Web Service Recommendation Based On User Trust Network And Preferences

As a new distributed computing model based on Internet standard and XML technology, Web service has played an increasingly important role in distributed platform, such as e-commerce and integrating enterprise applications. With the exponential growth of Web services, how to perceive user requirements proactively, mine potential user preferences and offer them their most interesting service lists, has become the hot topic in the research field of Service Computing. The most commonly used recommendation algorithm on Web service is Collaborative Filtering, including user-based CF, item-based CF and the hybrid algorithm. The problems often occur in CF algorithms are Data Sparsity and the Cold-start Problem. According to these shortcomings, we combine social network to service recommendation algorithms and propose an algorithm based on user trust network and preferences to offer more effective recommendations for users. The contribution of our paper is as following:First of all, we propose a multi-QoS based similarity computing method, which can calculate similarities of services with kinds of QoS values and users who invoke them directly. On the basis of this, we present a recommendation algorithm named URPC-Rec, i.e. User Relationship and Preference Clustering and Recommendation, which first clusters services, transform user-service matrix into user-service_category matrix to reduce the dimension and obtain sufficient user information and relationships. By digging out the associations between users and service categories, we can divide users into different Interest class and extract significant user characteristics. Then we can combine user information and interest label with them to solve the cold-start problem and make reasonable recommendations for new users.Secondly, we propose a method to construct the user trust network according to user information so that we can make the most of user information from social network and mine user potential relationships. Meanwhile, we can use Principal Component Analysis to optimize URPC algorithm (named URPC-PCA) as to make strict division of user relationships. Combining user trust network and URPC-PCA together, we can get the service recommendation algorithm based on user trust network and preferences (called UTWP-Rec). Compared with URPC-Rec, this algorithm extends the scope of services which can be recommended and raises the filtering standard at the same time, so it can make suitable and credible services.Finally, we develop the service recommendation tools and make sufficient experiments on URPC-Rec and UTWP-Rec algorithms. The results show that, both of them have higher recommendation accuracy than the traditional collaborative filtering algorithm and UTWP-Rec performs much better than URPC-Rec.