이선경 박사과정, 최민진 박사 졸업생, 최은성 석박통합과정, 김혜영 박사과정 ACL 2025 국제 학술대회 논문 채택

16 May 2025

DIAL 연구실 소속 인공지능학과 이선경(박사과정, 제1저자) 학생, 최민진(박사 졸업생, 제2저자), 최은성(석박통합과정, 제3저자) 학생, 김혜영(박사과정, 제4저자) 학생, 이종욱(교신저자) 교수가 참여한 논문 "GRAM: Generative Recommendation via Semantic-aware Multi-granular Late Fusion"이 자연어처리 분야 최우수 국제 학술대회인 The 63rd Annual Meeting of the Association for Computational Linguistics (ACL 2025) Main conference에 최종 게재가 승인되었으며 오는 7월에 발표될 예정입니다.

Abstract

Generative recommendation is an emerging paradigm that leverages extensive knowledge of large language models (LLMs) by formulating recommendations into a text-to-text generation task. However, existing studies face two key limitations in (i) incorporating implicit item relationships and (ii) utilizing rich yet lengthy item information. To address these challenges, we propose a Generative Recommender via semantic-Aware Multi-granular late fusion (GRAM), introducing two synergistic innovations. First, we design semantic-to-lexical translation to encode implicit hierarchical and collaborative item relationships into the vocabulary space of LLMs. Second, we present multi-granular late fusion to integrate rich semantics efficiently with minimal information loss. It employs separate encoders for multi-granular prompts, delaying the fusion until the decoding stage. Experiments on four benchmark datasets show that GRAM outperforms seven state-of-the-art generative recommenders, achieving significant improvements of 11.5-16.0% in Recall@5 and 5.3-13.6% in NDCG@5. The code will be available upon acceptance.