Research On Personalized Course Recommendation Algorithm Based On Deep Learning

Online learning platforms provide people with the opportunity to learn anytime,anywhere,and meet their individual needs.However,there are a large number of course resources on online learning platforms,and a large number of courses can meet the needs of different people,but users need to spend a lot of time selecting courses that meet their needs and style,thus creating the problem of information overload,in order to solve this problem,personalized course recommendation system was born.Although recommendation algorithms have achieved better results in other domains,they are not mature enough in online learning platforms.For one,learning is a sequential behavior,and the sequence is composed of sessions,where behaviors in the same session are closely related to each other and are highly homogeneous,and users’ casual behavior in the session deviates from the original expression of the session interest.In addition,there are still problems with data sparsity and cold starts in course recommendation systems.Finally,traditional recommendation algorithms,which tend to consider students’ preferences as static,ignore the fact that users’ long-term preferences change over time.In this paper,we propose a hybrid preference course recommendation model incorporating session interest extraction to solve the above problems.The details of the research are as follows.(1)A course recommendation model based on session interest extraction is proposed to address the problem that users’ casual behavior in a session deviates from the original expression of session interest and the items in the session are highly homogeneous.Considering the high-level complex user-item or item-item interactions,an attention layer is introduced in the model while the user embedding vector is contextual information to achieve the purpose of eliminating noisy courses and extracting session interests.Session interest at different moments has a different importance to the user,and the model uses the second attention layer to simulate the complex interactions between sessions and finally obtain the user interest representation.The experimental results show that the model can effectively improve the performance of the recommendation system.(2)A course recommendation model based on dynamic interests and short-term interests is proposed to address the problem that users’ interests change over time and the cold start problem in online course recommendation.The GRU in the model simulates the dynamically changing interests of users,and the hidden state of each moment of the GRU is input into the attention mechanism,so as to distinguish the importance of dynamic interests to users at different moments.At the same time,the amount of data required for short-term interest is small,and the most recently clicked course is an important reference factor in the recommendation,so the short-term interest of the user is integrated into the third attention layer to obtain a mixed interest representation of the user.Finally,the candidate course vector is combined to generate a recommendation rating ranking table for personalized recommendation service.The experimental results show that the algorithm has good recommendation performance.