Learning Sequence Recommendation Algorithm Based on Learner Interest and Neighborhood Information

Now, most of the sequential recommendation algorithms model learner interests and course information based on the current learner's learning sequence, lacking the use of course order information and mining course information in the neighborhood sequence, which makes it difficult to capture accurate learner interests and comprehensively describe course information. In response to the above issues, this paper fully uses the sequential recommendation method and proposes an online course Recommendation Algorithm Based on Learner Interests and Neighborhood Information (LINI-CR). The algorithm learns the contextual relationships of courses in an interaction sequence through a GRU (Gate Recurrent Unit) and superimposes the learning sequence information to model learner interests. Then, it searches for similar sequences based on the interaction sequences of learners and courses, constructs a neighborhood course sequence graph using the set of similar sequences, and uses directed graph attention propagation on the structure of this graph to mine the higher-order connectivity information of courses. On this basis, the probability of learner-course interaction is obtained through inner product operation and normalization, and course recommendation is based on the probability. Experiments were conducted on the MOOCCourse and MovieLens-1M datasets, and the results proved the accuracy and effectiveness of the algorithm, with an improvement in each evaluation metric compared to the better-performing baseline method, Precision@20 improves by 3.97%, 2.12% and NDCG@20 improves by 5.23%, 1.03%, respectively.