations from Transformer) transfer learning model is used to learn the behavioral dependencies of users at different time points. This paper improved the Transformer and improved the self-attention mechanism by combining time weighting and behavior frequency weighting. In addition, the relationship between knowledge points in courses is transformed into a knowledge graph, and GAT (Graph Attention Network) is introduced to consider the relationship between knowledge points in courses and further enhance the self-attention mechanism. The Transformer model can process input data in parallel, quickly identify unusual activity patterns, and warn of potential attacks. The learning path generation effect was verified on the MOOCCubeX-Small dataset. The similarity between the learning path generated in this paper and the expert path in a single course reached 97.2%, and the similarity in the combination of two courses reached 95.4%. The minimum fitness value of the model in this paper in a single course reached 2, the user retention rate in the 12th month reached 90.8%, and the class skipping rate was only 6.4%. In the case of data scarcity, the fitness value of the path generated in this paper still reaches 78.8% in the 12th month. The average abnormal traffic rate of the path generation method in this paper in different attacks is only 3.8%. Combining BERT4Rec with the learning path generated by the improved Transformer can improve the quality of the learning path and provide a new solution for the learning path generation of online education platforms.