Meng Kang, Fan Li, Ping Kuang · ACM International Conference on Information and Knowledge Management (CIKM) 2026 · 2026
DOI: 10.1145/3799682.3841083
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).
Diffusion recommender models have shown promising results by reconstructing user preference signals through denoising. Since user-item interactions typically follow a long-tailed distribution, the denoising process tends to favor signals from popular items with high-frequency interactions. This makes recommendation results overly concentrated on popular items and limits the exposure of long-tail items. Directly suppressing popular items may increase tail exposure, but it can weaken useful collaborative signals and degrade overall accuracy. To address this limitation, we propose Popularity-Aware Complementary Preference Distillation (PCPD), a diffusion-based recommendation method that mitigates popularity bias in the denoising process. Instead of simply suppressing popular items, PCPD incorporates mainstream collaborative signals and sparse long-tail preference signals as complementary supervision. Specifically, PCPD preserves mainstream collaborative preferences, calibrates the recovery of tail-item preferences, and distills the resulting complementary supervision into a unified diffusion recommender. At inference time, PCPD relies only on the distilled student model, without requiring auxiliary branches or online score fusion. Extensive experiments on three real-world datasets show that PCPD consistently improves long-tail recommendation performance and alleviates item-side exposure imbalance while maintaining competitive overall recommendation accuracy.
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
ACM International Conference on Information and Knowledge Management (CIKM) 2026 · 0 citations
No comments yet — start the discussion below.