Study Of Recommendation System Based On Double Clustering And Trust

With the popularity of the Internet and the rapid development of electronic commerce, Internet is flooded with a lot of redundant information, people are faced with how to quickly and accurately find useful information from the mass information which is called information overload. In this context recomm end system emerge as the times require. People generally believe that collabora tive filtering algorithm is proposed to mark the birth of the recommend system. Recommend system has been widely used from e-commerce, music video website, to the pillar of online advertising, and the Internet is filled with the figure of recommend system.However, because of the Long tail and Matthew effect, traditional collabor ative filtering recommendation algorithm is facing a serious problem of data sparse, cold start, security and extensible property.In recent years, in the study of recommend system, some researchers integ rate the clustering analysis technology in data mining into the recommendation system to solve some problems. The recommendation algorithm based on the fusion of clustering analysis firstly cluster it according to the attributes of users or items. Using cluster analysis according to the attributes of users or items clustering with similar properties of the user or program to gather a cluster, effe ctively reduces the size of the data objects, reduces the complexity of the algo rithm of recommendation system. But the traditional clustering algorithm does not take into account the impact of the user’s social relationship, and collaborative filtering algorithm based on clustering is still affected by data sparseness. This is because this kind of algorithm is still using the user-item rating to calc ulate the degree of similarity.Through the analysis of these problems, this thesis proposes a collaborative filtering algorithm based on double clustering and trust. This algorithm using user-clustering matrix to measure the user’s degree of similarity, which could relieve the sparsity problem and reduce the dimension of the user-item matrix. Taking into account the huge impact of expert users in users, we choose expert users as initial clustering centers. We introduce item-level trust into the process of predicting rating. Through several improvements of this thesis,we could effectively relieve the sparsity problem, and improve the accuracy of the results.