Abstract: Accurately capturing user preferences over time can improve the precision in recommender systems．Simple correlation over time is typically not meaningful,since users change their preferences due to different external events．User behavior can often be determined by individual’s long-term and short-term preferences．In order to capture it,we propose the Session-based Temporal Graph(STG)．Based on the STG model framework,we propose the Path Fusion Algorithm(PFA) to make accurate top-N recommendation.And we evaluate the effectiveness of our method using two real datasets,CiteULike and Delicious．Experimental results show that the proposed algorithm has higher accuracy than most of previous algorithms.

Key words: Recommendation algorithm,Session,Temporal graphs,Preference fusion