Zhan, H., Li, L., Geng, X., Lin, J., & Kot, A. C. (2022). Rule-guided knowledge-graph based negative sampling for outfit recommendation. In Proceedings of the 45th International ACM SIGIR Conference on Research and Development in Information Retrieval workshop on E-commerce (SIGIR Ecom'22). ACM, New York, NY, USA, 5 pages.
Abstract:
Recommender system (RS) has become increasingly prevalent among online service providers. The effectiveness of an accurate RS highly relies on the quality level of the selected negative instances for training. Most of the existing negative sampling strategies either merely leverage the interaction data which suffer from the sparsity challenge or fail to fully utilize the side information. To address these issues, we introduce rule-guided knowledge graph (RuleKG) by integrating the enriched relations of the knowledge graph (KG) and category-aware fashion rules into the outfit recommendation problem. The rules are further incorporated into a newly designed score function which represents the user’s preferences toward outfit in a more fine-grained perspective. Given a user-outfit pair, the negative candidates are explored in both outfit- and user-level. Also, the reinforcement learning (RL)-based strategy is developed to automatically choose the next state from the starting point over KG. Experimental results on the new top-bottom outfit dataset demonstrate the superiority of the proposed approach and the generality of the negative sampler model is validated on the music recommendation benchmark.
License type:
Publisher Copyright
Funding Info:
This research is supported by core funding from: A*STAR Institute for Infocomm Research (A*STAR I²R)
Grant Reference no. : N.A
Description:
Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit. Copyrights for components of this work owned by others than ACM must be honored. Abstracting with credit is permitted.