---
title: "Aligning Language Models to User Opinions — Hwang 2023"
description: "The authors show this approach misses much of the variation between individuals. 27 references, cited by 67 papers here."
authors: ["EunJeong Hwang", "Bodhisattwa Majumder", "Niket Tandon"]
year: 2023
venue: "Findings of the Association for Computational Linguistics: EMNLP 2023"
source: "https://aclanthology.org/2023.findings-emnlp.393/"
doi: "10.18653/v1/2023.findings-emnlp.393"
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/hwang-2023-aligning-language-models-user-opinions/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/hwang-2023-aligning-language-models-user-opinions/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2023](https://md.research.elamur.ai/free/en/papers/2023/page/6/) / Aligning Language Models to User Opinions

# Aligning Language Models to User Opinions

**Authors:** EunJeong Hwang, Bodhisattwa Majumder and [Niket Tandon](https://md.research.elamur.ai/free/en/authors/niket-tandon/) · **Year:** 2023 · **Venue:** [Findings of the Association for Computational Linguistics: EMNLP 2023](https://md.research.elamur.ai/free/en/venues/findings/) ([Findings of the Association for Computational Linguistics 2023](https://md.research.elamur.ai/free/en/venues/findings/2023/)) · **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/2023.findings-emnlp.393/) · [PDF](https://aclanthology.org/2023.findings-emnlp.393.pdf) · [DOI 10.18653/v1/2023.findings-emnlp.393](https://doi.org/10.18653/v1/2023.findings-emnlp.393) · [OpenAlex](https://openalex.org/W4389518978)

**Citations:** 23 citations (OpenAlex, 2026-10-14) · [67](https://md.research.elamur.ai/free/en/papers/hwang-2023-aligning-language-models-user-opinions/cited-by/) in our library, of the 141,840 papers whose bibliographies we read

## Summary
The paper asks what information an LLM needs to predict an individual's views: demographic traits, ideology, the person's own earlier answers, or a combination. Most persona prompting targets broad groups, such as asking a model to act as a member of a political party. The authors show this approach misses much of the variation between individuals. They analyzed PEW survey responses in the OpinionQA dataset and found that people with identical demographic profiles often answer quite differently. They also found that respondents with similar opinions frequently hold different ideologies. Building on this, they framed personalization as zero-shot multiple-choice QA over survey questions. Prompts were enriched step by step with ideology, demographics and a subset of the user's prior answers, selecting those most relevant to the target question. Combining all three sources gave the best alignment, raising accuracy by up to 7 points on some topics over demography- and ideology-based baselines. The work positions individual opinion history as a key ingredient for user-level alignment. It also discusses echo-chamber risks and possible mitigations.

## Key points
- **Data:** OpinionQA (Santurkar et al., 2023) is built from PEW surveys. It covers 15 topics, each with roughly 100 questions and about 5K respondents. Each respondent has eight demographic attributes plus political affiliation and ideology.
- **Agreement analysis:**
  - Users sharing all demographics reach only moderate agreement (Cohen's kappa around 0.4–0.5). Agreement is higher for Guns and Family & Relationships and lower for Race and America in 2050.
  - Among user pairs with similar opinions, about 80% differ in ideology.
- **Setup:**
  - 100 users were sampled per topic. 20% of each user's answers served as their persona and the remainder as test items.
  - Models tested were GPT-4, text-davinci-003, gpt-3.5-turbo, Vicuna-13b and LLaMA-7b.
- **Evaluation:** The authors report exact-choice accuracy and a "collapsed" accuracy, which merges fine-grained answer options into coarser classes.
- **Result:** Prompting with demographics, ideology and relevant past opinions together yields gains of up to about 7 accuracy points. Limitations include missing timestamps and restricting past opinions to the same topic.

