Research On Trust Based Collaborative Filtering In Dual-Structural Network

To alleviate the "information overloading" problem in the Internet, Dual-Structural Network was proposed which consists of traditional TCP/IP network as the primary structure and Broadcast-storage structure which means broadcast and distributed storage as secondary structure. However traditional collaborative filtering algorithms cannot suit this application environment well with many limitations such as data unreliability, data sparsity, user cold start, etc. In order to solve above problems, the thesis analyzed the role of trust in recommendation scenario of reality, people are often willing to accept recommendation from trustworthy people in life. Since trust information can be used for solving the data unreliability problem, supplementing sparse data and avoiding the negative effects of the data sparsity on recommendation. The thesis took advantage of trust to find the authority user in system to help new users make recommendations so as to effectively alleviate cold start problem. Therefore, this thesis proposed collaborative algorithms based on trust mechanism by introducing the concept of trust into traditional algorithm to solve those difficulties.This thesis took features and requirements of the Dual-Structural Network into account, and researched, designed, implemented the personalized recommendation mechanism in Dual-Structural Network based on trust mechanism. The main work of the thesis is reflected in the following aspects:? In order to solve the problems of data sparsity and unreliability, this thesis proposed a collaborative filtering algorithm called DS-CFAIT based on the implicit trust mechanism according to the features and requirements of Dual-Structural Network, which can be used to find out the implicit trust relationships and similarity relationships among users by virtue of their historical information and complete recommendations. On this basis, the thesis also researched how to solve the cold-start problem with implicit trust and introduced time to the implicit trust model.? To further improve the trust mechanism and satisfy the requirements of explicit trust between users in Dual-Structural Network, the thesis proposed an algorithm called DS-HTrustSVD, under which a mixed model was designed by merging the explicit and implicit trust relationships between users and recommendations were completed for users by using Single Value Decompose method. Therefore, adopting this algorithm can enrich trust relationships and acquire more accurate trust values, which can make more reliable recommendations.? All the improved algorithms were made experimental analysis and conducted performance verification on dataset. They were designed and implemented in the prototype system of personalized recommendation mechanism of Dual-Structural Network to verify feasibility. Experiment results showed that DS-CFAIT algorithm and DS-HTrustSVD algorithm could effectively improve the recommendation performance and the user experience in the Dual-Structural Network.