Recent Advancement of Emotion Cognition in Large Language Models
Yuyan Chen, Yanghua Xiao
TL;DR
The paper addresses how LLMs can model and utilize human emotions to improve performance on emotion-related tasks. It adopts Ulric Neisser’s cognitive stages to structure comparisons across sensation, perception, imagination, retention, recall, problem-solving, and thinking, framing emotion evaluation and emotion enhancement as core directions. It surveys methods including prompt engineering, embeddings-based fine-tuning, and knowledge augmentation, with results across emotion recognition, generation, and Theory of Mind, while noting limitations such as annotated-data dependence and explainability. It highlights future directions like unsupervised and contrastive learning to develop more robust, interpretable emotion-cognition LLMs with practical impact in education, mental health, and user interactions.
Abstract
Emotion cognition in large language models (LLMs) is crucial for enhancing performance across various applications, such as social media, human-computer interaction, and mental health assessment. We explore the current landscape of research, which primarily revolves around emotion classification, emotionally rich response generation, and Theory of Mind assessments, while acknowledge the challenges like dependency on annotated data and complexity in emotion processing. In this paper, we present a detailed survey of recent progress in LLMs for emotion cognition. We explore key research studies, methodologies, outcomes, and resources, aligning them with Ulric Neisser's cognitive stages. Additionally, we outline potential future directions for research in this evolving field, including unsupervised learning approaches and the development of more complex and interpretable emotion cognition LLMs. We also discuss advanced methods such as contrastive learning used to improve LLMs' emotion cognition capabilities.