---
title: "Context Attribution with Multi-Armed Bandit Optimization (2026)"
description: "The authors present CAMAB, which frames the task of attributing an answer to individual context segments as a combinatorial… 24 references."
authors: ["Deng Pan", "Keerthiram Murugesan", "Ting Hua", "Nuno Moniz", "Nitesh V. Chawla"]
year: 2026
venue: "Findings of the Association for Computational Linguistics: ACL 2026"
source: "https://aclanthology.org/2026.findings-acl.565/"
doi: "10.18653/v1/2026.findings-acl.565"
licence: "CC BY 4.0 (ACL Anthology)"
licence_url: "https://creativecommons.org/licenses/by/4.0/"
summary_licence: "Elamur Cosmic License"
canonical: "https://research.elamur.ai/free/en/papers/pan-2026-context-attribution-multi-armed-bandit/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/pan-2026-context-attribution-multi-armed-bandit/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2026](https://md.research.elamur.ai/free/en/papers/2026/page/26/) / Context Attribution with Multi-Armed Bandit Optimization

# Context Attribution with Multi-Armed Bandit Optimization

**Authors:** Deng Pan, [Keerthiram Murugesan](https://md.research.elamur.ai/free/en/authors/keerthiram-murugesan/), Ting Hua, [Nuno Moniz](https://md.research.elamur.ai/free/en/authors/nuno-moniz/) and [Nitesh V. Chawla](https://md.research.elamur.ai/free/en/authors/nitesh-v-chawla/) · **Year:** 2026 · **Venue:** [Findings of the Association for Computational Linguistics: ACL 2026](https://md.research.elamur.ai/free/en/venues/findings/) ([Findings of the Association for Computational Linguistics 2026](https://md.research.elamur.ai/free/en/venues/findings/2026/)) · **Publisher:** [Association for Computational Linguistics](https://md.research.elamur.ai/free/en/publishers/association-for-computational-linguistics/) · **Licence:** [CC BY 4.0 (ACL Anthology)](https://creativecommons.org/licenses/by/4.0/)

**Links:** [Publisher's page](https://aclanthology.org/2026.findings-acl.565/) · [PDF](https://aclanthology.org/2026.findings-acl.565.pdf) · [DOI 10.18653/v1/2026.findings-acl.565](https://doi.org/10.18653/v1/2026.findings-acl.565) · [OpenAlex](https://openalex.org/W4414988062)

**Citations:** 0 citations (OpenAlex, 2026-10-14)

## Summary
In retrieval-augmented generation, it helps to know which retrieved passages actually shaped an LLM's answer. Post-hoc perturbation methods such as SHAP and ContextCite can show this, but they need many costly model calls. The authors present CAMAB, which frames the task of attributing an answer to individual context segments as a combinatorial multi-armed bandit problem. Each context sentence is an arm, and each tested subset of sentences is a super-arm. The reward for a subset is the mean token log-probability of the original answer when the model sees only that subset. A linear reward model with Gaussian noise is fitted with Linear Thompson Sampling. In each round, the method samples weights from the posterior, keeps the segments with positive sampled weights, queries the model, and updates the posterior. The final posterior means are the attribution scores. The approach needs only log-probabilities, so it also works with API-based models. Across QA and summarization benchmarks, CAMAB matches or beats the baselines while using up to about 30% fewer queries. This makes faithful attribution more affordable when inference budgets are limited.

## Key points
- **Setup:** Experiments use HotpotQA, CNN/DailyMail and TyDi QA with LLaMA-3.1-8B, plus Qwen-2.5-8B (reported in the appendix). Attribution is measured by Top-k Log-Probability Drop and BERTScore, with query budgets s = 20, 40 and 60.
- **Low-budget efficiency:** On HotpotQA and TyDi QA, CAMAB at 40 queries often does better than SHAP at 60. SHAP degrades sharply at s = 20. ContextCite, which relies on Lasso regression, trails under tight budgets.
- **CNN/DailyMail:** CAMAB stays within 1% of SHAP. The authors explain this by the lead bias of news articles, which lets all methods converge by about 40 queries.
- **Agreement with human annotations:** On 200 HotpotQA samples with gold supporting facts (s = 40), CAMAB beats SHAP on every metric, including P@1 (0.780 vs 0.680), AUROC (0.855 vs 0.806) and AP (0.688 vs 0.598).
- **Interactions and limitations:** The posterior precision matrix implicitly captures dependencies between segments. The method requires token-level log-probabilities and may get stuck in local optima in noisy settings. Code is available at github.com/pd90506/camab.

