A New Recommender Algorithm Based on Signed Networks and Attitude Information Consistency on Online Rating Systems

In order to obtain the attitude information of users and sell items, many online platforms have established rating systems. the online rating systems generally contain information about users, objects and users' attitude towards objects. This user-object relationship can be represented by the signed bipartite network. In this paper, we use positive and negative edges to mean attitude that users like or dislike objects on signed bipartite networks. Therefore, we propose a new recommender algorithm, which uses attitude information consistency among users to get the personalized recommendation lists, and eliminate potential negative edges in recommendation lists through the negative edge prediction score. The results on several real datasets show that our algorithm has higher recommendation accuracy compared with GRM (global ranking method), CF (collaborative filtering) and Index P algorithms, and our algorithm can prohibit negative edges from appearing in the recommendation lists as much as possible. In addition, our algorithm performs better than classical algorithms in terms of Novelty, which indicates that it ensures a certain diversity when recommending items that users may like. Thus, our method can recommend more personalized objects accurately.