Research On Personalized Recommendation Method Based On Trust Relationship

Recent years,with the development of Internet and e-commerce,people can enjoy convenient life,but it makes the storage of information more and more difficult.Because the sharp growth of data,for people it is difficult to find the useful information from the massive data.So the benefits of personalized recommendation become more and more obvious.Personalized recommendation not only can help users filter out the products they are interested in by analyzing the feature of users and items,but also can help businesses increase their income through virtual marketing.In academia,it has become a hot topic in data mining and social network.Collaborative filtering is the most mature and widely used recommendation algorithms,but it still has some problems which need to be solved,such as data sparsity,cold start,scalability and so on.In this paper,multiple social relations among users are considered to integrate into the basic matrix factorization model.The measures of computing users' trust-similarity degree and expert-trust degree are designed and implemented,and then two kinds of recommendation algorithms are proposed.The first is trust-similarity degree based recommendation algorithm,and the second is a recommendation algorithm fusing expert-trust degree.It aims to use the users' social characteristics to alleviate the data sparsity problem which the traditional collaborative filtering can not handle and to improve the effect of recommendation.The main contents of this paper are as follows:1.For the data sparsity problem,a recommendation algorithm based on trust-similarity degree is proposed,which integrates users' social information into recommendation algorithm by using an effective trust-similarity measure.The advantage of this algorithm is that it takes into account the influences of both the users' social relations information and their historical score data.This relieves the problem of data sparsity.And the experiment on some datasets,such as Ciao DVD etc,verifys the effect of our algorithm.2.For the users' choices are often influenced by expert opinions.A measure to calculate the expert-trust degree of users is designed.And based on this new measure,a new recommendation algorithm TSCE fusing expert-trust degree of users is proposed.The algorithm improves the accuracy of recommendation by introducing the regularization term of expert-trust degree to constrain the objective function of matrix factorization.Moreover,users' preferences may be different according to different category of items.So the concept of circle is introduced,which is composed by the social relations of users who rate the same category items.Training and prediction process are respective for different circles.Finally,the experiment on Yelp dataset verifys the effectiveness of our algorithm.3.In this article,the Lib Rec recommend framework is expanded to implement the two recommendation algorithms.Firstly,add the trust-similarity functional module to calculate the trust-simlarity degree of users,and add the user-influence functional module to calculate user's expert-trust degree;secondly,in the original recommendation algorithm module add the implementation of TSMF,TSCE algorithms;finally,analyze the application scenarios of the two algorithms.