Improved Collaborative Filtering Recommendation Algorithm Based On Trust Relationship

With the increase of online user demand,the recommendation system came into being.The collaborative filtering recommendation algorithm is the most widely used algorithm in the recommendation system,but the traditional collaborative filtering recommendation algorithm has data sparse and cold start problems.In order to solve these two problems,the researchers tried to fuse the trust information to recommend the user,which resulted in a trust-based recommendation algorithm.The trust-based recommendation algorithm integrates the user's interest preferences and social relationship information.It has better performance on indicators such as accuracy and robustness of recommendations,but there are problems such as sparse trust data and unreliable trust relationships.In this paper,we focus on the problems of trust recommendation and improve the algorithm from the aspects of trust modeling and trust metrics.The details are as follows.This paper first analyzes the research status of collaborative filtering,briefly describes the existing problems of the recommendation algorithm and the research improvements made by researchers at home and abroad,and introduces the trust-based recommendation algorithm in detail.Secondly,for the problem of sparse data and single form of trust,the trust matrix is used as a supplement to the scoring matrix,together with the singular value decomposition algorithm.At the same time,the relationship between user trust and trust is considered,and trusted users and users are added to the predictive scoring model.Trust the user's influence in two aspects,and optimize the objective function to learn the best model parameters.In order to make full use of scoring and trust information to better build a recommendation model to improve the accuracy of scoring prediction.Thirdly,aiming at the problem that the trust relationship is not reliable and the recommendation accuracy is low,the Pearson similarity algorithm is improved by considering the degree of project feedback,and the trust matrix is de-pseudo-accepted according to the improved similarity.Then through the spread and aggregation of trust,consider the influence of trust attenuation,find more neighbors with different weights,and expand the trust matrix.In order to solve the problem of unsatisfactory recommendation.Finally,the above two algorithms are verified experimentally in real data sets.The results show that the two algorithms can fuse the trust information with the collaborative filtering recommendation algorithm well and improve the recommendation performance to some extent.