Hybrid Recommendation Algorithm Based On Similarity Of Feature Attributes

With the development of network technology,the Internet contains more and more data,which leads to the problem of information overload,it is difficult and timeconsuming for users to find information.Recommendation system is an important tool to solve this problem and is widely used in many aspects.All kinds of e-commerce systems need recommendation system to support.The key problem of recommendation system is the accuracy of prediction and recommendation.In order to solve the problem of prediction accuracy,the multi angle similarity of dynamic evolution clustering algorithm and matrix decomposition algorithm in collaborative filtering is mainly studied in this paper.(1)Multi-angle similarity algorithm of dynamic evolution clustering.Aiming at the problem of low precision of traditional collaborative filtering recommendation algorithm,a dynamic evolution clustering algorithm is proposed in this paper.And the proposed method improves the traditional similarity by considering the connections between the user and the project from several aspects.The heterogeneous matrix is constructed by user-item scores.Users and items are clustered by difference equations.The number of items to be shared by users and the user's interest in the project is not considered in traditional similarity.In this paper,user similarity score,user interest similarity and user common score are weighted fusion to get the final similarity.Our algorithm is verified on Movielens Data set,and the experiment results show that the algorithm improves the accuracy of predictive score.(2)Collaborative filtering recommendation algorithm based on similarity matrix decomposition.Observing the significance of the initial matrix representation in matrix decomposition,generate the initial matrix by singular value decomposition for useruser and item-item similarity matrix is proposed in this paper.It can fully explore the relationship between users and users,items and items,and it can reduce the errors caused by too many zero elements or random generation in the matrix.This algorithm is verified by two Movielens Data sets,and the experiment results showthat this algorithm improves the accuracy of predictive score and the recommended accuracy of the prediction.