Jiwon Son, Bohye Han, Sang‐Wook Kim · ACM International Conference on Information and Knowledge Management (CIKM) 2026 · 2026
DOI: 10.1145/3799682.3839971
Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).
Review-based recommendation systems (RBRS) leverage textual reviews to capture rich information about user preferences. A target review, which is directly associated with the target rating, contains highly relevant information for rating prediction but is unavailable during inference. Existing RBRS approaches try to reproduce the target review's semantic embedding by leveraging historical reviews. However, this semantic embedding often includes rating-irrelevant information, limiting its effectiveness in rating prediction. To address this issue, we propose TEDI, a novel RBRS based on privileged feature distillation (PFD). TEDI consists of a teacher module that distills rating-relevant knowledge from both target and historical reviews, and a student module that reproduces this knowledge by leveraging historical reviews during inference. Experiments on six datasets demonstrate that TEDI outperforms nine state-of-the-art RBRS methods, validating the effectiveness of utilizing knowledge from target reviews. Our code and data are available at https://github.com/Bigdasgit/TEDI.
ACM International Conference on Information and Knowledge Management (CIKM) 2026 · 0 citations
ACM International Conference on Information and Knowledge Management (CIKM) 2026 · 0 citations
ACM International Conference on Information and Knowledge Management (CIKM) 2026 · 0 citations
ACM International Conference on Information and Knowledge Management (CIKM) 2026 · 0 citations
No comments yet — start the discussion below.