MentalMAC: Enhancing Large Language Models for Detecting Mental Manipulation via Multi-Task Anti-Curriculum Distillation

2025-05-21

AI Summary

Detecting psychological manipulation in multi-turn dialogues remains challenging due to its high concealment and severe scarcity of annotated data. To address these issues, this paper proposes a multi-task anti-curriculum distillation framework. It introduces, for the first time, an “easy-to-hard” anti-curriculum learning paradigm—reversing conventional curriculum design—integrated with evolutionary semantic augmentation (EvoSA) and speech-act-theory-guided unsupervised sample construction, alongside multi-task supervised learning and human-in-the-loop annotation. Evaluated on ReaMent, a high-quality, real-world dialogue dataset (5,000 samples) curated by the authors, the framework significantly narrows the performance gap between teacher and student models, outperforming state-of-the-art large language models across F1-score, accuracy, and other key metrics. Our core contributions are threefold: (1) pioneering the application of anti-curriculum distillation to psychological manipulation detection; (2) establishing a theory-grounded, data-efficient, and scalable lightweight detection paradigm; and (3) releasing a benchmark dataset enabling reproducible research.

Technology Category

• Machine Learning: Adversarial Learning & Robustness
• Intelligent Robots: Manipulation
• Data Mining & Knowledge Management: Conversational Systems for Recommendation & Retrieval

Application Category

• Economics, Online Markets and Human Computation: Humans versus LLMs for data annotation and labeling
• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Mental manipulation is a subtle yet pervasive form of psychological abuse that poses serious threats to mental health. Its covert nature and the complexity of manipulation strategies make it challenging to detect, even for state-of-the-art large language models (LLMs). This concealment also hinders the manual collection of large-scale, high-quality annotations essential for training effective models. Although recent efforts have sought to improve LLM's performance on this task, progress remains limited due to the scarcity of real-world annotated datasets. To address these challenges, we propose MentalMAC, a multi-task anti-curriculum distillation method that enhances LLMs' ability to detect mental manipulation in multi-turn dialogue. Our approach includes: (i) EvoSA, an unsupervised data expansion method based on evolutionary operations and speech act theory; (ii) teacher-model-generated multi-task supervision; and (iii) progressive knowledge distillation from complex to simpler tasks. We then constructed the ReaMent dataset with 5,000 real-world dialogue samples, using a MentalMAC-distilled model to assist human annotation. Vast experiments demonstrate that our method significantly narrows the gap between student and teacher models and outperforms competitive LLMs across key evaluation metrics. All code, datasets, and checkpoints will be released upon paper acceptance. Warning: This paper contains content that may be offensive to readers.

Problem

• Detecting subtle mental manipulation in dialogues using LLMs
• Overcoming scarcity of annotated datasets for manipulation detection
• Enhancing LLM performance via multi-task anti-curriculum distillation

Innovation

• Unsupervised data expansion via EvoSA method
• Multi-task supervision from teacher models
• Anti-curriculum distillation from complex to simple