Research On Personalized Recommendation System Based On Social Trust Network

With the rapid pace of information technology and computer science developments, popularity of Internet is strongly growing and consequently, the amount of information on the web is also strongly expanding. Users can get more and more free information so that their requirements for information can be met. But the ability of people for finding satisfied information in an limited time is declining because of the huge amount information. Apperantly, the traditional artificial filtering search algorithm is unable to meet the current requirments. This situation leads to the emerging of personalized recommendation system. Under a growing number of scholars’ participation and frontier research work, theoretical foundation of recommendation system is at a rapid development.Recently, as the emerging of personalized recommendation system, the above problem called information overload has been a partial remission. Personalized recommendation system predicts user interest preferences based on historical user behavior and based on their preferences to perform personalized analysis. Then automatically matching the interests of users with the respective items. Personalized recommendation system has now been widely used for e-commerce, such as music website recommendation, video website recommendation, commodity recommendation and so on.Recommendation algorithm is crucial to the recommendation system, the most widely used and most classical algorthm is collaborative filtering. Collaborative filtering algorithm includes user-based collaborative filtering approach and item-based collaborative filtering approach. Collaborative filtering is facing some challenging problems such as rating matrix sparsity, users and items cold start problem. Aslo the recommendation quality and accuracy of collaborative filtering alogrithm is limited.Today, people often tend to choose an item recommended by their friends in a highly developed social network. Under the situation of limited user rating information or items newly added to the recommendation system however, recommendation with respecte to valuable information deduced from soci al network can overcome some critical challenges experienced by collaborative filtering.Considering the shortcomings of above algorithm and the positive influences that can possibly obtianed from soical network analysis, a combination algorithm of collaborative filtering algorithm and soical network analysis is proposed in this thesis. First, traditional collaborative filtering algorithm is used to predict the rating of target items. Then a social network trust matrix is constructed based on tag information and user-to-friends information. Then a random walk algorithm is used to predict the rating of target item based on the trust analysis matrix. Finally, the two predicted rating is combined as weighted accumulation. Experiment results show that the proposed algorithm can improve the accuracy of recommendation system and bring better matched items for user.