Improvement Of Collaborative Filtering Recommendation Algorithm

With the popularity of e-commerce and the exponential increase in the number of websites,the user's life has become rich and colorful.At the same time,it also brings certain troubles.Users often feel at a loss in front of massive amounts of data and information,that is,information overload occurs.To solve this problem,a personalized recommendation system was born,which provided personalized recommendations for specific users.The collaborative filtering algorithm in personalized recommendation system,because of its simple and easy-to-implement characteristics,has been widely studied and applied in a large number of applications.However,the accuracy of the collaborative filtering recommendation algorithm is not high because the user scoring matrix is often very sparse and the similarity calculation is not reasonable.In addition,the purpose of the recommendation system is that the recommended item is approved by the user,and accuracy is one of the aspects.It is also important that the user feels novel about the recommended item.Therefore,from the aspects of precision and novelty,this article has done the following work to improve the user's satisfaction with recommended items.(1)In order to improve the novelty of the recommended items,a collaborative filtering algorithm based on the degree of controversy and trust was proposed.The algorithm first introduces the degree of controversy between users into the calculation of similarity,obtains the similarity of the controversy,and uses the similarity of the controversy to fill the sparse matrix.Second,calculate the indirect confidence of the filled matrix,and the direct trust between users is calculated from indirect trust and similarity.Finally,the users are predicted by combining the degree of controversy and the degree of trust to generate recommended items.(2)In order to improve the novelty of the recommended items,a collaborative filtering algorithm based on novelty was proposed.The algorithm first considers that each individual has different scoring criteria,and corrects the scoring matrix through the baseline offset.Second,the algorithm improves the novelty of the recommended item by reducing the popularity.Finally,the algorithm considers the impact of time on the recommendation and weights the time factor into the traditional similarity calculation formula.The improved algorithm synthesizes the above factors to get the comprehensive similarity,and uses the similarity degree of similarity to select the nearest neighbor for the target user and predict the recommendation result.