Research On Personalized Recommendation Algorithm Integrating Social Trust And Influence

The 21st century is an era of information overload.Every day,countless news,advertisements,emails and other new information are generated at a rapidly increasing rate.In order to help people find the information they need more accurately and efficiently,Recommendation Systems(RS)are increasingly being applied to various portals and e-commerce systems.Despite this,cold-start and data sparsity still affect the performance of RS.With the popularity of social network platform,social network information is integrated into the traditional Collaborative Filtering(CF)to alleviate the impact of sparse rating information on recommendation performance.However,the existing social recommendation algorithms generally focus on trust relationship,but ignore the influential users,that is,those who can influence the opinions and opinions of other users through the network structure also play a great role in the final decision of the user.At the same time,most existing trust-aware recommendation systems simply utilized the social network information with binary format directly,and thus ignore the fact that trust values between each pair of users are ful l of variety.Based on these two main issues,this thesis conducted the following research to further enhance the performance of the recommended algorithm:(1)In order to solve the problems of trust data sparseness and binary trust relationship,a new trust metric model combining user interaction information and preference is proposed to explore the implicit trust relationship between social network users and build a social trust network.(2)A key node identification algorithm based on structural holes is proposed,which uses the self-property of the nodes in the complex network and the topological structure of their neighbors to evaluate the importance of nodes,so as to mine the influential users(key nodes)in the social network.(3)Aiming at the Top-N recommendation scenario,considering the influence of user trust and influence on the recommendation effect,a top-N recommendation model FSTID based on the combination of user and item similarity,trust and social influence is proposed.Aiming at the insufficiency of the traditional collaborative filtering algorithm to randomly initialize the user and project feature vector,the Auto encoder is used to unsupervised learning the initial feature of user behavior to improve the recommendation performance.Finally,the comparison is verified on the three standard datasets of FilmTrust,Epinions and Ciao.The experimental results prove the efficiency of our algorithm.(4)Based on the SVD(Singular Value Decomposition)model,a personalized recommendation algorithm ITSVD(Influence and Trust based on SVD),which integrates social trust and influence,is proposed for the recommendation task in rating prediction scenarios.When considering the explicit and implicit influences of trusted users on the target users' rating prediction,the influence of high-impact users is further integrated.Experiments on two real world education datasets show that our algorithm can effectively improve the accuracy of rating prediction.