Research On Course Recommendation Based On Self-attention Mechanism

With the development of information technology,people’s access to knowledge becomes more information-based,and online learning has gradually become an indispensable way to acquire knowledge,helping users to acquire knowledge anytime and anywhere.However,with the continuous increase of educational resources,more and more courses are available for users to choose from,and users gradually find it difficult to find appropriate learning materials from the mass of information.Due to its high efficiency of filtering information,the course recommendation system can help users quickly find personalized resources suitable for themselves,thus alleviating the problem of information overload in the learning environment.Therefore,it is of great practical significance to build an efficient course recommendation system to help users quickly find the courses they may be interested in and recommend resources to them according to their learning situation,which is of great practical significance to improve the learning efficiency and effect of users.At present,there are still few researches on curriculum recommendation system.Most of the existing researches on curriculum recommendation system directly carry the popular algorithms in the existing recommendation system,ignoring many important characteristics of the curriculum learning environment.For example,students usually choose courses in a certain order,and learners usually choose courses according to their own learning reserves and interests.Ignoring these characteristics will lead to the recommendation effect made by the recommendation system is not optimal.In this thesis,a curriculum recommendation algorithm based on self-attention mechanism is designed and proposed according to the actual needs of learners for online learning platform.Specifically,this thesis makes innovative research on course recommendation in the following four aspects:(1)We propose a new classification framework.According to whether the time order of user’s course selection record is considered,the existing recommendation model is divided into general recommendation model and sequential recommendation model.Then we introduce the representative research work under the two categories in detail and point out that the current research on course recommendation system ignores the order of course selection,which leads to the recommendation effect is not optimal.(2)We propose a course recommendation algorithm model based on self-attention mechanism and apply self-attention mechanism to course recommendation for the first time.By using the selfattention mechanism to model the user’s interest in selecting courses,the dynamic change of user interest with time is captured,and the features of the user’s historical course selection sequence data are efficiently extracted.(3)Traditional course recommendation only considers the user’s course selection record,ignoring other auxiliary information of the user and the course itself,which is valuable to the course recommendation system.Making full use of this information may improve the accuracy of recommender systems.Therefore,this thesis proposes to add the type information of the course to verify the influence of the addition of side information on the recommendation effect.(4)The model was further improved by designing a building module of user’s short-term preference to capture the influence of the user’s recent courses on user preference,adding user representation and introducing random sharing embedding to enhance user’s personalized representation.