Elamur Research⌕
☰
home / papers / 2025 / Improving Contextual Faithfulness of Large Language Models via Retrieval Heads-Induced Optimization

Improving Contextual Faithfulness of Large Language Models via Retrieval Heads-Induced Optimization

Paper · 2025 · Lei Huang, Xiaocheng Feng, Weitao Ma et al.

Summary

The paper addresses faithfulness hallucination in retrieval-augmented long-form question answering (LFQA), where a model writes paragraph-length answers containing claims that the retrieved documents do not support or that contradict them. A pilot study on CLAPNQ with Llama-2 models (7B–70B) found that disabling retrieval heads, the attention heads that copy information from context, steadily reduces faithfulness. Masking random non-retrieval heads has no such effect. The errors caused by masking also resemble the model's own natural mistakes.

The authors build on this finding with RHIO (Retrieval Heads-Induced Optimization), which has three parts:

  1. Negative data generation. Unfaithful training answers are produced by masking the top retrieval heads.
  2. Faithfulness-Aware Tuning. The model is trained on both faithful and unfaithful answers, each labelled with a control token.
  3. Contrastive decoding. At inference, the model generates under both tokens, and the gap between the two outputs is amplified.

They also release GroundBench, an evaluation set built from five existing LFQA datasets. In every GroundBench example, the context is guaranteed to contain enough information to answer the question. On this benchmark, RHIO clearly improves faithfulness and outperforms GPT-4o. This points to model internals as a way to make RAG systems more trustworthy.

Key points

  • Error patterns match. Answers produced with retrieval heads masked show an error mix close to the model's natural errors:
    • masked heads: 50% incomplete, 36% fabricated, 14% inconsistent;
    • natural errors: 48% incomplete, 40% fabricated, 12% inconsistent.
  • Method details. Negative samples come from zeroing the attention of the top N=100 retrieval heads. These heads are found with the detection algorithm of Wu et al. (2024). The control tokens are [POS] (faithful) and [NEG] (unfaithful).
  • Main results. Average faithfulness gains on GroundBench are 12.84% for the 7B model and 12.59% for the 13B model, exceeding GPT-4o. Baselines include Self-RAG and RECOMP.
  • Limitations.
    • Experiments use only Llama-2 models.
    • GroundBench excludes unfaithfulness caused by retrieval failures.
    • The method cannot target specific error types.
  • Availability. Code is released at github.com/LuckyyySTA/faithful-LFQA. The paper appeared at ACL 2025.

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

🎓Harbin Institute of TechnologyHarbin · China185 works · 29,140 citations🎓University of Hong KongPok Fu Lam · China118 works · 13,496 citations🔬Peng Cheng LaboratoryShenzhen · China47 works · 2,699 citations🎓Northeastern UniversityShenyang · China53 works · 2,205 citations🏢IFlyTekHefei · China11 works · 2,810 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 worksInformation retrieval & search · 552 works

Cite

Cite as Lei Huang, Xiaocheng Feng, Weitao Ma, Yuchun Fan, Xiachong Feng, Yangfan Ye, Weihong Zhong, Yuxuan Gu, Baoxin Wang, Dayong Wu, Guoping Hu and Bing Qin. 2025. Improving Contextual Faithfulness of Large Language Models via Retrieval Heads-Induced Optimization. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers).

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2025.acl-long.826

@inproceedings{huang2025improving,
  title = {{Improving Contextual Faithfulness of Large Language Models via Retrieval Heads-Induced Optimization}},
  author = {Lei Huang, Xiaocheng Feng, Weitao Ma, Yuchun Fan, Xiachong Feng, Yangfan Ye, Weihong Zhong, Yuxuan Gu, Baoxin Wang, Dayong Wu, Guoping Hu and Bing Qin},
  booktitle = {Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)},
  year = {2025},
  doi = {10.18653/v1/2025.acl-long.826},
  url = {https://doi.org/10.18653/v1/2025.acl-long.826}
}

