snap-stanford/humanual-politics

---
license: cc-by-nc-4.0
language:
- en
tags:
- user-simulation
- humanlm
- persona
- long-form-content-and-politics
pretty_name: Humanual-Politics
size_categories:
- 10K<n<100K
---

# Humanual-Politics

[](https://humanlm.stanford.edu)
[](https://humanlm.stanford.edu/HumanLM_paper.pdf)
[](https://github.com/zou-group/humanlm)
[](https://huggingface.co/collections/snap-stanford/humanual-datasets)

Medium users responding to blog posts on political topics, featuring diverse political stances from users spanning different cultural backgrounds. This dataset is part of the **[HumanLM](https://humanlm.stanford.edu)** benchmark for training user simulators that accurately reflect real user behavior.

**Source:** RapidAPI Medium endpoint · **Domain:** Long-form Content & Politics · **Date Range:** 2022-04-01 to 2025-11-04

The dataset contains **47,905** comments from **5,300** users across **14,724** posts, with an average of **1.73** turns per conversation. Each example includes the user's persona, conversation context, and ground-truth response.

**Splits:** train (45,429) · val (489) · test (1,987)

| Column | Description |
|--------|-------------|
| `prompt` | Medium article content as a list of messages with `role` and `content` fields |
| `completion` | The ground-truth user comment to generate |
| `persona` | User's commenting history and political stances on Medium |
| `post_id` | Medium article ID |
| `user_id` | SHA-256 hashed Medium user ID (for privacy) |
| `timestamp` | Unix timestamp of when the comment was posted |
| `turn_id` | Position in the comment thread |
| `metadata` | Article metadata as JSON (title, claps, author, tags, etc.) |

## Quick Start

```python
from datasets import load_dataset

dataset = load_dataset("snap-stanford/humanual-politics")
sample = dataset["train"][0]

print(sample["persona"]) # User persona
print(sample["prompt"]) # Conversation context
print(sample["completion"]) # Ground-truth response
```

## Citation

```bibtex
@article{wu2026humanlm,
title={HUMANLM: Simulating Users with State Alignment Beats Response Imitation},
url={https://humanlm.stanford.edu/},
author={Wu, Shirley and Choi, Evelyn and Khatua, Arpandeep and Wang, Zhanghan and He-Yueya, Joy and Weerasooriya, Tharindu Cyril and Wei, Wei and Yang, Diyi and Leskovec, Jure and Zou, James},
year={2026}
}
```

Released under [CC BY-NC 4.0](https://creativecommons.org/licenses/by-nc/4.0/).