YSGao/ReaMent

---
license: cc-by-nc-nd-4.0
task_categories:
- text-classification
language:
- en
tags:
- agent
pretty_name: ReaMent
size_categories:
- 1M<n<10M
---

<h1>Boosting Large Language Models for Mental Manipulation Detection via Data Augmentation and Distillation</h1>

[](https://arxiv.org/abs/2505.15255)

✨ Like ReaMent? Give us a ⭐ Star on GitHub! Your support keeps us going! [**Yuansheng-Gao/MentalMAD**](https://github.com/Yuansheng-Gao/MentalMAD)
# 🌿 ReaMent Dataset Card
A multi-round, real-world conversation-based mental manipulation detection dataset.

# 🧠 Dataset Summary
The ReaMent dataset was created to address the lack of real-world data in the field of mental manipulation detection.

- **Source**: The dataset is built from the YTD-18M corpus, which contains over 18 million dialogue-like segments extracted from unscripted interactions in web videos. These dialogues cover a wide range of everyday scenarios, such as interviews, group discussions, and situational conversations.
- **Size**: The final dataset consists of 5,000 high-quality annotated dialogues.
- **Diversity**: ReaMent captures a broader range of conversational contexts compared to scripted data, providing more natural and spontaneous interaction patterns.
- **Statistics**: Around 68.3% of dialogues in ReaMent were labeled as containing mental manipulation, while 31.7% were labeled as non-manipulative. The dataset has an average of 4 dialogue turns and 80 words per dialogue.

# 🤗 Key Contributions
- **Real-World Representation**: Unlike scripted or domain-specific datasets (e.g., MentalManip and LegalCon), ReaMent captures natural dialogues, making it valuable for detecting real-world mental manipulation.
- **Scalability**: It complements smaller datasets, offering richer and more representative data for training models that aim to detect manipulative behaviors in social interactions.

# 💻 Usage
```python
from datasets import load_dataset
ds = load_dataset("YSGao/ReaMent")
```

# 📝 Citation
```markdown
@inproceedings{gao2026boosting,
title={Boosting Large Language Models for Mental Manipulation Detection via Data Augmentation and Distillation},
author={Gao, Yuansheng and Gao, Peng and Bao, Han and Li, Bin and Luo, Jixiang and Wang, Zonghui and Chen, Wenzhi},
booktitle={Proceedings of the ACM Web Conference 2026},
pages={9033--9043},
year={2026}
}
```