김현수 석사과정, 김준영 석사과정, 이선경 박사과정, 최민진 석박통합과정, 이종욱 교수과정 CIKM 2024 국제 학술대회 논문 채택

23 Jul 2024

DIAL 연구실 소속 인공지능학과 김현수(석사과정, 공동 1저자) 학생, 김준영(석사과정, 공동 1저자) 학생, 이선경(박사과정, 제3저자) 학생, 최민진(석·박통합과정, 4저자) 학생, 이종욱(교신저자) 교수가 참여한 “MARS: Matching Attribute-aware Representations for Text-based Sequential Recommendation” 논문이 데이터마이닝 분야 최우수 국제 학술대회인 ACM International Conference on Information and Knowledge Management (CIKM 2024, short paper)에 최종 게재가 승인되었으며 오는 10월에 발표될 예정입니다. 

Abstract

Sequential recommendation aims to infer the next item a user is likely to prefer based on the user’s sequential interaction. Recently, text-based sequential recommendation has been proposed to exploit textual item features with collaborative signals. They successfully transfer knowledge from training data to new datasets by encoding text-based items with pre-trained language models. Despite their advantages, text-based recommender models still face two challenges for fine-grained recommendation: (i) how to represent the user/item with multiple attributes and (ii) how to match the appropriate item considering complex user interests. To this end, we propose a novel model, namely Matching Attribute-aware Representations for Text-based Sequential Recommendation (MARS). First, we extract fine-grained user/item representations by attribute-aware text encoding. In particular, we fully utilize multiple attribute-aware representations to reflect diverse user intents. Then, a user-item score is computed via attribute-wise interaction matching to capture the user’s attribute-level preference. Extensive experiments show that MARS surpasses existing sequential models by up to 24.43% and 29.26% in Recall@10 and NDCG@10 on five benchmark datasets

Model Architecture