Shiyu Song, Xiaofei Ding, Jie Gao, Yile Liang, Dongjing Wang, Xudong Shen, Xin Zhang, Tangjie Lv, Runze Wu · ACM International Conference on Information and Knowledge Management (CIKM) 2026 · 2026
DOI: 10.1145/3799682.3840122
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).
Generative recommendation has emerged as a transformative approach in industrial-scale systems, yet faces critical challenges in production environments, including Semantic Drift/Gap and cold start, and preference inconsistency. As NetEase's game platform serving more than 6 Million users, we identified that hash ID-based generative models suffer 84.6% performance degradation on the 60% long-tail items that generate minimal revenue but are crucial for user retention. Although multimodal content features could enrich item representations via semantic IDs, extracting and storing them can be tens of times higher than compact semantic IDs. Moreover, existing semantic IDs can be inconsistent with collaborative signals in industrial scenarios. To address this limitation, we propose LASER, a Label-Anchored Semantic Encoding with Multimodal-Collaborative Signals Harmonization for Industrial-Scale Generative Recommendation. LASER introduces two novel components: (1) Multimodal Semantic ID Learning, which constructs semantic IDs from multimodal features and aligns them with collaborative signals, via label-driven residual quantization and Swing-weighted pairwise alignment; and (2) Preference-aligned generative sequential recommendation, which leverages hierarchical semantic IDs to improve robustness under sparse interactions and cold-start, via popularity-aware fusion and preference-aligned autoregressive training. We conduct extensive experiments on real-world datasets and industrial scenarios to demonstrate the effectiveness of our model.
ACM International Conference on Information and Knowledge Management (CIKM) 2026 · 2 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
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.