The paper titled “An Appraisal-Based Chain-Of-Emotion Architecture for Affective Language Model Game Agents,” authored by Maximilian Croissant and colleagues, explores the development of digital agents that convincingly imitate human emotions. The study addresses challenges related to creating believable and interactive agents, emphasizing the potential of large language models (LLMs) to enhance emotional intelligence simulations through situational appraisal patterns. Through three empirical experiments, the research examines LLMs’ effectiveness in emotional tasks and introduces a novel chain-of-emotion architecture grounded in psychological appraisal theory. The findings indicate that this new architecture significantly outperforms standard LLM frameworks across various user experience and content analysis metrics. Overall, the study lays the groundwork for constructing affective agents that integrate cognitive processes as represented in language models, contributing to the advancement of emotionally intelligent digital agents in gaming and beyond.
Source link