*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

- [IBM](https://md.research.elamur.ai/free/en/companies/ibm/) — Company, Armonk, United States · 139 works · 70,691 citations
- [University of Notre Dame](https://md.research.elamur.ai/free/en/universities/university-of-notre-dame/) — University, Notre Dame, United States · 51 works · 36,717 citations

📚 Shelves (6): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 16,667 works · [🤖 Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) · 16,194 works · [🎭 Arts and Humanities](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/) · 16,039 works · [🗣️ Language and Linguistics](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/language-and-linguistics/) · 15,891 works · [🎯 Decision Sciences](https://md.research.elamur.ai/free/en/shelves/decision-sciences/) · 209 works · [🎯 Management Science and Operations Research](https://md.research.elamur.ai/free/en/shelves/decision-sciences/management-science-and-operations-research/) · 108 works

## Cite

Cite as: Deng Pan, Keerthiram Murugesan, Ting Hua, Nuno Moniz and Nitesh V. Chawla. 2026. Context Attribution with Multi-Armed Bandit Optimization. Findings of the Association for Computational Linguistics: ACL 2026.

The paper itself, not this page. BibTeX from the library's record of the work.

```bibtex
@inproceedings{pan2026context,
  title = {{Context Attribution with Multi-Armed Bandit Optimization}},
  author = {Deng Pan, Keerthiram Murugesan, Ting Hua, Nuno Moniz and Nitesh V. Chawla},
  booktitle = {Findings of the Association for Computational Linguistics: ACL 2026},
  year = {2026},
  doi = {10.18653/v1/2026.findings-acl.565},
  url = {https://doi.org/10.18653/v1/2026.findings-acl.565}
}
```

- [← previous contestant001 at SemEval-2026 Task 13 Stylometric and TF-IDF-Based Detection of Machine-Generated Code](https://md.research.elamur.ai/free/en/papers/ozaylar-2026-contestant001-semeval-2026-task-13/)
- [next → Context-8: A Data Set for Evaluating Context Sensitivity in Machine Translation](https://md.research.elamur.ai/free/en/papers/wang-2026-context-8-data-set-evaluating/)
- [↑ back to papers of 2026](https://md.research.elamur.ai/free/en/papers/2026/page/26/)

