Collaborative Filtering Recommendation Model Considering Integration of User Rating and Attribute Similarity

Abstract

IntroductionWith the development of e-commence, service providers must clearly grasp users' needs and preferences and provide users with services to their satisfaction in the premise that product quality is guaranteed. Therefore, excellent recommendation systems are very significant in e-commence application and also become one of the researchers' focuses of attention [1,2]. Basic thought of collaborative filtering algorithm is to judge whether the evaluation is of value to target user through evaluation of other users having similar interest preference with target user on target item and then to decide whether to recommend target item to target user further. Nearest neighboring collaborative filtering recommendation is the most successful recommendation technology at present and the algorithm recommends items to target user based on rating data of nearest neighbor having similar rating. As nearest neighbor's rating on item is similar to that of target user, target user's rating on unrated item can be predicted through nearest neighbor's rating on the item with weighting method. As collaborative filtering algorithm is usually influenced by data sparsity and cold start, a collaborative filtering recommendation algorithm based on improvement nearest neighbor is proposed. Firstly, the current user's k nearest neighbor and reverse k nearest neighbor are obtained on the basis of the similarity algorithm, which are used to compute positive and negative credibility values respectively based on their predicted ratings and the current user ratings. Then modifications of constraint are made for the users who are both the k nearest neighbor and the reverse k nearest neighbor and the hot resources. Finally, the collaborative filtering recommendation algorithm based on weight fusion is derived and a comparative experiment of simulation is conducted on MovieLens. The result shows that the algorithm in the Thesis decreases the mean absolute error value while improving the accuracy of recommendation [3][4][5][6][7][8].

Adopting explicit and implicit social relations by SVD++ for recommendation system improvement

“…A recommendation system (RS) can be defined as a tool that can recommend a list of items to interested users [3]. A recommendation system comes in a variety of types, collaborative filtering being the most common and most widely used [4,5]. Collaborative filtering estimates unknown ratings based on user history ratings and behaviors of similar users.…”