Cross-Intent Graph Contrastive Learning for Fashion Sequential Recommendation

Abstract

Sequential recommendation in e-commerce seeks to recommend items to users based on their past purchase preferences and has received much research attention. However, most approaches are so far restricted to modeling the user's long-term intent (user-to-item) and short-term intent (item-to-item) separately, which may extract the user's purchase interest but disregard possible synergistic interplay between the long-term and short-term intents on the user's overall purchase interest. To balance the users' long-term and short-term intents, this paper develops a cross-intent graph contrastive learning framework. By encouraging alignment of graph network representations among multiple intentions, we capture complementary information from different intentions of the same user. We extensively evaluate this framework by conducting experiments on two publicly available fashion datasets. The results demonstrate the effectiveness and efficiency of cross-intent graph contrastive learning.