Research Of Recommendation Algorithm Based On Social Network

As the advent of the information age and the popularity of smart phones, the volume of information has been rapidly growing within every aspect of our lives, people hope to find their desiring content and target from this giant information network. To meet this need, personalized recommendation technology has thus been widely used in various fields, such as search engines, e-commerce, social networking, news portals, etc.Social networks provide a platform, where the personalized recommendation technology can continue to mature gradually. With the help of recommendation engine, many personalized needs can be thus satisfied by applying various algorithms to the daily and history data from online user visiting and rating activity. However, several problems have still remained to be tackled in this area of technology, for example, data sparsity problem, cold-start problem, and black swan problem. A main problem is due to the raw data avaliable to the recommender system, which is characterized by limited activeness of large number of users and items, thus the density of dataset is relatively small and the sparsity is high. To solve this problem, many researches have taken other explicit and implicit information into consideration, namely, user interaction information or item interaction information, such as social trust network, similarity network, influence network, etc. Thus, the adoption of various networks is the trend of research of recommendation algorithm.In this paper, the author at first review the development of recommendation technology, where the collaborative filtering recommendation algorithm and model-based algorithm have been highlighted including existant probabilistic matrix factorization model and the extension based on social network, then introduce several classical types of users in social network.As for the rating prediction problems, the author have made use of the rating matrix, further concern the implicit influence network, then propose a new matrix factorization model-based recommendation algorithm, the last part is consisted of the experiment and result analysis.