Just a Scratch: Enhancing LLM Capabilities for Self-harm Detection through Intent Differentiation and Emoji Interpretation

2025-06-05

AI Summary

Detecting covert self-harm expressions in social media is highly challenging due to semantic ambiguity and strong contextual dependence, rendering existing large language models (LLMs) prone to misinterpreting intent and failing to provide interpretable justifications. To address this, we propose an intention-decoupled multi-task fine-tuning framework. Our approach introduces two novel resources: the Centennial Emoji Sensitivity Matrix (CESM-100), quantifying emoji-level affective sensitivity, and SHINES, a multi-level annotated dataset for self-harm intent and rationale. We integrate emoji-aware input enhancement, span-level intention classification, and rationale generation into a three-stage fine-tuning pipeline applied to Llama 3, Mental-Alpaca, and MentalLlama. Experiments demonstrate significant improvements in F1-score for self-harm detection and rationale consistency across zero-shot, few-shot, and full fine-tuning settings. All resources—including SHINES, CESM-100, and implementation code—are publicly released.

Technology Category

Machine Learning: Large Multimodal Models (LMMs)Natural Language Processing: Safety and RobustnessCognitive Modeling & Cognitive Systems: Affective Computing

Application Category

Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMsUser Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendationEconomics, Online Markets and Human Computation: Humans versus LLMs for data annotation and labeling

Abstract

Self-harm detection on social media is critical for early intervention and mental health support, yet remains challenging due to the subtle, context-dependent nature of such expressions. Identifying self-harm intent aids suicide prevention by enabling timely responses, but current large language models (LLMs) struggle to interpret implicit cues in casual language and emojis. This work enhances LLMs' comprehension of self-harm by distinguishing intent through nuanced language-emoji interplay. We present the Centennial Emoji Sensitivity Matrix (CESM-100), a curated set of 100 emojis with contextual self-harm interpretations and the Self-Harm Identification aNd intent Extraction with Supportive emoji sensitivity (SHINES) dataset, offering detailed annotations for self-harm labels, casual mentions (CMs), and serious intents (SIs). Our unified framework: a) enriches inputs using CESM-100; b) fine-tunes LLMs for multi-task learning: self-harm detection (primary) and CM/SI span detection (auxiliary); c) generates explainable rationales for self-harm predictions. We evaluate the framework on three state-of-the-art LLMs-Llama 3, Mental-Alpaca, and MentalLlama, across zero-shot, few-shot, and fine-tuned scenarios. By coupling intent differentiation with contextual cues, our approach commendably enhances LLM performance in both detection and explanation tasks, effectively addressing the inherent ambiguity in self-harm signals. The SHINES dataset, CESM-100 and codebase are publicly available at: https://www.iitp.ac.in/~ai-nlp-ml/resources.html#SHINES .

Problem

• Enhancing LLMs to detect self-harm intent in social media posts
• Improving emoji interpretation for context-dependent self-harm signals
• Addressing ambiguity in casual language for suicide prevention

Innovation

• Enhances LLMs with emoji sensitivity matrix
• Fine-tunes LLMs for multi-task learning
• Generates explainable self-harm prediction rationales