Cognition Chain for Explainable Psychological Stress Detection on Social Media

TL;DR

This work tackles the explainability gap in social-media-based stress detection by introducing Cognition Chain, a four-step reasoning framework guided by cognitive appraisal theory: Stimulus $\mathcal{S}$, Evaluation $\mathcal{E}$, Reaction $\mathcal{R}$, and Stress State $\mathcal{A}$. It formalizes this reasoning in a probabilistic chain and embeds it into a dedicated prompt template, CogInstruct, to generate high-quality, self-refining cognitive explanations. Using a three-stage self-reflection pipeline with GPT-4o and a human-filtered RoBERTa classifier, the authors create 4,282 instruction-tuning examples and instruction-tune Llama-3 (CogLLM) via LoRA. Experimental results show CogLLM achieves state-of-the-art performance among open-source models and superior explainability compared with baselines, highlighting the potential of integrating cognitive theory into LLM reasoning for trustworthy psychological detection.

Abstract

Stress is a pervasive global health issue that can lead to severe mental health problems. Early detection offers timely intervention and prevention of stress-related disorders. The current early detection models perform "black box" inference suffering from limited explainability and trust which blocks the real-world clinical application. Thanks to the generative properties introduced by the Large Language Models (LLMs), the decision and the prediction from such models are semi-interpretable through the corresponding description. However, the existing LLMs are mostly trained for general purposes without the guidance of psychological cognitive theory. To this end, we first highlight the importance of prior theory with the observation of performance boosted by the chain-of-thoughts tailored for stress detection. This method termed Cognition Chain explicates the generation of stress through a step-by-step cognitive perspective based on cognitive appraisal theory with a progress pipeline: Stimulus $\rightarrow$ Evaluation $\rightarrow$ Reaction $\rightarrow$ Stress State, guiding LLMs to provide comprehensive reasoning explanations. We further study the benefits brought by the proposed Cognition Chain format by utilising it as a synthetic dataset generation template for LLMs instruction-tuning and introduce CogInstruct, an instruction-tuning dataset for stress detection. This dataset is developed using a three-stage self-reflective annotation pipeline that enables LLMs to autonomously generate and refine instructional data. By instruction-tuning Llama3 with CogInstruct, we develop CogLLM, an explainable stress detection model. Evaluations demonstrate that CogLLM achieves outstanding performance while enhancing explainability. Our work contributes a novel approach by integrating cognitive theories into LLM reasoning processes, offering a promising direction for future explainable AI research.

Figures (6)

• Figure 1: An example of our Cognition Chain. The left part is an individual's post. The right part is our Cognition Chain.
• Figure 2: Illustration of our proposed prompt template for the Reasoning Chain. The detailed prompt template is presented in Appendix.
• Figure 3: Illustration of the constuction process for our CogInstruct.
• Figure 4: The word cloud of each Cognition Chain step in our CongInstruct dataset
• Figure 5: Human evaluation for explanations. The CO, DE, RE, LO, OV represent comprehension, depth, relevance, logic, overall, respectively.