MentalManip: A Dataset For Fine-grained Analysis of Mental Manipulation in Conversations

TL;DR

A new dataset, named "$4,000$ annotated movie dialogues", is introduced, which enables a comprehensive analysis of mental manipulation, pinpointing both the techniques utilized for manipulation and the vulnerabilities targeted in victims.

Abstract

Mental manipulation, a significant form of abuse in interpersonal conversations, presents a challenge to identify due to its context-dependent and often subtle nature. The detection of manipulative language is essential for protecting potential victims, yet the field of Natural Language Processing (NLP) currently faces a scarcity of resources and research on this topic. Our study addresses this gap by introducing a new dataset, named ${\rm M{\small ental}M{\small anip}}$, which consists of $4,000$ annotated movie dialogues. This dataset enables a comprehensive analysis of mental manipulation, pinpointing both the techniques utilized for manipulation and the vulnerabilities targeted in victims. Our research further explores the effectiveness of leading-edge models in recognizing manipulative dialogue and its components through a series of experiments with various configurations. The results demonstrate that these models inadequately identify and categorize manipulative content. Attempts to improve their performance by fine-tuning with existing datasets on mental health and toxicity have not overcome these limitations. We anticipate that ${\rm M{\small ental}M{\small anip}}$ will stimulate further research, leading to progress in both understanding and mitigating the impact of mental manipulation in conversations.

Figures (13)

• Figure 1: An example dialogue that contains elements of mental manipulation which GPT-4 fails to identify ($Temperature = 0$). The manipulative parts are highlighted in red.
• Figure 2: Multi-level taxonomy of $\textsc{MentalManip}$.
• Figure 3: Statistics of $\textsc{MentalManip}_{\text{con}}$ and $\textsc{MentalManip}_{\text{maj}}$. The x-axis ticks in the left two panels are abbreviations for techniques and vulnerabilities (see Appendix \ref{['appendix:definition']}). The emotion distribution of $\textsc{MentalManip}_{\text{maj}}$ dataset is in Appendix \ref{['appendix:statistics_maj']}.
• Figure 4: Co-occurrence heat maps among techniques (left), vulnerabilities (center), and techniques and vulnerabilities (right) in $\textsc{MentalManip}_{\text{con}}$ dataset. Darker cell indicates a higher co-occurrence. The same figures showing results on $\textsc{MentalManip}_{\text{maj}}$ dataset are in Appendix \ref{['appendix:statistics_maj']}.
• Figure 5: t-SNE visualization of Sentence Transformer embeddings of manipulative and non-manipulative dialogues in $\textsc{MentalManip}_{\text{con}}$ (left) and the distribution of $\textsc{MentalManip}$ and other dialogical datasets (right).