References

39 references, 27 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Akari Asai, Zeqiu Wu, Yizhong Wang, Avirup Sil, and Hannaneh Hajishirzi. 2024a. Self-rag: Learning to retrieve, generate, and critique through self-reflection. In The Twelfth International Conference on Learning Representations, ICLR 2024, Vienna, Austria, May 7-11, 2024. OpenReview.net.
  2. Akari Asai, Zexuan Zhong, Danqi Chen, Pang Wei Koh, Luke Zettlemoyer, Hannaneh Hajishirzi, and Wen-tau Yih. 2024b. Reliable, adaptable, and attributable language models with retrieval. CoRR, abs/2403.03187.
  3. Hung-Ting Chen, Fangyuan Xu, Shane A. Arora, and Eunsol Choi. 2023. Understanding retrieval augmentation for long-form question answering. CoRR, abs/2310.12150.
  4. Hyung Won Chung, Le Hou, Shayne Longpre, Barret Zoph, Yi Tay, William Fedus, Yunxuan Li, Xuezhi Wang, Mostafa Dehghani, Siddhartha Brahma, Albert Webson, Shixiang Shane Gu, Zhuyun Dai, Mirac Suzgun, Xinyun Chen, Aakanksha Chowdhery, Alex Castro-Ros, Marie Pellat, Kevin Robinson, Dasha Valter, Sharan… find
  5. Angela Fan, Yacine Jernite, Ethan Perez, David Grangier, Jason Weston, and Michael Auli. 2019. ELI5: long form question answering. In Proceedings of the 57th Conference of the Association for Computational Linguistics, ACL 2019, Florence, Italy, July 28- August 2, 2019, Volume 1: Long Papers, pages…
  6. Yuchun Fan, Yongyu Mu, Yilin Wang, Lei Huang, Junhao Ruan, Bei Li, Tong Xiao, Shujian Huang, Xiaocheng Feng, and Jingbo Zhu. 2025. SLAM: towards efficient multilingual reasoning via selective language alignment. In Proceedings of the 31st International Conference on Computational Linguistics, COLING…
  7. Tianyu Gao, Howard Yen, Jiatong Yu, and Danqi Chen. 2023a. Enabling large language models to generate text with citations. In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, EMNLP 2023, Singapore, December 6-10, 2023, pages 6465–6488. Association for Computati…
  8. Yunfan Gao, Yun Xiong, Xinyu Gao, Kangxiang Jia, Jinliu Pan, Yuxi Bi, Yi Dai, Jiawei Sun, Qianyu Guo, Meng Wang, and Haofen Wang. 2023b. Retrieval-augmented generation for large language models: A survey. CoRR, abs/2312.10997.
  9. Rujun Han, Yuhao Zhang, Peng Qi, Yumo Xu, Jenyuan Wang, Lan Liu, William Yang Wang, Bonan Min, and Vittorio Castelli. 2024. Rag-qa arena: Evaluating domain robustness for long-form retrieval augmented question answering. Preprint, arXiv:2407.13998. find
  10. Lei Huang, Xiaocheng Feng, Weitao Ma, Yuxuan Gu, Weihong Zhong, Xiachong Feng, Weijiang Yu, Weihua Peng, Duyu Tang, Dandan Tu, and Bing Qin. 2024a. Learning fine-grained grounded citations for attributed large language models. In Findings of the Association for Computational Linguistics, ACL 2024, B… find
  11. Lei Huang, Xiaocheng Feng, Weitao Ma, Liang Zhao, Yuchun Fan, Weihong Zhong, Dongliang Xu, Qing Yang, Hongtao Liu, and Bing Qin. 2024b. Advancing large language model attribution through self-improving. In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, EMNLP… find
  12. Lei Huang, Weijiang Yu, Weitao Ma, Weihong Zhong, Zhangyin Feng, Haotian Wang, Qianglong Chen, Weihua Peng, Xiaocheng Feng, Bing Qin, and Ting Liu. 2023. A survey on hallucination in large language models: Principles, taxonomy, challenges, and open questions. CoRR, abs/2311.05232.
