Research On Personalized Recommendation Algorithm In Trust Social Network

With recently rapid Internet developing,various information on social networks has exploded massively,which makes information-overload problem be further serious.The one of user-faced problems is how to find interested or valuable information in such a huge data environment quickly and efficiently.Hence,various recommendation systems have emerged as powerful tools to provide efficient user decision-making.In this regard,a user-preferred and recommended-precisely algorithm is crucial.Currently,collaborative filtering algorithm is a mainstream one in personalized recommendation systems,in which social network information plays a pivotal role on preference-mining.But,defections such as cold start,data sparseness,information timeliness neglect,etc.still exist,which can cause unexpected recommendations.To solve above-mentioned problems,the work of this paper is arranged as follows:As current collaborative filtering algorithms cannnot calculate similarity for users on different-user-scored items,a personalized recommendation algorithm based on user clustering in trusted social networks(TSN)is proposed in this paper.And,the TSN-based similar-user set of target users is used to maintain inter-items similarity transfer,so that the proposed algorithm alleviates the recommendation cold start problem to certain extent by target user's neighbor set.To improve recommendation accuracy,an information-timeliness-integrated personalized recommendation method is proposed.Firstly,guide the recommendations with same implicit features existed between user rating matrix and user relation matrix in social network.Secondly,adjust the calculation method of items similarity with items information timeliness.Thirdly,optimize the recommendation list from target users to obtain efficient recommendation results.To deal with user multi-feedback information and weight balance ignored by current social network recommendation methods,a graph-entropy based personalized recommendation algorithm is proposed to adapt to TSN.Inspired by random walk algorithm,walk on user graph and trust graph until all similarity values converge.To get final similarities for recommendation,graph-entropy is employed for balancing the two similarity sets.Finally,verification experiments on all proposed recommendation algorithms are carried out.The results show that not only cold start,data sparseness and information overload are significantly alleviated,but also the recommendation quality is effectively improved.