*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

- [Allen Institute for Artificial Intelligence](https://md.research.elamur.ai/free/en/labs/allen-institute-for-artificial-intelligence/) — Research institute, Seattle, United States · 219 works · 89,712 citations
- [University of British Columbia](https://md.research.elamur.ai/free/en/universities/university-of-british-columbia/) — University, Vancouver, Canada · 46 works · 33,673 citations
- [Vector Institute](https://md.research.elamur.ai/free/en/labs/vector-institute/) — Research institute, Toronto, Canada · 19 works · 4,281 citations
- [Allen Institute](https://md.research.elamur.ai/free/en/labs/allen-institute/) — Research institute, Seattle, United States · 5 works · 4,252 citations

📚 Shelves (6): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 9,086 works · [🤖 Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) · 8,889 works · [🎭 Arts and Humanities](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/) · 9,296 works · [🗣️ Language and Linguistics](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/language-and-linguistics/) · 9,206 works · [🤝 Social Sciences](https://md.research.elamur.ai/free/en/shelves/social-sciences/) · 744 works · [🤝 Safety Research](https://md.research.elamur.ai/free/en/shelves/social-sciences/safety-research/) · 221 works

🧭 Topics (1): [🧠 Language models & pretraining](https://md.research.elamur.ai/free/en/topics/language-models/) · 1,343 works

## Cite

Cite as: EunJeong Hwang, Bodhisattwa Majumder and Niket Tandon. 2023. Aligning Language Models to User Opinions. Findings of the Association for Computational Linguistics: EMNLP 2023.

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

```bibtex
@inproceedings{hwang2023aligning,
  title = {{Aligning Language Models to User Opinions}},
  author = {EunJeong Hwang, Bodhisattwa Majumder and Niket Tandon},
  booktitle = {Findings of the Association for Computational Linguistics: EMNLP 2023},
  year = {2023},
  doi = {10.18653/v1/2023.findings-emnlp.393},
  url = {https://doi.org/10.18653/v1/2023.findings-emnlp.393}
}
```

- [← previous Skywork: A More Open Bilingual Foundation Model](https://md.research.elamur.ai/free/en/papers/wei-2023-skywork-more-open-bilingual/)
- [next → Cross-Lingual Consistency of Factual Knowledge in Multilingual Language Models](https://md.research.elamur.ai/free/en/papers/qi-2023-cross-lingual-consistency-factual-knowledge/)
- [↑ back to papers of 2023](https://md.research.elamur.ai/free/en/papers/2023/page/6/)

