Course Recommendation System Based On Weighted Modified Item-CF Similarity

This paper mainly implements the recommendation of the course through collaborative filtering recommendation algorithm,collaborative filtering algorithms have the advantages of simple engineering implementation and strong model versatility.However,recommendation systems based on collaborative filtering still has many deficiencies in practical applications,such as the problems of cold start and scalability in the application of course recommendation.Cold start problems cause new projects or new users to get inappropriate recommendations or even never be recommended,scalability problem in the rapidly increasing of users and projects become more and more serious.If the computational complexity of the algorithm is too high,the recommendation timeliness will be degraded.The scalability of the recommendation algorithm in mass data needs to be solved.From the above two perspectives,this paper has done the following research:First of all,consider the practical application of course recommendations,this paper user Item-CF to comply the recommendation system.Through the analysis and experiment of different similarity measure methods,then observe the influence of the recommendation results' accuracy in different similarity measure methods.The experiments show that the recommendation accuracy rate based on the cosine is the highest.Secondly,considering the sparsity of the user rating matrix in Item-CF will result in some courses can't calculate the similarity,this paper proposes to fill the similarity of course preferences by the similarity of course attributes.On the one hand,through the filling of similarities,the problem that the similarity between some courses cannot be calculated is solved.On the other hand,the course attributes changes slowing and the course data's features are available from the description of courses,somehow,problem of cold start is also mitigated.Experiments show that although the proposed algorithm has higher accuracy than the traditional Item-CF when the neighbor size is small,then improved the efficiency with the same recommendation accuracy,but the improvement is not significant,only about 3.5%,and far less than the maximum accuracy rate that the algorithm can achieve.Finally,based on the analysis of experimental and student learning behaviors,in order to improve efficiency without losing the recommendation accuracy,this paper proposes a weighted modified similarity algorithm(WHSCF)based on popularity of course.Experiments show that WHSCF can improve the efficiency and accuracy in the course recommendation compared to the traditional Item-CF.When course neighbor numbers are small,the accuracy can increased by 11.8%,and the accuracy is close to WHSCF and Item-CF.At the same time,the traditional Item-CF algorithm needs the neighbor numbers are greater than 40.The WHSCF greatly improves the efficiency of the course recommendation,greatly mitigation the scalability issue of recommendation system.