YSGao/MentalMAD_Augmented_Data

---
license: cc-by-nc-nd-4.0
task_categories:
- question-answering
language:
- en
tags:
- agent
pretty_name: mentalmad_augmented_data
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 MentalMAD? Give us a ⭐ Star on GitHub! Your support keeps us going! [**Yuansheng-Gao/MentalMAD**](https://github.com/Yuansheng-Gao/MentalMAD)

# 🧠 Summary

Except for “LegalCon_naive_data.json”, all other data are generated by LLaMA3-70B-Instruct under the EvoSA or complementary task settings. “LegalCon_naive_data.json” is used for comparison with the EvoSA-based LLM-based label-preserving augmentation method.

# 🤗 Explanation

DDD_EvoSA_data.json represents the data expanded by EvoSA on the DDD dataset.\
DDD_EvoSA_TTT_data.json represents the data of the complementary task TTT, which was enhanced after EvoSA expanded the dataset on the DDD dataset.\
DDD_TTT_data.json represents the data of the complementary task TTT, which was enhanced on the DDD dataset.

Among them, DDD can be MentalManip_con, ReaMent_con, or LegalCon. TTT can be correct, incorrect, or feedback.