Research On Course Recommendation Technology Based On Online Education Platform Behavior Lo

The proportion of online education in daily educational activities has increased rapidly and has become an essential educational method.In the face of massive curriculum resources,the traditional information search methods have some problems,such as information trek,curriculum resources overloaded and so on.Course recommendation technology is the key to the rapid matching between learners and course resources.Most of the traditional sequence recommendation methods aim at the user’s own sequence coding,ignoring the relationship between sequences.At the same time,there is a cold start problem for users with insufficient behavior logs.This paper conducts relevant research to address the above-mentioned problems of behavior log recommendation algorithms,and the main research results are as follows.(1)SRSGNN,a course recommendation algorithm based on behavior logs,is proposed.Most sequence recommendation algorithms model just for a single sequence so that ignore the connection between different sequences and it affects the effectiveness of the recommendation.To address the problem,the sequence recommendation algorithm SRSGNN is proposed.It uses the Min Hash algorithm to divide the user’s behavior logs into neighborhoods,then the global-level course interaction representation is learned using behavior logs within the same neighborhood,and the user-level course interaction representation is learned using the behavior logs of individual users.Fusing these two levels of course representations for course recommendation makes the recommended results take into account both user information and neighborhood information.(2)Focor,a course recommendation algorithm based on the course selection dataset,is proposed to address the user cold-start problem of the behavior log recommendation algorithm.Focor is a course recommendation algorithm based on course selection probability calculation.Focor uses a genetic algorithm to filter the optimal feature subset from the feature set composed of the course selection dataset,and uses Light GBM on the optimal feature subset to train a binary classification model for whether to select a course or not,and then recommends courses based on the course selection probability output by the model.The experimental results show that the SRSGNN course recommendation algorithm outperforms SRGNN and traditional sequential recommendation models in terms of Recall@20 and MRR@20 metrics on Tracking Log,Yoo Choose,and Diginetica data.Focor is experimented and performance evaluated on real datasets with algorithms such as Light GBM,XGBoost,decision tree,random forest and Logistic Regression,and achieved the best performance on F1 score.