Elamur Research⌕
☰
home / papers / 2026 / An Efficient Framework for Whole-Page Reranking via Single-Modal Supervision

An Efficient Framework for Whole-Page Reranking via Single-Modal Supervision

Paper · 2026 · Zishuai Zhang, Sihao Yu, Xiewenyi et al.

Summary

Search result pages mix items from several sources, such as web text, videos, images and LLM answer cards. Ordering them jointly (whole-page reranking) usually requires expensive human labels, because annotators must judge relevance across modalities. This industry paper from Beihang University and Baidu proposes SMAR, which uses existing single-modal rankers to reduce that labeling burden.

Upstream per-modality rankers do two jobs. First, they choose which items get human cross-modal labels:

  • Top-P labels only the highest-ranked fraction of each modality list.
  • Iso-label anchors use binary search to find cross-modal items with matching upstream scores, which helps align score scales between modalities.

Second, their intra-modality orderings act as cheap supervision through pairwise or ListMLE losses. These losses need only relative order, so binned or coarse upstream signals still work.

The reranker itself fuses visual and textual features with a gating mechanism and adds user personalization through cross-attention. On Qilin and CrossRank, SMAR needs 70–90% less annotation yet beats fully annotated baselines. Online A/B tests at Baidu also showed gains, making the approach a practical, budget-friendly option for industrial search.

Key points

  • Annotation savings: SMAR cuts labeling cost by 70–90% on Qilin and CrossRank while outperforming fully supervised fine-tuning. For example, in the paper's CrossRank table with Bert-base-chinese, MRR@1 is about 0.759 with SMAR on 30% of the data versus about 0.741 for SFT on the full data. Qwen3-Reranker-4B was also evaluated.
  • Motivating experiment: On Qilin, items that upstream rankers place near the top are more valuable to annotate than random, middle or tail items.
  • Comparison with other low-label methods: SMAR outperforms knowledge-distillation and pseudo-labeling baselines and achieves the best α-NDCG@10 (α = 0.5) diversity score.
  • Online A/B test against Baidu's production reranker:
    • Reported gains are roughly +0.86% NDCG, +0.25% CTR (top-8) and +1.58% ∆GSB on random queries.
    • Next-day retention rose by about 0.3%; the paper gives 0.33% in one place and 0.31% in another.
    • CTR@1 dipped slightly, which the authors attribute to LLM answer cards at the top satisfying users without a click.
  • Released resources: Code and a large industrial dataset (CrossRank) are available at github.com/zzs97str/SMAR (arXiv 2510.16803).

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Institutions

🏢BaiduBeijing · China84 works · 17,681 citations🎓Beihang UniversityBeijing · China86 works · 9,097 citations

 Fields, topics and concepts

📚 Shelves 5The field of science this work belongs to (OpenAlex fields).Computer Science · 18,614 worksArtificial Intelligence · 17,965 worksArts and Humanities · 17,450 worksLanguage and Linguistics · 17,296 worksInformation Systems · 1,730 works

🧭 Topics 1Research subjects and languages the work is about.Information retrieval & search · 552 works

Cite

Cite as Zishuai Zhang, Sihao Yu, Xiewenyi, Ying Nie, Junfeng Wang, Zhiming Zheng, Dawei Yin and Hainan Zhang. 2026. An Efficient Framework for Whole-Page Reranking via Single-Modal Supervision. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 6: Industry Track).

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.acl-industry.67

@inproceedings{zhang2026efficient,
  title = {{An Efficient Framework for Whole-Page Reranking via Single-Modal Supervision}},
  author = {Zishuai Zhang, Sihao Yu, Xiewenyi, Ying Nie, Junfeng Wang, Zhiming Zheng, Dawei Yin and Hainan Zhang},
  booktitle = {Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 6: Industry Track)},
  year = {2026},
  doi = {10.18653/v1/2026.acl-industry.67},
  url = {https://doi.org/10.18653/v1/2026.acl-industry.67}
}

