Recommender Systems Research Based On Social Influence

With the explosive growth of information,the emergence of recommender systems provides an effective means to solve the problem of information overload.It can help users to extract useful information content from large amounts of information on the Internet.First,it analyzes the historical behavior data generated by the user on the Internet,and then builds the users interest model on that basis to provide them with content that they like or need.In recent years,recommender system research is widely used in computing advertising,electricity and video sites and other fields,to enhance the commercial value.However,the traditional recommendation algorithm ignores the social influence of social networks in recommender systems.In fact,in a real life,people often choose products that friends recommend for them.As a result,social influence is also an important source of enhancing the recommendation performance.With Microblog,WeChat and other social media popular,more and more people began to study on the basis of social networks,the use of social relations and other information to improve the recommendation quality.On the basis of analyzing the related research in the field of recommendation system,this paper finds out that the social influence of users can improve the recommendation performance effectively from the social relations among users.However,at present,in some social recommendation algorithms,they just think that the social impact is single and static,ignoring the diversity,dynamic changes and other characteristics of the social influence.It is clear that this is not consistent with the facts.In addition,users in the social network tend to show different behavioral tendencies.Users' role information and behavioral characteristics are closely related,and some of the existing social recommendation algorithms often overlook the user's role diversity.Therefore,this article mainly aims at solving the existing problems in the social recommendation system from these two aspects.Recommender systems need to mine the users' interest and predict their behavior,and the social influence is an important means to explore the user interaction.By analyzing the user interaction network,this paper learns the important role of multi-dimensional social influence in mining users' potential interests.However,the users' interest is changing over time,the social influence is also dynamic,and so the purpose of this paper is to use dynamic social relations of users to recommend interesting goods.This paper models the users' interests changes caused by users' interactions,proposed a probabilistic graphic model referred as IRDMSI to integrate into the dynamic multi-dimensional social influence.This paper finds that the model not only brings a good recommendation performance,but also reveals some common rules of dynamic interactive behavior of users in social network.In addition to the social network between users,we analyze the social influence from the social theory point of view.Then,different types of users(users of different roles)may have different conformity tendency.Most of the existing recommendation algorithms assume that the users' roles are single and ignore the users' diverse role information which plays an important role in the social recommendation system.Therefore,we have studied how the conformity tendency changed with users' roles in recommender systems.We first define a utility function to formalize the conformity influence,and then put forward a probabilistic graphic model to combine users' roles and conformity influence,that is,role-conformity recommender systems(RCRS).The model allows us to recommend items for users by using potential factors and potential roles as features.We evaluate the model on several data sets,and the experimental results show that our model is far more than some of the existing benchmark algorithms.