Research On Trust Network Based Rating Prediction Algorithm

Information overload is more and more serious with the rapid development of the Internet. Recommender system is an effective tool to solve the information overload. As an important means of information filtering, it can present information which people may be interested to them and provide personalized service. Trust-based recommender algorithms improved cold start and data sparse to some extern existing in traditional collaborative filtering algorithm, which became the research focus in the field of recommendation.Because of the complex trust relationship in the large-scale trust network, the cost of recommendation algorithm is very huge. Therefore, proposing a network simplification algorithm and a rating prediction algorithm based trust network. Analyzing and researching trust network from constructing, path dependent elimination and simplification of trust path in the network simplification algorithm improves the accuracy of trust measurement and reduces the cost of the rating prediction algorithm. Taking into account that trust-based recommendation algorithms didn't utilize users clustering feature and historical recommendation information, the cost of algorithm is very expensive. The rating prediction algorithm divides users into different communities based on trust relationship, and uses ant colony optimization to search target item which considers the pheromone as experience information in different communities and trust value as heuristic information between users, and completed the rating prediction algorithm after network simplification.In order to verify network simplification algorithm and rating prediction algorithm,the experiments utilized the Epinions dataset are carried out by Matlab platform. The result showed that, the proposed trust network simplification algorithm can remove a lot of nodes and trust relationship, and a large amount of important information can be retained. Rating prediction algorithm performs better than TidalTrust, TrustWalker and T-bar in accuracy, coverage and cost. And rating prediction algorithm shows better performance after the simplification of trust network.