A Trust Network Based Recommendation System

With the rapid development of networking and communications technology, the worldwide information technology has changed the way people live. The emergence of the Internet, take us into the era of an information explosion. Facing of large amounts of information, people will often feel at a loss, it is difficult to find the suitable information for their own and the efficiency of the use of information is low. This is the so-called information overload problem. Recommendation system as an important means of information filtering is an effective way to solve the information overload problem. Traditional collaborative filtering recommendation systems and trust-based recommendation systems have been widely used, but there are still many deficiencies and challenges in the data sparsity problem, new user problem and recommendation accuracy. The existing recommendation systems still have the following problems:· Data sparsity problem. Traditional collaborative filtering method recommends by calculating similarity between users. Due to the sparsity of the rating matrix, it is difficult to calculate user similarity. Traditional collaborative filtering methods consider the similarity of the user while ignore the social relationships between users. Trust-based methods use trust value instead of user similarity, and use the transfer of trust value to find nearest neighbor from the non-direct neighbors. But it may find some users with different taste, and decline the accuracy of recommendation.· New user problem. New users just joined the recommendation system and have no evaluation or little evaluation. So it is difficult to calculate the similarity of the new user with other users. Meanwhile, the new users are often isolated nodes in the trust network, or only have a few trust relationships. They do not know whom to trust. Therefore, whether the early content-based methods, collaborative filtering-based methods and trust-based methods, can not solve new user problem in the recommendation system properly.In order to solve the above problem, provide users with more effective personalized recommendation in an open environment, and solve the problem of information overload properly, we analyzed the research status and existing research results of recommendation system, used the trust network and expert users to solve the problems of the recommendation system. The main work is summarized as follows:· A framework of recommendation system based on trust networks. We analyzed current recommendation systems. To solve the existing data sparse problem and new user problem, we introduced the trust network and expert users and proposed a hybrid recommendation method combining trust value with similarity and an expert based recommendation method. Finally we proposed a framework of recommendation system based on trust networks on this basis.· A hybrid recommendation method combining trust value with similarity. To solve the data sparsity problem, we proposed a hybrid recommendation method combining trust value with similarity. Through the introduction of user similarity as a supplement to trust, considering the social networks of trust and user similarity. It is more accurate to find the nearest neighbors of the target user collection, thereby improving the accuracy of the nearest neighbor search to improve the effect of the recommendation system.·A recommendation method based on expert users. To solve the new user problem, we proposed s recommendation method based on expert users. We discover expert users in the trust network, and then use these expert users to extend the new user’s trust network, so we can provide useful information to new users. This method solved the new user problem to a certain extent.· Design and Implementation of the recommendation system on the android mobile platform. Based on the recommendation system model, we designed and implemented an Android platform for mobile software recommendation system, the system can effectively recommend personalized phone software to the user. It preliminarily verified the feasibility of our recommendation system model.