## References

24 references, 8 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Shipra Agrawal and Navin Goyal. 2013. Thompson sampling for contextual bandits with linear payoffs. In International conference on machine learning, pages 127–135. PMLR. [find](https://scholar.google.com/scholar?q=Shipra+Agrawal+and+Navin+Goyal%2E+2013%2E+Thompson+sampling+for+contextual+bandits+with+linear+payoffs%2E+In+International+conference+on+machine+learning%2C+pages+127%E2%80%93135%2E+PMLR%2E)
2. Yunfan Gao, Yun Xiong, Xinyu Gao, Kangxiang Jia, Jinliu Pan, Yuxi Bi, Yixin Dai, Jiawei Sun, Haofen Wang, and Haofen Wang. 2023b. [Retrieval-augmented generation for large language models: A survey](https://md.research.elamur.ai/free/en/papers/gao-2023-retrieval-augmented-generation-large-language/). arXiv preprint arXiv:2312.10997, 2:1.
3. Aaron Grattafiori, Abhimanyu Dubey, Abhinav Jauhri, Abhinav Pandey, Abhishek Kadian, Ahmad Al-Dahle, Aiesha Letman, Akhil Mathur, Alan Schelten, Alex Vaughan, and 1 others. 2024. [The llama 3 herd of models](https://md.research.elamur.ai/free/en/papers/grattafiori-2024-llama-3-herd-models/). arXiv preprint arXiv:2407.21783.
4. Chengyu Huang, Zeqiu Wu, Yushi Hu, and Wenya Wang. 2024. Training language models to generate text with citations via fine-grained rewards. arXiv preprint arXiv:2402.04315. [find](https://scholar.google.com/scholar?q=Chengyu+Huang%2C+Zeqiu+Wu%2C+Yushi+Hu%2C+and+Wenya+Wang%2E+2024%2E+Training+language+models+to+generate+text+with+citations+via+fine%2Dgrained+rewards%2E+arXiv+preprint+arXiv%3A2402%2E04315%2E)
5. Neil Jethani, Mukund Sudarshan, Ian Connick Covert, Su-In Lee, and Rajesh Ranganath. 2021. Fastshap: Real-time shapley value estimation. In International conference on learning representations. [find](https://scholar.google.com/scholar?q=Neil+Jethani%2C+Mukund+Sudarshan%2C+Ian+Connick+Covert%2C+Su%2DIn+Lee%2C+and+Rajesh+Ranganath%2E+2021%2E+Fastshap%3A+Real%2Dtime+shapley+value+estimation%2E+In+International+conference+on+learning+representations%2E)
6. Scott M Lundberg and Su-In Lee. 2017. [A unified approach to interpreting model predictions](https://md.research.elamur.ai/free/en/papers/lundberg-2017-unified-approach-interpreting-model-predictions/). In Advances in neural information processing systems, pages 4765–4774.
7. Jacob Menick, Maja Trebacz, Vladimir Mikulik, John Aslanides, Francis Song, Martin Chadwick, Mia Glaese, Susannah Young, Lucy Campbell-Gillingham, Geoffrey Irving, and 1 others. 2022. [Teaching language models to support answers with verified quotes](https://md.research.elamur.ai/free/en/papers/menick-2022-teaching-language-models-support-answers/). arXiv preprint arXiv:2203.11147.
8. Sumana Sanyasipura Nagaraju. 2025. Automation and feature selection enhancement with reinforcement learning (rl). [find](https://scholar.google.com/scholar?q=Sumana+Sanyasipura+Nagaraju%2E+2025%2E+Automation+and+feature+selection+enhancement+with+reinforcement+learning+%28rl%29%2E)
9. Hanjie Chen, Guangtao Zheng, and Yangfeng Ji. 2020. Generating hierarchical explanations on text classification via feature interaction detection. arXiv preprint arXiv:2004.02015. [find](https://scholar.google.com/scholar?q=Hanjie+Chen%2C+Guangtao+Zheng%2C+and+Yangfeng+Ji%2E+2020%2E+Generating+hierarchical+explanations+on+text+classification+via+feature+interaction+detection%2E+arXiv+preprint+arXiv%3A2004%2E02015%2E)
