Improvement Of Collaborative Filtering Recommendation Algorithm

With the growth of information on the Internet and the explosive growth of data,the way users get information is constantly changing,which has experienced the evolution of portal,search engine and recommendation system.Recommendation system is also a popular recommendation system for the current personalized recommendation system.The so-called personalized recommendation refers to the recommendation of the same time,to ensure that different users get in line with their personal information,with a better user experience.Collaborative filtering technology is personalized recommendation technology is the most successful and most widely used technique that by calculating the similarity between users to select the target user's nearest neighbor set,and then by scoring nearest neighbor set to predict the target users Unrated The project score,resulting in recommendations.However,since the user program ratings extreme sparse matrix and other reasons,collaborative filtering recommendation precision technology still has much room for improvement.From collaborative filtering algorithm to improve the precision in recommending starting to do the following two things work.(1)The degree of trust between the introduction of user similarity computation formula,is proposed a user confidence improved algorithm based on collaborative filtering.As the traditional collaborative filtering algorithm ignores the trust between different users direct impact on the calculation of similarity,leading to the recommendation accuracy is not precise enough.Improved algorithm,first by similar rates among users,authority and universality to calculate direct trust between users,and then calculate the indirect trust between users through trust transfer mechanism,the resulting composite confidence;then,the traditional collaboration Comprehensive filtering trust similarity between users combine to generate integrated similarity;and finally,the use of composite similarity score to predict,and ultimately generate recommendations.(2)Considering the effect of the average score,the popularity of the project,and the different characteristics of the user's score on the similarity computation of traditional collaborative filtering algorithm,a new collaborative filtering algorithm based on different features and reducing popularity is proposed.In the process of similarity calculation,the authors consider that the user's score behavior is different,so the similarity calculation formula can be weighted by different features.Considering that the average score of the two users can be more similar,so that the degree of popularity of the project will affect the user's similarity.With calculate the comprehensive similarity between users,and make a comprehensive evaluation of the target user score prediction,finally we can recommend.