Xintian Han, Honggang Chen, Quan Lin, Jingyue Gao, Xiangyuan Ren, Lifei Zhu, Zhi‐Sheng Ye, Shikang Wu, Xionghang Xie, Xiaochu Gan, Bingzheng Wei, Peng Xu, Zhe Wang, Yu Zheng, J.-L. Lin, Di Wu, Junfeng Ge · ACM International Conference on Information and Knowledge Management (CIKM) 2026 · 2026
DOI: 10.1145/3799682.3840084
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).
Traditional ID-based recommender systems often struggle with cold-start and generalization challenges. Multimodal recommendation systems, which leverage textual and visual data, offer a promising solution to mitigate these issues. However, existing industrial approaches typically adopt a two-stage training paradigm: first pretraining a multimodal model, then applying its frozen representations to train the recommendation model. This decoupled framework suffers from misalignment between multimodal learning and recommendation objectives, as well as an inability to adapt dynamically to new data. To address these limitations, we propose LEMUR, the first large-scale multimodal recommender system trained end-to-end from raw data. By jointly optimizing both the multimodal and recommendation components, LEMUR ensures tighter alignment with downstream objectives while enabling real-time parameter updates. Constructing multimodal sequential representations from user history often entails prohibitively high computational costs. To alleviate this bottleneck, we propose a novel memory bank mechanism that incrementally accumulates historical multimodal representations throughout the training process. In a 14-day online A/B test on Douyin Search, LEMUR has led to a significant 0.843% reduction in query change rate decay, a 0.558% increase in staytime and a 0.81% improvement in QAUC. Additionally, LEMUR has shown significant gains across key online or offline metrics for Douyin Advertisement, Tiktok Search and Ecom Search. Our results validate the superiority of end-to-end multimodal recommendation in real-world industrial scenarios.
Last synced
This work has 12 recorded citations, but citing papers have not been linked locally yet.
No comments yet — start the discussion below.