  13. Albert Q. Jiang, Alexandre Sablayrolles, Arthur Mensch, Chris Bamford, Devendra Singh Chaplot, Diego de Las Casas, Florian Bressand, Gianna Lengyel, Guillaume Lample, Lucile Saulnier, Lélio Renard Lavaud, Marie-Anne Lachaux, Pierre Stock, Teven Le Scao, Thibaut Lavril, Thomas Wang, Timothée Lacroix,… find
  14. Ehsan Kamalloo, Aref Jafari, Xinyu Zhang, Nandan Thakur, and Jimmy Lin. 2023. HAGRID: A human-llm collaborative dataset for generative information-seeking with attribution. CoRR, abs/2307.16883. — [arXiv:2307.16883](https://arxiv.org/abs/2307.16883) find
  15. Tom Kwiatkowski, Jennimaria Palomaki, Olivia Redfield, Michael Collins, Ankur P. Parikh, Chris Alberti, Danielle Epstein, Illia Polosukhin, Jacob Devlin, Kenton Lee, Kristina Toutanova, Llion Jones, Matthew Kelcey, Ming-Wei Chang, Andrew M. Dai, Jakob Uszkoreit, Quoc Le, and Slav Petrov. 2019. Natur…
  16. Woosuk Kwon, Zhuohan Li, Siyuan Zhuang, Ying Sheng, Lianmin Zheng, Cody Hao Yu, Joseph E. Gonzalez, Hao Zhang, and Ion Stoica. 2023. Efficient memory management for large language model serving with pagedattention. In Proceedings of the ACM SIGOPS 29th Symposium on Operating Systems Principles.
  17. Junyi Li, Xiaoxue Cheng, Xin Zhao, Jian-Yun Nie, and Ji-Rong Wen. 2023a. Halueval: A large-scale hallucination evaluation benchmark for large language models. In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, EMNLP 2023, Singapore, December 6-10, 2023, pages…
  18. Xiang Lisa Li, Ari Holtzman, Daniel Fried, Percy Liang, Jason Eisner, Tatsunori Hashimoto, Luke Zettlemoyer, and Mike Lewis. 2023b. Contrastive decoding: Open-ended text generation as optimization. In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1:…
  19. Ximing Lu, Sean Welleck, Liwei Jiang, Jack Hessel, Lianhui Qin, Peter West, Prithviraj Ammanabrolu, and Yejin Choi. 2022. Quark: Controllable text generation with reinforced unlearning. CoRR, abs/2205.13636.
  20. Chaitanya Malaviya, Subin Lee, Sihao Chen, Elizabeth Sieber, Mark Yatskar, and Dan Roth. 2023. Expertqa: Expert-curated questions and attributed answers. CoRR, abs/2309.07852.
  21. Alex Mallen, Akari Asai, Victor Zhong, Rajarshi Das, Daniel Khashabi, and Hannaneh Hajishirzi. 2023. When not to trust language models: Investigating effectiveness of parametric and non-parametric memories. In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (V… find
  22. Abhika Mishra, Akari Asai, Vidhisha Balachandran, Yizhong Wang, Graham Neubig, Yulia Tsvetkov, and Hannaneh Hajishirzi. 2024. Fine-grained hallucination detection and editing for language models. CoRR, abs/2401.06855.
  23. Reiichiro Nakano, Jacob Hilton, Suchir Balaji, Jeff Wu, Long Ouyang, Christina Kim, Christopher Hesse, Shantanu Jain, Vineet Kosaraju, William Saunders, Xu Jiang, Karl Cobbe, Tyna Eloundou, Gretchen Krueger, Kevin Button, Matthew Knight, Benjamin Chess, and John Schulman. 2021. Webgpt: Browser-assis…
  24. Xuan-Phi Nguyen, Shrey Pandit, Senthil Purushwalkam, Austin Xu, Hailin Chen, Yifei Ming, Zixuan Ke, Silvio Savarese, Caiming Xong, and Shafiq Joty. 2024. Sfr-rag: Towards contextually faithful llms. Preprint, arXiv:2409.09916. — [arXiv:2409.09916](https://arxiv.org/abs/2409.09916) find
