Study On User Interest Evolution And Personalized Recommendation Under Social Influence

User interests are always changing over time,which makes time a very critical contextual information in the recommendation systems.Therefore,many studies have proposed the time-based dynamic recommendation algorithm to track changes in user preferences.However,few of these methods have noticed that social information also has a significant impact on the user's changing interests.In addition,the success of online social networks has made users' behavior in social network available for further study together with information used in traditional recommender systems.Nowadays,most studies incorporating social information into recommender systems to improve the performances have been developed based on the homophily principle,which is that users and their social friends tend to share certain common interests or have similar tastes.This assumption ignores the most critical part,that is,the true role played by social friends especially in situations where the user's interest changes over time.According to the social influence theory,users' attitude and behavior are always infected based on information,attitudes,and behaviors of social friends in the network.Few studies pay attention to this perspective.In this thesis,we analyze the impact of social information on the change process of user interest based on real datasets and introduce two assumptions about the impact mechanism.Through the different design of the input-output hidden Markov model structure,we model this two influence mechanisms and proposes three dynamic recommendation models based on social influence.The expectation maximization algorithm is used in this thesis for learning of model parameters.The proposed model learns the dynamic changing process of user interest and integrates the influence of social friends during it,which can better predict the user's future interests and make recommendations.We compare the proposed method with various types of recommendation methods through experiments using two real-world datasets.The experimental results show that the algorithm can effectively improve the recommendation quality of the recommendation system.Furthermore,the validity of the social impact mechanism proposed in this thesis is further verified,and the value of social information and dynamic time information in the recommendation system is confirmed again.