## References

27 references, 6 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Ante Busic-Sontic, Natalia V Czap, and Franz Fuerst. 2017. The role of personality traits in green decision-making. Journal of Economic Psychology, 62:313– 328. [find](https://scholar.google.com/scholar?q=Ante+Busic%2DSontic%2C+Natalia+V+Czap%2C+and+Franz+Fuerst%2E+2017%2E+The+role+of+personality+traits+in+green+decision%2Dmaking%2E+Journal+of+Economic+Psychology%2C+62%3A313%E2%80%93+328%2E)
2. Eugene Y. Chan and Mauricio Palmeira. 2021. Political ideology moderates consumer response to brand crisis apologies for data breaches. Computers in Human Behavior, 121:106801. [find](https://scholar.google.com/scholar?q=Eugene+Y%2E+Chan+and+Mauricio+Palmeira%2E+2021%2E+Political+ideology+moderates+consumer+response+to+brand+crisis+apologies+for+data+breaches%2E+Computers+in+Human+Behavior%2C+121%3A106801%2E)
3. Wei-Lin Chiang, Zhuohan Li, Zi Lin, Ying Sheng, Zhanghao Wu, Hao Zhang, Lianmin Zheng, Siyuan Zhuang, Yonghao Zhuang, Joseph E. Gonzalez, Ion Stoica, and Eric P. Xing. 2023. Vicuna: An open-source chatbot impressing gpt-4 with 90% chatgpt quality. [find](https://scholar.google.com/scholar?q=Wei%2DLin+Chiang%2C+Zhuohan+Li%2C+Zi+Lin%2C+Ying+Sheng%2C+Zhanghao+Wu%2C+Hao+Zhang%2C+Lianmin+Zheng%2C+Siyuan+Zhuang%2C+Yonghao+Zhuang%2C+Joseph+E%2E+Gonzalez%2C+Ion+Stoica%2C+and+Eric+P%2E+Xing%2E+2023%2E+Vicuna%3A+An+open%2Dsource+cha)
4. Eric Chu, Jacob Andreas, Stephen Ansolabehere, and Deb Roy. 2023. Language models trained on media diets can predict public opinion. [find](https://scholar.google.com/scholar?q=Eric+Chu%2C+Jacob+Andreas%2C+Stephen+Ansolabehere%2C+and+Deb+Roy%2E+2023%2E+Language+models+trained+on+media+diets+can+predict+public+opinion%2E)
5. Jacob Cohen. 1960. A coefficient of agreement for nominal scales. Educational and Psychological Measurement, 20:37 – 46. [find](https://scholar.google.com/scholar?q=Jacob+Cohen%2E+1960%2E+A+coefficient+of+agreement+for+nominal+scales%2E+Educational+and+Psychological+Measurement%2C+20%3A37+%E2%80%93+46%2E)
6. David Crockett and Melanie Wallendorf. 2004. The role of normative political ideology in consumer behavior. Journal of Consumer Research, 31:511–528. [find](https://scholar.google.com/scholar?q=David+Crockett+and+Melanie+Wallendorf%2E+2004%2E+The+role+of+normative+political+ideology+in+consumer+behavior%2E+Journal+of+Consumer+Research%2C+31%3A511%E2%80%93528%2E)
7. Michela Del Vicario, Gianna Vivaldo, Alessandro Bessi, Fabiana Zollo, Antonio Scala, Guido Caldarelli, and Walter Quattrociocchi. 2016. Echo chambers: Emotional contagion and group polarization on facebook. Scientific reports, 6(1):37825. [find](https://scholar.google.com/scholar?q=Michela+Del+Vicario%2C+Gianna+Vivaldo%2C+Alessandro+Bessi%2C+Fabiana+Zollo%2C+Antonio+Scala%2C+Guido+Caldarelli%2C+and+Walter+Quattrociocchi%2E+2016%2E+Echo+chambers%3A+Emotional+contagion+and+group+polarization+on+fac)
8. Ameet Deshpande, Vishvak Murahari, Tanmay Rajpurohit, Ashwin Kalyan, and Karthik Narasimhan. 2023. [Toxicity in chatgpt: Analyzing persona-assigned language models](https://md.research.elamur.ai/free/en/papers/deshpande-2023-toxicity-chatgpt-analyzing-persona-assigned/). arXiv preprint arXiv:2304.05335.
9. Yao Fu, Litu Ou, Mingyu Chen, Yuhao Wan, Hao-Chun Peng, and Tushar Khot. 2023. [Chain-of-thought hub: A continuous effort to measure large language models’ reasoning performance](https://md.research.elamur.ai/free/en/papers/fu-2023-chain-thought-hub-continuous-effort/). ArXiv, abs/2305.17306.
