Research Of Social Recommendation Algorithms Based On Trust Relations

Recommendation System is one of the key technologies to solve the information overload problem. By mining the user's behavior information, an effective recommendation algorithm is proposed to recommend the content which may be of interest to the user. In recent years, with the popularity of Facebook, Weibo and other Social Networking Platforms, Social Recommender System has gradually become an important research hotspot in the field of Recommender System. Compared with the traditional recommender algorithm, the social recommender algorithm can alleviate the cold start problem to some extent, improve the recommendation accuracy. However, since the user's online activities and interactive behaviors change with time, the data information continues to expand, which undoubtedly increases the cost of computing the social recommender algorithm. So, the social recommender algorithm's scalability problem needs to be solved urgently.This dissertation explores how to solve the problem of cold start,data sparseness and scalability in the recommendation process. It deeply studies the traditional collaborative filtering recommendation algorithm and the corresponding optimization algorithms, including the social collaborative filtering and the incremental collaborative filtering.Following aspects are main contributions of this dissertation:1) In order to solve the cold start and data sparseness problems of the traditional recommendation algorithm, a new social recommendation algorithm based on the trust relations named SocialSVD++ is proposed.It incorporates the trust relations into the collaborative filtering algorithm.First, trust value of social datasets are always binary number in [0,1],which cannot reflect the degree of trust between users, so the Degree Centrality is used to digitize the original trust data. Then we bulid SVD++ model based on the effects of explicit and implicit of rating data and the effects of new trust data.2) In order to solve the poor realtime ability problem of the SocialSVD++ algorithm, an incremental updating algorithm named Incremental SocialSVCD++ is proposed to be used for the dynamic recommendation of ratings changing. First, The SocialSVD++ model was trained with original rating matrix and original digital trust matrix to get the static prediction ratings. Then, each time when the number of new ratings reaches the update threshold, incremental rating matrix and digital trust matrix were dynamically sampled from the current total matrix according to the users, items of the new ratings and SocialSVD++model was trained again with the two matrices obtained by sampling to get the incremental prediction ratings. Finally, the prediction ratings of static training and incremental training were merged with the linear weighting to get the final predicted ratings.Experimental results show that the SocialSVD++ algorithm is superior to the other social collaborative filtering algorithms on two social dataset, and alleviates the users' cold start and data sparseness problem to a certain extent. Compared with the Social SVD++ algorithm,the Incremental Social SVD++ reduces the computation cost in the premise of ensuring high recommended accuracy.