Research On Key Technologies Of MOOCs Based On Modeling Of Interactive Behaviors

In recent years,with the development of 5G and the Internet,online education has become popular.Massive Open Online Courses(MOOCs)in online education have attracted tens of millions of users worldwide with their abundant course resources and quality course services.However,MOOCs also face problems such as high dropout rates and low resource utilization.Therefore,predicting user dropouts and recommending courses can effectively improve these issues.In the MOOCs system,there are a large number of historical interaction records between users and the system.Reasonably modeling these interaction behaviors and applying them to dropout prediction and course recommendation is of great significance.However,most existing MOOCs dropout prediction methods mainly focus on the sequential dependency of behavior sequences and lack capturing the relationship between different time steps of behavior,which leads to difficulty in obtaining an effective representation of user behavior sequences.Existing course recommendation methods do not fully explore the non-linear relationships among "user-course" interaction behaviors and ignore the guidance of knowledge in course recommendation for user needs.To address these issues,this thesis conducts research on dropout prediction and course recommendation based on interaction behavior modeling,with the main contributions as follows:From the perspective of dropout prediction,this thesis proposes a pre-training method based on behavior sequence to predict dropouts.This method first compresses the behavior sequence to reduce its sparsity and then uses a bidirectional Transformer encoder to pre-train the behavior sequence,thereby capturing the relationship between behaviors at different time intervals and obtaining better behavior sequence features.Then,the explicit information of users and courses is encoded and fused,and dropout prediction is performed jointly with user behavior sequences.Experimental comparisons verify that the proposed dropout prediction method improves the accuracy of dropout prediction.From the perspective of course recommendation,this thesis proposes a course recommendation method based on multi-modal knowledge space interactive graph convolution.This method first constructs a "user-behavior-course" heterogeneous graph using the interaction relationship between users and courses,and then uses multi-modal learning to fuse course knowledge into multiple modalities to enhance course vectors.The user representation vector and course representation vector are obtained through graph convolutional learning of multiple behavior subgraphs and course subgraphs,respectively,and the recommendation is completed by calculating the similarity between the user representation vector and the course representation vector.Experimental results show that the proposed course recommendation method improves the effectiveness of course recommendation.Based on the above methods and theories,this thesis develops a mooc online learning prototype system,which mainly implements core functions such as online learning,dropout prediction,and course recommendation through steps such as demand analysis,summary design,detailed design,and specific implementation.the feasibility and effectiveness of the proposed dropout prediction and course recommendation methods are validated.