Elamur Research⌕
☰
home / papers / 2026 / A Novel Matching Paradigm: Unified Generative and Discriminative LLM with Prompt Compression for Relevance Learning

A Novel Matching Paradigm: Unified Generative and Discriminative LLM with Prompt Compression for Relevance Learning

Paper · 2026 · Guoliang Zhao, Zixin Cui, Chao Ye et al.

Summary

The paper addresses query–landing-page relevance in industrial search advertising. In this setting, LLMs usually act only as feature extractors, and two-tower or single-tower matchers return bare scores with no explanation. The authors, from Baidu Search Ads and Xi'an Jiaotong University, propose UGD (Unified Generative and Discriminative LLM). UGD places two-tower scoring, single-tower scoring and generation of a textual relevance reason inside one LLM by partitioning the attention mask across special-token segments of a CoT-style prompt. Multi-task fine-tuning lets reason generation act as extra supervision, and KL losses transfer knowledge from the single-tower head to the two-tower head.

Long landing pages make serving slow, so a self-distilled variant shrinks each page into a few learned vectors. A KMeans-initialised linearized RQ-VAE then quantises those vectors into compact codebook indices. The authors also release the QLQ dataset. In this setup, page input drops from 600 to 40 tokens with only a small accuracy loss. The system is deployed in Baidu's ad engine, where advertisers receive both scores and explanations.

Key points

  • Dataset: The authors introduce QLQ (Query Landing Page Quality), built from an industrial search-ads engine. For the compression experiments, they add 213,367 snapshot samples and a Conditional Fine-Tuning strategy, which raised UGD STM to 0.7855 ACC / 0.9445 AUC (+3.3% ACC over the non-snapshot setting).
  • Self-distillation: Mask isolation keeps the full ⟨Query, LP⟩ view and the compressed view separate during training. This adds +2.11% ACC / +0.95% AUC over compression with placeholder tokens alone.
  • Quantization: Plain VQVAE drops ACC to 0.6234. Moving through RQVAE, then linearized RQVAE, then the KMeans warm start reaches 0.7798 / 0.9389. That is about 0.57 points below the uncompressed baseline, with only 8 × 40 codebook indices stored per page.
  • Online A/B test: Over at least one week, the share of 0-score landing pages fell by 1.87% with UGD TTM and by 3.2% with UGD STM, compared with the previously deployed model.
  • Limitations: Production systems lack a systematic evaluation of the generated reasons. Quantization costs more accuracy than compression does.

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🎓Xi'an Jiaotong UniversityXi'an · China42 works · 36,431 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 2Research subjects and languages the work is about.Language models & pretraining · 3,894 worksEfficient models & compression · 297 works

Cite

Cite as Guoliang Zhao, Zixin Cui, Chao Ye, Dengwu He, Fei Huang, Yubo Liu, Shuanglong Li, Tzungren Kuo, Bin Ding, Shuang Zhang, KunhongZhu, Zhi Guo and Liu Lin. 2026. A Novel Matching Paradigm: Unified Generative and Discriminative LLM with Prompt Compression for Relevance Learning. 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.122

@inproceedings{zhao2026novel,
  title = {{A Novel Matching Paradigm: Unified Generative and Discriminative LLM with Prompt Compression for Relevance Learning}},
  author = {Guoliang Zhao, Zixin Cui, Chao Ye, Dengwu He, Fei Huang, Yubo Liu, Shuanglong Li, Tzungren Kuo, Bin Ding, Shuang Zhang, KunhongZhu, Zhi Guo and Liu Lin},
  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.122},
  url = {https://doi.org/10.18653/v1/2026.acl-industry.122}
}