10. Reiichiro Nakano, Jacob Hilton, Suchir Balaji, Jeff Wu, Long Ouyang, Christina Kim, Christopher Hesse, Shantanu Jain, Vineet Kosaraju, William Saunders, and 1 others. 2021. [Webgpt: Browser-assisted question-answering with human feedback](https://md.research.elamur.ai/free/en/papers/nakano-2021-webgpt-browser-assisted-question-answering/). arXiv preprint arXiv:2112.09332.
11. Jonathan H Clark, Eunsol Choi, Michael Collins, Dan Garrette, Tom Kwiatkowski, Vitaly Nikolaev, and Jennimaria Palomaki. 2020. Tydi qa: A benchmark for information-seeking question answering in ty pologically di verse languages. Transactions of the Association for Computational Linguistics, 8:454–47… [find](https://scholar.google.com/scholar?q=Jonathan+H+Clark%2C+Eunsol+Choi%2C+Michael+Collins%2C+Dan+Garrette%2C+Tom+Kwiatkowski%2C+Vitaly+Nikolaev%2C+and+Jennimaria+Palomaki%2E+2020%2E+Tydi+qa%3A+A+benchmark+for+information%2Dseeking+question+answering+in+ty+pol)
12. Lucas Monteiro Paes, Dennis Wei, Hyo Jin Do, Hendrik Strobelt, Ronny Luss, Amit Dhurandhar, Manish Nagireddy, Karthikeyan Natesan Ramamurthy, Prasanna Sattigeri, Werner Geyer, and 1 others. 2024. Multi-level explanations for generative language models. arXiv preprint arXiv:2403.14459. [find](https://scholar.google.com/scholar?q=Lucas+Monteiro+Paes%2C+Dennis+Wei%2C+Hyo+Jin+Do%2C+Hendrik+Strobelt%2C+Ronny+Luss%2C+Amit+Dhurandhar%2C+Manish+Nagireddy%2C+Karthikeyan+Natesan+Ramamurthy%2C+Prasanna+Sattigeri%2C+Werner+Geyer%2C+and+1+others%2E+2024%2E+Mult)
13. Benjamin Cohen-Wang, Harshay Shah, Kristian Georgiev, and Aleksander Madry. 2024. Contextcite: Attributing model generation to context. Advances in Neural Information Processing Systems, 37:95764– 95807. [find](https://scholar.google.com/scholar?q=Benjamin+Cohen%2DWang%2C+Harshay+Shah%2C+Kristian+Georgiev%2C+and+Aleksander+Madry%2E+2024%2E+Contextcite%3A+Attributing+model+generation+to+context%2E+Advances+in+Neural+Information+Processing+Systems%2C+37%3A95764%E2%80%93+958)
14. Deng Pan, Xin Li, and Dongxiao Zhu. 2021. Explaining deep neural network models with adversarial gradient integration. In Thirtieth International Joint Conference on Artificial Intelligence (IJCAI). [find](https://scholar.google.com/scholar?q=Deng+Pan%2C+Xin+Li%2C+and+Dongxiao+Zhu%2E+2021%2E+Explaining+deep+neural+network+models+with+adversarial+gradient+integration%2E+In+Thirtieth+International+Joint+Conference+on+Artificial+Intelligence+%28IJCAI%29%2E)
15. Yue Dong, Yikang Shen, Eric Crawford, Herke van Hoof, and Jackie Chi Kit Cheung. 2018. Banditsum: Extractive summarization as a contextual bandit. arXiv preprint arXiv:1809.09672. [find](https://scholar.google.com/scholar?q=Yue+Dong%2C+Yikang+Shen%2C+Eric+Crawford%2C+Herke+van+Hoof%2C+and+Jackie+Chi+Kit+Cheung%2E+2018%2E+Banditsum%3A+Extractive+summarization+as+a+contextual+bandit%2E+arXiv+preprint+arXiv%3A1809%2E09672%2E)
16. Tianyu Gao, Howard Yen, Jiatong Yu, and Danqi Chen. 2023a. [Enabling large language models to generate text with citations](https://md.research.elamur.ai/free/en/papers/gao-2023-enabling-large-language-models-generate/). arXiv preprint arXiv:2305.14627.
17. Deng Pan, Nuno Moniz, and Nitesh Chawla. 2025. Fast explanations via policy gradient-optimized explainer. Preprint, arXiv:2405.18664. — [arXiv:2405.18664](https://arxiv.org/abs/2405.18664) [find](https://scholar.google.com/scholar?q=Deng+Pan%2C+Nuno+Moniz%2C+and+Nitesh+Chawla%2E+2025%2E+Fast+explanations+via+policy+gradient%2Doptimized+explainer%2E+Preprint%2C+arXiv%3A2405%2E18664%2E+%E2%80%94+%5BarXiv%3A2405%2E18664%5D%28https%3A%2F%2Farxiv%2Eorg%2Fabs%2F2405%2E18664%29)
