Research On User Characteristics And Item Association Degree Based Collaborative Filtering Recommendation Algorithms

The advent of the information age makes it difficult for users to quickly obtain data that is useful to them.In order to help users obtain effective information from massive data,personalized recommendation algorithms are applied.According to the user's historical browsing records,interest categories and specific search terms,it can mine the information that users really need,and then generate recommendations for users.At present,collaborative filtering algorithm has the advantages of personalized recommendation,high automation,effective use of feedback information from other similar users,and accelerating personalized learning speed.It has become a hot spot in the research of recommendation search algorithm.However,with the increasing amount of information,this algorithm has the problems of cold start,recommendation accuracy,data sparsity and recommendation singularity.To solve the above problems,this paper proposes a collaborative filtering recommendation algorithm based on user characteristics and item association degree.The research contents of this paper are as follows:Firstly,aiming at user-based collaborative filtering recommendation algorithm which only considers user ratings,this paper proposes three key issues to characterize user characteristics,including:(1)To solve the problem of cold start,this paper puts forward a comprehensive consideration of the user's multiple characteristics,such as gender,age,occupation,etc.By weighting the influence of the three characteristics on similarity according to the corresponding importance with weight coefficient,the similarity of attributes among users can be obtained,the recommendation of new users can be realized,and the cold start problem can be improved.(2)To solve the problem of low recommendation accuracy,a method of establishing user interest characteristics by user rating is proposed,which calculates the similarity of user interest,pushes the user's favorite items,and improves the recommendation accuracy.(3)To solve the problem of data sparsity,a method of building user trust features is proposed.The trust features are divided into direct trust and indirect trust.Direct trust represents that users can trust each other directly.Indirect trust can be transferred through direct trust between users.The trust feature matrix obtained by indirect trust transfer can be filled in the original score matrix.The problem of matrix sparsity is improved.Secondly,aiming at the recommendation singularity of recommendation algorithm,this paper proposes a method that uses association rules to find frequent itemsets among items,and then combines it with collaborative filtering algorithm based on user characteristics to make hybrid recommendation.This method can recommend related items for users and improve the singularity of push results.This paper uses the Movielens site data set to verify the accuracy of the algorithm.The experimental results show that the proposed algorithm has a significant improvement in push accuracy compared with similar algorithms.