References

40 references, 15 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Gediminas Adomavicius and Alexander Tuzhilin. 2005. Toward the next generation of recommender systems: A survey of the state-of-the-art and possible extensions. IEEE transactions on knowledge and data engineering, 17(6):734–749. find
  2. Samuel R Bowman, Gabor Angeli, Christopher Potts, and Christopher D Manning. 2015. A large annotated corpus for learning natural language inference. arXiv preprint arXiv:1508.05326.
  3. Jing Chen, Qingcai Chen, Xin Liu, Haijun Yang, Daohe Lu, and Buzhou Tang. 2018. The bq corpus: A large-scale domain-specific chinese corpus for sentence semantic equivalence identification. In Proceedings of the 2018 conference on empirical methods in natural language processing, pages 4946–4951. find
  4. Lizhe Chen, Binjia Zhou, Yuyao Ge, Jiayi Chen, and Shiguang Ni. 2025. Pis: Linking importance sampling and attention mechanisms for efficient prompt compression. arXiv preprint arXiv:2504.16574. — [arXiv:2504.16574](https://arxiv.org/abs/2504.16574) find
  5. Paul Covington, Jay Adams, and Emre Sargin. 2016. Deep neural networks for youtube recommendations. In Proceedings of the 10th ACM conference on recommender systems, pages 191–198. find
  6. Jacob Devlin. 2018. Bert: Pre-training of deep bidirectional transformers for language understanding. arXiv preprint arXiv:1810.04805.
  7. 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,…
  8. Li Dong, Nan Yang, Wenhui Wang, Furu Wei, Xiaodong Liu, Yu Wang, Jianfeng Gao, Ming Zhou, and Hsiao-Wuen Hon. 2019. Unified language model pre-training for natural language understanding and generation. Advances in neural information processing systems, 32.
  9. Zhengxiao Du, Yujie Qian, Xiao Liu, Ming Ding, Jiezhong Qiu, Zhilin Yang, and Jie Tang. 2021. Glm: General language model pretraining with autoregressive blank infilling. arXiv preprint arXiv:2103.10360.
  10. Hector Garcia-Molina, Georgia Koutrika, and Aditya Parameswaran. 2011. Information seeking: convergence of search, recommendations, and advertising. Communications of the ACM, 54(11):121–130. find
  11. Jui-Ting Huang, Ashish Sharma, Shuying Sun, Li Xia, David Zhang, Philip Pronin, Janani Padmanabhan, Giuseppe Ottaviano, and Linjun Yang. 2020. Embedding-based retrieval in facebook search. In Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, pages 2553… find
  12. Po-Sen Huang, Xiaodong He, Jianfeng Gao, Li Deng, Alex Acero, and Larry Heck. 2013. Learning deep structured semantic models for web search using clickthrough data. In Proceedings of the 22nd ACM international conference on Information & Knowledge Management, pages 2333–2338. find
  13. Jiho Jang, Chaerin Kong, Donghyeon Jeon, Seonhoon Kim, and Nojun Kwak. 2023. Unifying vision-language representation space with single-tower transformer. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 37, pages 980–988. find
  14. Huiqiang Jiang, Qianhui Wu, Chin-Yew Lin, Yuqing Yang, and Lili Qiu. 2023. Llmlingua: Compressing prompts for accelerated inference of large language models. arXiv preprint arXiv:2310.05736.
  15. Omar Khattab and Matei Zaharia. 2020. Colbert: Efficient and effective passage search via contextualized late interaction over bert. In Proceedings of the 43rd International ACM SIGIR conference on research and development in Information Retrieval, pages 39– 48.
  16. Wonjae Kim, Bokyung Son, and Ildoo Kim. 2021. Vilt: Vision-and-language transformer without convolution or region supervision. In International conference on machine learning, pages 5583–5594. PMLR.
  17. Mei Kobayashi and Koichi Takeda. 2000. Information retrieval on the web. ACM computing surveys (CSUR), 32(2):144–173. find
  18. Solomon Kullback and Richard A Leibler. 1951. On information and sufficiency. The annals of mathematical statistics, 22(1):79–86. find
  19. Huong T Le, Dung T Cao, Trung H Bui, Long T Luong, and Huy Q Nguyen. 2021. Improve quora question pair dataset for question similarity task. In 2021 RIVF International Conference on Computing 1789 7 and Communication Technologies (RIVF), pages 1–5. IEEE. find
  20. Doyup Lee, Chiheon Kim, Saehoon Kim, Minsu Cho, and Wook-Shin Han. 2022. Autoregressive image generation using residual quantization. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 11523–11532. find
  21. Zongqian Li, Yinhong Liu, Yixuan Su, and Nigel Collier. 2025a. Prompt compression for large language models: A survey. In Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers), p…
