Recommendation Algorithm Based On Time Function And User Attribute

With the development of science and technology,the internet has became the main source of information for people.The data generated by the internet is growing rapidly,making it difficult for users to find what they really want.How to quickly find the u ser's favorite content from the "information ocean",is a hot topic of current researches.The recommendation algorithm uses user evaluational data to calculate the matching degree between users or projects,and achieve the purpose of pushing items to users.But the recommendation system still has problems such as the sparseness,cold start,prediction accuracy of user evaluational data,and the challenge of user interested changes.This paper research and analy these challenges.Based on the user's interest drift and multi-angle description of user preferences,the recommendation algorithm of fusion time function and user attribute is proposed,which improves the accuracy and diversity of recommendation.The main contributions of this paper are as follows:1.The matrix of user evaluational data is sparse,matrix decomposition model is proposed to reduce and fill the score matrix to obtain a new user score matrix,which provides a data foundation for the input of the latter algorithm.2.According to the problem of user interested changes with time,the time function is fitted according to the forgetting curve to characterize the change law of user interest.The weights of the predicted scores are given at different times and integrated into the calculation of the user similarity,thereby improving the effectiveness of the user's evaluational score.3.The problems of cold start and prediction accuracy occur in the recommendation process,user attributes are introduced to diversify the user's characteristics,the attributes of gender,age,occupation,region are discretized.the user attribute matrix is constructed to calculate user attribute similarity,finally,we can weight the user evaluational score and user attribute similarity to obtain a new comprehensive similarity.According to the comprehensive similarity,The paper calculate the nearest neighbor set,then push the items to target user,thereby this process improve the pushing quality of the recommendation system.This paper propose a recommendation algorithm based on time function and user attribute.Algorithm is tested on the MovieLens data,one experiment is parameter setting on the matrix decomposition filling algorithm and the model,another experiment is the effectiveness and superiority compares in the recommendation algorithm based on time function and user attribute proposed in this paper with other algorithms.The experimental results show the recommendation algorithm proposed in this paper is the best and the factor are as follows: 1.the regularization coefficients of users and projects are 50 and 10 in the matrix decomposition filling algorithm.2.The number of hidden factors is 175.3.The weights of user scores is 0.7,and the weights of user attribute similarities is 0.3.In addition,when the number of nearest neighbors is 30,each algorithm have the highest prediction accuracy.The accuracy of recommendation algorithm based time function and user attribute is improved 12.04%,10.15%,and 2.85% with recommendation algorithms of SC-CF,NCF and SS-MF.