10. Yunfan Gao, Tao Sheng, Youlin Xiang, Yun Xiong, Haofen Wang, and Jiawei Zhang. 2023a. Chat-rec: Towards interactive and explainable llms-augmented recommender system. arXiv preprint arXiv:2303.14524. — [arXiv:2303.14524](https://arxiv.org/abs/2303.14524) [find](https://scholar.google.com/scholar?q=Yunfan+Gao%2C+Tao+Sheng%2C+Youlin+Xiang%2C+Yun+Xiong%2C+Haofen+Wang%2C+and+Jiawei+Zhang%2E+2023a%2E+Chat%2Drec%3A+Towards+interactive+and+explainable+llms%2Daugmented+recommender+system%2E+arXiv+preprint+arXiv%3A2303%2E14524%2E+)
11. Yunfan Gao, Tao Sheng, Youlin Xiang, Yun Xiong, Haofen Wang, and Jiawei Zhang. 2023b. Chat-rec: Towards interactive and explainable llms-augmented recommender system. [find](https://scholar.google.com/scholar?q=Yunfan+Gao%2C+Tao+Sheng%2C+Youlin+Xiang%2C+Yun+Xiong%2C+Haofen+Wang%2C+and+Jiawei+Zhang%2E+2023b%2E+Chat%2Drec%3A+Towards+interactive+and+explainable+llms%2Daugmented+recommender+system%2E)
12. Shibani Santurkar, Esin Durmus, Faisal Ladhak, Cinoo Lee, Percy Liang, and Tatsunori Hashimoto. 2023. [Whose opinions do language models reflect?](https://md.research.elamur.ai/free/en/papers/santurkar-2023-whose-opinions-language-models-reflect/) ArXiv, abs/2303.17548.
13. Xiangnan He, Lizi Liao, Hanwang Zhang, Liqiang Nie, Xia Hu, and Tat-Seng Chua. 2017. Neural collaborative filtering. [find](https://scholar.google.com/scholar?q=Xiangnan+He%2C+Lizi+Liao%2C+Hanwang+Zhang%2C+Liqiang+Nie%2C+Xia+Hu%2C+and+Tat%2DSeng+Chua%2E+2017%2E+Neural+collaborative+filtering%2E)
14. Hugo Touvron, Thibaut Lavril, Gautier Izacard, Xavier Martinet, Marie-Anne Lachaux, Timothée Lacroix, Baptiste Rozière, Naman Goyal, Eric Hambro, Faisal Azhar, Aurelien Rodriguez, Armand Joulin, Edouard Grave, and Guillaume Lample. 2023. [Llama: Open and efficient foundation language models](https://md.research.elamur.ai/free/en/papers/touvron-2023-llama-open-efficient-foundation-language/).
15. Liwei Jiang, Jena D. Hwang, Chandra Bhagavatula, Ronan Le Bras, Maxwell Forbes, Jon Borchardt, Jenny Liang, Oren Etzioni, Maarten Sap, and Yejin Choi. 2021. Delphi: Towards machine ethics and norms. ArXiv, abs/2110.07574. [find](https://scholar.google.com/scholar?q=Liwei+Jiang%2C+Jena+D%2E+Hwang%2C+Chandra+Bhagavatula%2C+Ronan+Le+Bras%2C+Maxwell+Forbes%2C+Jon+Borchardt%2C+Jenny+Liang%2C+Oren+Etzioni%2C+Maarten+Sap%2C+and+Yejin+Choi%2E+2021%2E+Delphi%3A+Towards+machine+ethics+and+norms%2E+A)
16. Urvashi Khandelwal, Omer Levy, Dan Jurafsky, Luke Zettlemoyer, and Mike Lewis. 2020. Generalization through memorization: Nearest neighbor language models. In International Conference on Learning Representations. [find](https://scholar.google.com/scholar?q=Urvashi+Khandelwal%2C+Omer+Levy%2C+Dan+Jurafsky%2C+Luke+Zettlemoyer%2C+and+Mike+Lewis%2E+2020%2E+Generalization+through+memorization%3A+Nearest+neighbor+language+models%2E+In+International+Conference+on+Learning+Repr)
17. Junsol Kim and Byungkyu Lee. 2023. Ai-augmented surveys: Leveraging large language models for opinion prediction in nationally representative surveys. [find](https://scholar.google.com/scholar?q=Junsol+Kim+and+Byungkyu+Lee%2E+2023%2E+Ai%2Daugmented+surveys%3A+Leveraging+large+language+models+for+opinion+prediction+in+nationally+representative+surveys%2E)
18. Marco Lauriola and Irwin P Levin. 2001. Personality traits and risky decision-making in a controlled experimental task: An exploratory study. Personality and individual differences, 31(2):215–226. [find](https://scholar.google.com/scholar?q=Marco+Lauriola+and+Irwin+P+Levin%2E+2001%2E+Personality+traits+and+risky+decision%2Dmaking+in+a+controlled+experimental+task%3A+An+exploratory+study%2E+Personality+and+individual+differences%2C+31%282%29%3A215%E2%80%93226%2E)
19. Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Brian Ichter, Fei Xia, Ed Chi, Quoc Le, and Denny Zhou. 2023. [Chain-of-thought prompting elicits reasoning in large language models](https://md.research.elamur.ai/free/en/papers/wei-2022-chain-thought-prompting-elicits-reasoning/).
20. Simone Weil. 1957. Écrits de Londres et dernières lettres. Gallimard. [find](https://scholar.google.com/scholar?q=Simone+Weil%2E+1957%2E+%C3%89crits+de+Londres+et+derni%C3%A8res+lettres%2E+Gallimard%2E)
21. Leor Zmigrod, Ian Eisenberg, Patrick Bissett, Trevor Robbins, and Russell Poldrack. 2021. The cognitive and perceptual correlates of ideological attitudes: A data-driven approach. Philosophical Transactions of the Royal Society B: Biological Sciences, 376:20200424. [find](https://scholar.google.com/scholar?q=Leor+Zmigrod%2C+Ian+Eisenberg%2C+Patrick+Bissett%2C+Trevor+Robbins%2C+and+Russell+Poldrack%2E+2021%2E+The+cognitive+and+perceptual+correlates+of+ideological+attitudes%3A+A+data%2Ddriven+approach%2E+Philosophical+Transa)
22. Shuyang Li, Bodhisattwa Prasad Majumder, and Julian McAuley. 2021. Self-supervised bot play for conversational recommendation with justifications. [find](https://scholar.google.com/scholar?q=Shuyang+Li%2C+Bodhisattwa+Prasad+Majumder%2C+and+Julian+McAuley%2E+2021%2E+Self%2Dsupervised+bot+play+for+conversational+recommendation+with+justifications%2E)
23. Aman Madaan, Niket Tandon, Peter Clark, and Yiming Yang. 2022. Memory-assisted prompt editing to improve GPT-3 after deployment. In Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, pages 2833–2861, Abu Dhabi, United Arab Emirates. Association for Computational… [find](https://scholar.google.com/scholar?q=Aman+Madaan%2C+Niket+Tandon%2C+Peter+Clark%2C+and+Yiming+Yang%2E+2022%2E+Memory%2Dassisted+prompt+editing+to+improve+GPT%2D3+after+deployment%2E+In+Proceedings+of+the+2022+Conference+on+Empirical+Methods+in+Natural+L)
24. Bodhisattwa Prasad Majumder, Shuyang Li, Jianmo Ni, and Julian McAuley. 2019. Generating personalized recipes from historical user preferences. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Pro… [find](https://scholar.google.com/scholar?q=Bodhisattwa+Prasad+Majumder%2C+Shuyang+Li%2C+Jianmo+Ni%2C+and+Julian+McAuley%2E+2019%2E+Generating+personalized+recipes+from+historical+user+preferences%2E+In+Proceedings+of+the+2019+Conference+on+Empirical+Metho)
25. David McLellan. 1989. pessimist. Springer. [find](https://scholar.google.com/scholar?q=David+McLellan%2E+1989%2E+pessimist%2E+Springer%2E)
26. Simone Weil: utopian Xiaoman Pan, Kai Sun, Dian Yu, Jianshu Chen, Heng Ji, Claire Cardie, and Dong Yu. 2019. Improving question answering with external knowledge. In Proceedings of the 2nd Workshop on Machine Reading for Question Answering, pages 27–37, Hong Kong, China. Association for Computationa… [find](https://scholar.google.com/scholar?q=Simone+Weil%3A+utopian+Xiaoman+Pan%2C+Kai+Sun%2C+Dian+Yu%2C+Jianshu+Chen%2C+Heng+Ji%2C+Claire+Cardie%2C+and+Dong+Yu%2E+2019%2E+Improving+question+answering+with+external+knowledge%2E+In+Proceedings+of+the+2nd+Workshop+on)
27. Alireza Salemi, Sheshera Mysore, Michael Bendersky, and Hamed Zamani. 2023. [Lamp: When large language models meet personalization](https://md.research.elamur.ai/free/en/papers/salemi-2024-lamp-when-large-language-models/).

*Updated 24 October 2026*