  22. Zongqian Li, Yixuan Su, and Nigel Collier. 2025b. 500xcompressor: Generalized prompt compression for large language models. In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 25081–25091.
  23. Zhenghao Lin, Yeyun Gong, Xiao Liu, Hang Zhang, Chen Lin, Anlei Dong, Jian Jiao, Jingwen Lu, Daxin Jiang, Rangan Majumder, and 1 others. 2023. Prod: Progressive distillation for dense retrieval. In Proceedings of the ACM Web Conference 2023, pages 3299–3308. find
  24. Aixin Liu, Bei Feng, Bing Xue, Bingxuan Wang, Bochao Wu, Chengda Lu, Chenggang Zhao, Chengqi Deng, Chenyu Zhang, Chong Ruan, and 1 others. 2024. Deepseek-v3 technical report. arXiv preprint arXiv:2412.19437. find
  25. Xin Liu, Qingcai Chen, Chong Deng, Huajun Zeng, Jing Chen, Dongfang Li, and Buzhou Tang. 2018. Lcqmc: A large-scale chinese question matching corpus. In Proceedings of the 27th international conference on computational linguistics, pages 1952–1962.
  26. Yinhan Liu. 2019. Roberta: A robustly optimized bert pretraining approach. arXiv preprint arXiv:1907.11692, 364.
  27. Haoyu Lu, Nanyi Fei, Yuqi Huo, Yizhao Gao, Zhiwu Lu, and Ji-Rong Wen. 2022. Cots: Collaborative two-stream vision-language pre-training model for cross-modal retrieval. In Proceedings of the IEEE/CVF conference on computer Vision and pattern recognition, pages 15692–15701. find
  28. N Reimers. 2019. Sentence-bert: Sentence embeddings using siamese bert-networks. arXiv preprint arXiv:1908.10084.
  29. Ying Shan, T Ryan Hoens, Jian Jiao, Haijing Wang, Dong Yu, and JC Mao. 2016. Deep crossing: Web-scale modeling without manually crafted combinatorial features. In Proceedings of the 22nd ACM SIGKDD international conference on knowledge discovery and data mining, pages 255–262. find
  30. Liangcai Su, Fan Yan, Jieming Zhu, Xi Xiao, Haoyi Duan, Zhou Zhao, Zhenhua Dong, and Ruiming Tang. 2023. Beyond two-tower matching: Learning sparse retrievable cross-interactions for recommendation. In Proceedings of the 46th International ACM SIGIR Conference on Research and Development in Informat… find
  31. Kimi Team, Yifan Bai, Yiping Bao, Guanduo Chen, Jiahao Chen, Ningxin Chen, Ruijue Chen, Yanru Chen, Yuankun Chen, Yutian Chen, and 1 others. 2025. Kimi k2: Open agentic intelligence. arXiv preprint arXiv:2507.20534.
  32. Aaron Van Den Oord, Oriol Vinyals, and 1 others. 2017. Neural discrete representation learning. Advances in neural information processing systems, 30. find
  33. Zhe Wang, Liqin Zhao, Biye Jiang, Guorui Zhou, Xiaoqiang Zhu, and Kun Gai. 2020. Cold: Towards the next generation of pre-ranking system. arXiv preprint arXiv:2007.16122. — [arXiv:2007.16122](https://arxiv.org/abs/2007.16122) find
  34. Dongbo Xi, Zhen Chen, Peng Yan, Yinger Zhang, Yongchun Zhu, Fuzhen Zhuang, and Yu Chen. 2021. Modeling the sequential dependence among audience multi-step conversions with multi-task learning in targeted display advertising. In Proceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery & D… find
  35. An Yang, Anfeng Li, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Gao, Chengen Huang, Chenxu Lv, and 1 others. 2025. Qwen3 technical report. arXiv preprint arXiv:2505.09388.
  36. Ji Yang, Xinyang Yi, Derek Zhiyuan Cheng, Lichan Hong, Yang Li, Simon Xiaoming Wang, Taibai Xu, and Ed H Chi. 2020. Mixed negative sampling for learning two-tower neural networks in recommendations. In Companion proceedings of the web conference 2020, pages 441–447. find
  37. Xinyang Yi, Ji Yang, Lichan Hong, Derek Zhiyuan Cheng, Lukasz Heldt, Aditee Kumthekar, Zhe Zhao, Li Wei, and Ed Chi. 2019. Sampling-bias-corrected neural modeling for large corpus item recommendations. In Proceedings of the 13th ACM conference on recommender systems, pages 269–277. find
  38. Yantao Yu, Weipeng Wang, Zhoutian Feng, and Daiyue Xue. 2021. A dual augmented two-tower model for online large-scale recommendation. DLP-KDD. find
  39. Zheng Zhang, Jinyi Li, Yihuai Lan, Xiang Wang, and Hao Wang. 2025. An empirical study on prompt compression for large language models. arXiv preprint arXiv:2505.00019. — [arXiv:2505.00019](https://arxiv.org/abs/2505.00019) find
  40. Yongxin Zhu, Bocheng Li, Yifei Xin, Zhihua Xia, and Linli Xu. 2025. Addressing representation collapse in vector quantized models with one linear layer. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 22968–22977. 1790 8 find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/zhao-2026-novel-matching-paradigm-unified-generative/