18. Qwen, :, An Yang, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chengyuan Li, Dayiheng Liu, Fei Huang, Haoran Wei, Huan Lin, Jian Yang, Jianhong Tu, Jianwei Zhang, Jianxin Yang, Jiaxi Yang, Jingren Zhou, and 25 others. 2025. Qwen2.5 technical report. Preprint, arXiv:2412.15115. [find](https://scholar.google.com/scholar?q=Qwen%2C+%3A%2C+An+Yang%2C+Baosong+Yang%2C+Beichen+Zhang%2C+Binyuan+Hui%2C+Bo+Zheng%2C+Bowen+Yu%2C+Chengyuan+Li%2C+Dayiheng+Liu%2C+Fei+Huang%2C+Haoran+Wei%2C+Huan+Lin%2C+Jian+Yang%2C+Jianhong+Tu%2C+Jianwei+Zhang%2C+Jianxin+Yang%2C+Jiaxi+)
19. Marco Tulio Ribeiro, Sameer Singh, and Carlos Guestrin. 2016. [" why should i trust you?" explaining the predictions of any classifier](https://md.research.elamur.ai/free/en/papers/ribeiro-2016-why-should-i-trust-explaining/). In Proceedings of the 22nd ACM SIGKDD international conference on knowledge discovery and data mining, pages 1135– 1144.
20. Abigail See, Peter J Liu, and Christopher D Manning. 2017. Get to the point: Summarization with pointer-generator networks. arXiv preprint arXiv:1704.04368. [find](https://scholar.google.com/scholar?q=Abigail+See%2C+Peter+J+Liu%2C+and+Christopher+D+Manning%2E+2017%2E+Get+to+the+point%3A+Summarization+with+pointer%2Dgenerator+networks%2E+arXiv+preprint+arXiv%3A1704%2E04368%2E)
21. Mahesh Sudhakar, Sam Sattarzadeh, Konstantinos N Plataniotis, Jongseong Jang, Yeonjeong Jeong, and Hyunwoo Kim. 2021. Ada-sise: adaptive semantic input sampling for efficient explanation of convolutional neural networks. In ICASSP 2021-2021 IEEE International Conference on Acoustics, Speech and Sign… [find](https://scholar.google.com/scholar?q=Mahesh+Sudhakar%2C+Sam+Sattarzadeh%2C+Konstantinos+N+Plataniotis%2C+Jongseong+Jang%2C+Yeonjeong+Jeong%2C+and+Hyunwoo+Kim%2E+2021%2E+Ada%2Dsise%3A+adaptive+semantic+input+sampling+for+efficient+explanation+of+convolutio)
22. Zhilin Yang, Peng Qi, Saizheng Zhang, Yoshua Bengio, William W Cohen, Ruslan Salakhutdinov, and Christopher D Manning. 2018. [Hotpotqa: A dataset for diverse, explainable multi-hop question answering](https://md.research.elamur.ai/free/en/papers/yang-2018-hotpotqa-dataset-diverse-explainable-multi/). arXiv preprint arXiv:1809.09600.
23. Jiajie Zhang, Yushi Bai, Xin Lv, Wanjun Gu, Danqing Liu, Minhao Zou, Shulin Cao, Lei Hou, Yuxiao Dong, Ling Feng, and 1 others. 2024. Longcite: Enabling llms to generate fine-grained citations in long-context qa. arXiv preprint arXiv:2409.02897. [find](https://scholar.google.com/scholar?q=Jiajie+Zhang%2C+Yushi+Bai%2C+Xin+Lv%2C+Wanjun+Gu%2C+Danqing+Liu%2C+Minhao+Zou%2C+Shulin+Cao%2C+Lei+Hou%2C+Yuxiao+Dong%2C+Ling+Feng%2C+and+1+others%2E+2024%2E+Longcite%3A+Enabling+llms+to+generate+fine%2Dgrained+citations+in+long)
24. Tianyi Zhang, Varsha Kishore, Felix Wu, Kilian Q Weinberger, and Yoav Artzi. 2019. Bertscore: Evaluating text generation with bert. arXiv preprint arXiv:1904.09675. [find](https://scholar.google.com/scholar?q=Tianyi+Zhang%2C+Varsha+Kishore%2C+Felix+Wu%2C+Kilian+Q+Weinberger%2C+and+Yoav+Artzi%2E+2019%2E+Bertscore%3A+Evaluating+text+generation+with+bert%2E+arXiv+preprint+arXiv%3A1904%2E09675%2E)

*Updated 1 November 2026*