SSM-based Sequential Recommendation Considering Short-term Dependencies of Time and Adjacency

Keito Kozaki, Keigo Sakurai, Ren Togo, Takahiro Ogawa, Miki Haseyama

Abstract: Sequential recommendation focuses on capturing user interests based on the chronological order of user’s interaction history. Recently, State-Space Model (SSM), a state-of-the-art time-series machine learning model, has attracted attention in sequential recommendation modeling. SSM-based sequential recommendation models can effectively capture long-term user preferences, but modeling short-term dependencies is difficult. In this paper, we propose a new sequential recommendation model, which complements the short-term dependencies of SSM-based models in two aspects: item adjacencies and time intervals. To capture the adjacency between items, a graph structure is constructed from the sequences, and the time interval between items is captured in the time-aware SSM layer. Extensive experiments on two real-world datasets show that our model consistently outperforms state-of-the-art recommendation models.