Interest-Forgetting Markov Model for Next-Basket Recommendation

Recommendation systems provide users with ranked items based on individual’s preferences. Two types of preferences are commonly used to generate ranking lists: long-term preferences which are relatively stable and short-term preferences which are constantly changeable. But short-term preferences have an important real-time impact on individual’s current preferences. In order to predict personalized sequential patterns, the long-term user preferences and the short-term variations in preference need to be jointly considered for both personalization and sequential transitions. In this paper, a IFNR model is proposed to leverage long-term and short-term preferences for Next-Basket recommendation. In IFNR, similarity was used to represent long-term preferences. Personalized Markov model was exploited to mine short-term preferences based on individual’s behavior sequences. Personalized Markov transition matrix is generally very sparse, and thus it integrated Interest-Forgetting attribute, social trust relation and item similarity into personalized Markov model. Experimental results are on two real data sets, and show that this approach can improve the quality of recommendations compared with the existed methods.