  25. Catherine Olsson, Nelson Elhage, Neel Nanda, Nicholas Joseph, Nova DasSarma, Tom Henighan, Ben Mann, Amanda Askell, Yuntao Bai, Anna Chen, Tom Conerly, Dawn Drain, Deep Ganguli, Zac Hatfield-Dodds, Danny Hernandez, Scott Johnston, Andy Jones, Jackson Kernion, Liane Lovitt, Kamal Ndousse, Dario Amode… — In-context Learning and Induction Heads
  26. OpenAI. 2023. GPT-4 technical report. abs/2303.08774.
  27. CoRR, Rafael Rafailov, Archit Sharma, Eric Mitchell, Stefano Ermon, Christopher D. Manning, and Chelsea Finn. 2023. Direct preference optimization: Your language model is secretly a reward model. CoRR, abs/2305.18290.
  28. Jeff Rasley, Samyam Rajbhandari, Olatunji Ruwase, and Yuxiong He. 2020. Deepspeed: System optimizations enable training deep learning models with over 100 billion parameters. In KDD ’20: The 26th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, Virtual Event, CA, USA, August 23-27, 2020… find
  29. Sara Rosenthal, Avirup Sil, Radu Florian, and Salim Roukos. 2024. CLAPNQ: cohesive long-form answers from passages in natural questions for RAG systems. CoRR, abs/2404.02103.
  30. Tal Schuster, Ádám D. Lelkes, Haitian Sun, Jai Gupta, Jonathan Berant, William W. Cohen, and Donald Metzler. 2024. SEMQA: semi-extractive multi-source question answering. In Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Langu… find
  31. Weijia Shi, Xiaochuang Han, Mike Lewis, Yulia Tsvetkov, Luke Zettlemoyer, and Wen-tau Yih. 2024. Trusting your evidence: Hallucinate less with context-aware decoding. In Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language…
  32. Ivan Stelmakh, Yi Luan, Bhuwan Dhingra, and Ming-Wei Chang. 2022. ASQA: factoid questions meet long-form answers. CoRR, abs/2204.06092.
  33. Alessandro Stolfo. 2024. Groundedness in retrieval-augmented long-form generation: An empirical study. CoRR, abs/2404.07060. find
  34. Liyan Tang, Philippe Laban, and Greg Durrett. 2024. Minicheck: Efficient fact-checking of llms on grounding documents. CoRR, abs/2404.10774.
  35. Hugo Touvron, Louis Martin, Kevin Stone, Peter Albert, Amjad Almahairi, Yasmine Babaei, Nikolay Bashlykov, Soumya Batra, Prajjwal Bhargava, Shruti Bhosale, Dan Bikel, Lukas Blecher, Cristian Canton-Ferrer, Moya Chen, Guillem Cucurull, David Esiobu, Jude Fernandes, Jeremy Fu, Wenyin Fu, Brian Fuller,… — Llama 2: Open Foundation and Fine-Tuned Chat Models
  36. Han Wang, Archiki Prasad, Elias Stengel-Eskin, and Mohit Bansal. 2024. Adacad: Adaptively decoding to balance conflicts between contextual and parametric knowledge. Preprint, arXiv:2409.07394. find
  37. Wenhao Wu, Yizhong Wang, Guangxuan Xiao, Hao Peng, and Yao Fu. 2024. Retrieval head mechanistically explains long-context factuality. CoRR, abs/2404.15574.
  38. Fangyuan Xu, Weijia Shi, and Eunsol Choi. 2023a. RECOMP: improving retrieval-augmented lms with compression and selective augmentation. CoRR, abs/2310.04408.
  39. Fangyuan Xu, Yixiao Song, Mohit Iyyer, and Eunsol Choi. 2023b. A critical evaluation of evaluations for long-form question answering. In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), ACL 2023, Toronto, Canada, July 9-14, 2023, pages…

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/huang-2025-improving-contextual-faithfulness-large-language/