Sequential Recommendation Based on Long-Term and Short-Term User Behavior with Self-attention.

Product recommenders based on users’ interests are becoming increasingly essential in e-commerce. With the continuous development of the recommendation system, the available information is further enriched. In the case, user’s click or purchase behavior could be a visual representation of his or her interest. Due to the rapid update of products, users’ interests are not static, but change over time. In order to cope with the users’ interest changes, we propose a desirable work on the basis of representative recommendation algorithm. The sequence of user interaction behavior is thoroughly utilized, and the items that users interact at different times have different significance for the reflect of users’ interests. By considering the user’s sequential behaviors, this paper focuses on the recent ones to obtains the real interest of user. In this process, user behavior is divided into long-term and short-term, modeled by LSTM and Attention-based model respectively for user’s next click recommendation. We refer this model as LANCR and analyze the model in experiment. The experiment demonstrates that the proposed model has superior improvement compared with standard approaches. We deploy our model on two real datasets to verify the superior performance made in predicting user preferences.