Research On Collaborative Filtering Recommendation Algorithm Based On User Interest

With the rapid development of Internet technology,the recommended system(RS)can help users to make the appropriate decisions quickly,are widely used in major e-commerce sites,recommended promote the user from the browser to the conversion between the buyers to the production Business opportunities.Collaborative filtering(CF)is the main technology in the field of RS,which can effectively solve the information overload problem.CF is based on the user is more likely to adopt the RS given by friends,using the nearest neighbor to study the user's interest characteristics,and to personalize the RS by predicting the user's interest.In the user-project matrix,the amount of scoring data is rather sparse compared to the number of user items,resulting in a decrease in recommended adoption rate and poor user experience.In addition,CF also has problems such as cold start,poor scalability and lack of consideration of dynamic changes in user interests,resulting in low accuracy of RS results.Therefore,we need to further study the above issues.(1)The neighborhood-based CF algorithm only uses the common score between users,and the data shortage and the size of the project are large,which leads to the seriousness of the user-project matrix sparseness,and the similarity measure and the actual deviation,CF based on Bhattacharyya and Jaccard coefficient was proposed.This method measured the similarity by introducing Bhattacharyya and Jaccard coefficient,Bhattacharyya coefficient could utilize all ratings made by a pair of users to get rid of common rating restrictions,Jaccard coefficient could provide more importance to the number of common items,which improved higher accuracy to select nearest neighborhood and optimize the preference prediction and personalized recommendation of the active user.The algorithm is published in the Computer Application and has been published.(2)The user-based CF focus on how to efficiently use historical score to calculate the similarity,ignoring the timeliness of scoring data.In this paper,considering the influence of time factors,a recommendation algorithm for user interest offset and clustering is proposed.Firstly,the project similarity degree and project association graph are introduced.The project association graph summarizes the project aggregation and shortens the time of recommendation list generation.Then,the interest model is established according to the association graph.The use of sequence partition algorithm to identify the user interest with time changes,the deviation of a certain moment of interest to the user only to retain the time after the historical data;Finally,the use of neighbors to predict the recommendation.Experiments are carried out from clustering and recommendation,and it is proved that the algorithm can improve the accuracy of clustering and recommendation.This algorithm has been submitted.