Top-N selection based on user behavior history formobile applications recommendation

Being one of the most essential techniques in recommender systems, current researches on collaborative filtering (CF) focus on how to improve the accuracy. However, it is of the same significance to recommend more potential interests to users because of their expectations for recommendation list besides the accuracy. Since current recommender systems rarely address this problem, this paper focuses on how to help users find more interests in the recommendation list. A sampling-based algorithm, namely, probabilistic top-N selection (PTN), is proposed to recommend potential interests for users. The PTN algorithm computes the proportion of each category of mobile applications in accordance with the users' historical downloads first. Then a candidate recommendation list, which is longer than needed, is generated with CF algorithm. Last, the mobile applications in the candidate list are further adjusted according to a certain probability related with the proportion and a final recommendation list is produced. A series of experiments are conducted against a mobile application dataset obtained from an application market in China. The experimental results demonstrate that our algorithm is effective for the recommendation.