References

19 references, 5 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Gavin Brown, Adam Pocock, Ming-Jie Zhao, and Mikel Luján. 2012. Conditional likelihood maximisation: a unifying framework for information theoretic feature selection. The journal of machine learning research, 13(1):27–66. find
  2. Christopher Burges, Robert Ragno, and Quoc Le. 2006. Learning to rank with nonsmooth cost functions. Advances in neural information processing systems, 19. find
  3. Jia Chen, Qian Dong, Haitao Li, Xiaohui He, Yan Gao, Shaosheng Cao, Yi Wu, Ping Yang, Chen Xu, Yao Hu, and 1 others. 2025. Qilin: A multimodal information retrieval dataset with app-level user sessions. In Proceedings of the 48th International ACM SI-GIR Conference on Research and Development in Inf… find
  4. Aleksandr Chuklin, Ilya Markov, and Maarten De Rijke. 2022. Click models for web search. Springer Nature. find
  5. Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. Bert: Pre-training of deep bidirectional transformers for language understanding. In Proceedings of the 2019 conference of the North American chapter of the association for computational linguistics: human language technologies,…
  6. Huifeng Guo, Ruiming Tang, Yunming Ye, Zhenguo Li, and Xiuqiang He. 2017. Deepfm: a factorization-machine based neural network for ctr prediction. arXiv preprint arXiv:1703.04247. find
  7. Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. 2016. Deep residual learning for image recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 770– 778.
  8. Edward J Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, Weizhu Chen, and 1 others. 2022. Lora: Low-rank adaptation of large language models. ICLR, 1(2):3.
  9. Canjia Li, Xiaoyang Wang, Dongdong Li, Yiding Liu, Yu Lu, Shuaiqiang Wang, Zhicong Cheng, Simiu Gu, and Dawei Yin. 2023. Pretrained language model based web search ranking: From relevance to satisfaction. arXiv preprint arXiv:2306.01599. — [arXiv:2306.01599](https://arxiv.org/abs/2306.01599) find
  10. Yuchen Li, Hao Zhang, Haojie Zhang, Hengyi Cai, Xinyu Ma, Shuaiqiang Wang, Haoyi Xiong, Zhaochun Ren, Maarten de Rijke, and Dawei Yin. 2025. Fultr: A large-scale fusion learning to rank dataset and its application for satisfaction-oriented ranking. In Proceedings of the 31st ACM SIGKDD Conference on… find
  11. Haitao Mao, Lixin Zou, Yujia Zheng, Jiliang Tang, Xiaokai Chu, Jiashu Zhao, Qian Wang, and Dawei Yin. 2024. Whole page unbiased learning to rank. In Proceedings of the ACM Web Conference 2024, pages 1431–1440. find
  12. Junwei Pan, Wei Xue, Ximei Wang, Haibin Yu, Xun Liu, Shijie Quan, Xueming Qiu, Dapeng Liu, Lei Xiao, and Jie Jiang. 2024. Ads recommendation in a collapsed and entangled world. In Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, pages 5566–5577. find
  13. Przemysław Pobrotyn, Tomasz Bartczak, Mikołaj Syn-owiec, Radosław Białobrzeski, and Jarosław Bojar. 2020. Context-aware learning to rank with self-attention. arXiv preprint arXiv:2005.10084. — [arXiv:2005.10084](https://arxiv.org/abs/2005.10084) find
  14. Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, and 1 others. 2021. Learning transferable visual models from natural language supervision. In International conference on machine learning, pages 8748–8…
  15. Qi Wan, Xiangnan He, Xiang Wang, Jiancan Wu, Wei Guo, and Ruiming Tang. 2022. Cross pairwise ranking for unbiased item recommendation. In Proceedings of the ACM web conference 2022, pages 2370– 2378. find
  16. Peng Wang, Shuai Bai, Sinan Tan, Shijie Wang, Zhihao Fan, Jinze Bai, Keqin Chen, Xuejing Liu, Jialin Wang, Wenbin Ge, Yang Fan, Kai Dang, Mengfei Du, Xuancheng Ren, Rui Men, Dayiheng Liu, Chang Zhou, Jingren Zhou, and Junyang Lin. 2024. Qwen2-vl: Enhancing vision-language model’s perception of the w…
  17. Yue Wang, Dawei Yin, Luo Jie, Pengyuan Wang, Makoto Yamada, Yi Chang, and Qiaozhu Mei. 2016. Beyond ranking: Optimizing whole-page presentation. In Proceedings of the Ninth ACM International Conference on Web Search and Data Mining, pages 103–112. find
  18. Fen Xia, Tie-Yan Liu, Jue Wang, Wensheng Zhang, and Hang Li. 2008. Listwise approach to learning to rank: theory and algorithm. In Proceedings of the 25th international conference on Machine learning, pages 1192–1199. find
  19. Algorithm 1 Binary search for iso-label anchors find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/zhang-2026-efficient-framework-whole-page-reranking/