An Improved Algorithm Of Neighborhood-based Collaborative Filtering

With the advent of the information age,people's connection with the Internet has become increasingly close.With the rapid development of communication technology and Internet technology,the transmission and acquisition of information have become very convenient and efficient.However,in the face of an increasingly large scale of information,there is a conflict between the user's demand for fast access to effective information and the massive growth of information.As an effective way to alleviate this problem,personalized recommendation service can provide personalized recommendation service according to the behavior characteristics and interest preferences of different users.Collaborative filtering algorithm is the most widely used recommendation technology in the field of recommendation systems.However,with the growth of the number of Internet users and the change of application platform architecture,collaborative filtering algorithm is facing the problems of sparse data and inaccurate similarity calculation.Aiming at the above problems,this paper makes an in-depth study,analyzes the shortcomings of traditional similarity calculation methods,and proposes a measurement method based on user attribute behavior similarity.This method uses fuzzy set theory to construct a fuzzy relation matrix based on user demographic information and uses fuzzy distance function to measure user attribute similarity.Furthermore,the user behavior similarity is generated by using all the user score information.By integrating the above two similarity methods,the user's interest preferences are comprehensively depicted from a multi-dimensional perspective.At the same time,this paper combines the proposed user attribute behavior similarity method with UCF algorithm and proposes a collaborative filtering algorithm based on user attributes and behavior.Through the introduction of user attribute information and rating information,the problem of low recommendation accuracy of traditional UCF algorithm in sparse data sets is alleviated.In the experimental part,Movie Lens data set is used to compare the proposed algorithm in three aspects: the accuracy of scoring prediction,the effectiveness of alleviating data sparsity,and the accuracy of recommendation.Mae,recall,and precision was selected as the evaluation indexes.Experimental results show that the accuracy of the proposed algorithm is improved by 7.5% compared with the traditional Pearson method.On the data set with large sparsity,the recall rate is not easily affected by data sparsity,and the best value can be stable at 18%.The accuracy of recommendation is 16% higher than that of the NHSM algorithm.The above experimental results effectively verify that the proposed algorithm can effectively alleviate the problem of inaccurate similarity calculation caused by data sparsity,and improve the accuracy and stability of recommendation.