Towards Efficient Collaborative Filtering Using Parallel Graph Model and Improved Similarity Measure

The recommender system can partially alleviate user's difficulty on information filtering and discover valuable information for the active user. Collaborative filtering has been widely used for recommendation. Because of facing the large-scale and sparse user-item rating matrix, high precision and better performance are always the big challenges for recommender system. In this paper, we model the rating data as a graph. We also make an improvement for similarity measure. Through the message propagation in the graph, candidate similar users will be selected firstly before the calculation of users similarity. Based on the topology of the user-item bipartite graph, the candidate items can be quickly located through the shortest path algorithm. Building on the parallel graph framework, the proposed approach is efficient and scalable. The experiments on real world cluster show that compared with the traditional collaborative filtering, our approach can better improve recommendation precision and the rating accuracy. It also has good scalability and real-time performance.