Research On Collaborative Filtering Recommendation Algorithm Based On User Trust And Interest

With the continuous popularization of information technology and the rapid development of the Internet,the problem of "Information Overload" is getting worse.How to provide users with better products that may be of interest to them and help them to confuse the confusion caused by a large amount of information has become a research hotspot in recent years.The recommender system provides users with personalized recommendations based on their interests and hobbies,to a certain extent,mitigated the negative impact of "information overload",especially since the 1990 s,the research results of recommendation systems have been emerging.At the same time,with the number of Internet users has exploded in recent years,social networking emerged suddenly.Research conducted by the famous Nielsen research firm in the United States on "influencing users to trust a recommendation" showed that about 90% of users tended to trust recommendations from friends.Related experts and scholars began to study the recommendation algorithm based on social trust.The results show that the introduction of social trust into the recommended areas can significantly improve the quality of recommendations.This paper mainly studies based on Collaborative Filtering Recommendation(CF)which is widely used in the field of recommendation system.Traditional CF has sparse data,cold start,new users and other issues,affecting the recommended accuracy and quality.On the other hand,due to the huge number of users,there is data sparse problems of direct trust between users.Therefore,firstly,we propose the user behavior coefficient to mining the user's implicit trust relationship,and propose the user trust function combined with user's explicit trust relationship to alleviate data sparseness of user's trust relationship further.Secondly,we propose the concept of “user interest similarity” combined with user rating and the project attributes tag to mine user's potential interest;Thirdly,we use probability matrix factorization model to conduct matrix decomposition of user ratings information,users trust relationship,user interest label information,and further excavate the user latent characteristics to ease data sparseness.Then we propose a probability matrix factorization model(STUIPMF)which alleviates data sparseness.Finally,we use experiment based on the Epinions dataset to verify our proposed method.The results show that the proposed method can improve the recommendation accuracy to some extent,ease cold-start and new user problems.Meanwhile,the STUIPMF approach we propose also has good scalability.