YSGao/ReaMent

Boosting Large Language Models for Mental Manipulation Detection via Data Augmentation and Distillation

✨ Like ReaMent? Give us a ⭐ Star on GitHub! Your support keeps us going! 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

from datasets import load_dataset
ds = load_dataset("YSGao/ReaMent")

📝 Citation

@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}
}