Collaborative Filtering Through Combining Bipartite Graph Projection and Ranking
Enhong Chen · Journal of Chinese Computer Systems · 2010
Collaborative Filtering is the most widely used approach in recommender systems. This paper proposes a novel collaborative filtering approach through combining bipartite graph projection and ranking. In our approach bipartite graph is projected based on structural similarity and random walk is used to rank the nodes. This method can not only deal with cold start problem to get high precision but also have good scalability. Moreover,our method can generate the rank based on the similarity between all user-item pairs,thus making it avoids inaccuracy caused by low coverage rate. The algorithm is tested in both the item-collaborative-based item ranking way and the user-collaborative-based user ranking way. The experimental results obtained on a benchmark dataset Movielens clearly show the effectiveness of our proposed approach.