Research On Recommendation Algorithm Based On Users' Similarity And Social Trust Relationship

With continually expanding of the network size,colorful and abundant information has been presented exponential growth.The large amount of redundant information is very serious problem for customer and researchers.Currently,recommender system can be the best way to solve this problem.The main idea is to form a certain regular model and predict the behavior of users in the future through the analysis of their historical data.So recommender system's research and development has important significance for academia or industry.Traditional collaborative filtering algorithm only depends on the user-item rating matrix,due to the inherent defect of the score matrix that makes the final prediction results has been limited.But in real life,users are often inclined to recommend by their trust people,so if we can find a group of the user's trust people,which can improve recommender systems and provide more personalized recommendation results.But we can directly obtain the trust relationship is very few between users,there is a difficult problem for analysis.In fact,under the background of the same social networking platform,because of most of users do not know each other,which make them have distrust relationships.If two users have similar factors such as interest,age and work environment,we can affirm that has a potential trust relationship between two people in the same context,it is also called implicit trust relationship.Therefore,if we can find the implicit trust relationship between users,it may bring a great help for recommendation.In order to overcome above problems,we propose one social recommendation method that utilize user's classic Pearson similarity and heterogeneous network similarity to form trust hierarchy to improve the prediction accuracy based on the popular recommendation algorithm model,it named IMtrustSVD.The experiment shows that our method not only has much better than other state-of-the-art recommendation algorithm,but also effectively solve the data sparseness and user cold start problem on data set MovieLens100 K.