Exploring Periodicity and Interactivity in Multi-Interest Framework for Sequential Recommendation

• DOI: 10.24963/ijcai.2021/197
• Publication Date: 2021-08-11T11:00:49Z
• Authors: Gaode Chen; Xinghua Zhang; Yanyan Zhao; Cong Xue; Ji Xiang
• Keywords: Interactivity

Sequential recommendation systems alleviate the problem of information overload, and have attracted increasing attention in literature. Most prior works usually obtain an overall representation based on user's behavior sequence, which can not sufficiently reflect multiple interests user. To this end, we propose a novel method called PIMI to mitigate issue. model multi-interest effectively by considering both periodicity interactivity item sequence. Specifically, design periodicity-aware module utilize time interval between behaviors. Meanwhile, ingenious graph is proposed enhance items capture global local features. Finally, extraction applied describe obtained representation. Extensive experiments two real-world datasets Amazon Taobao show that outperforms state-of-the-art methods consistently.