Improving LLM emotional benchmark performance with ECOT

Several benchmarks have been developed to measure emotional intelligence (EI) in LLMs’ outputs (Gao et al., 2025; Lian et al., 2024; Liu et al., 2025; Paech, 2024). However, while LLMs have achieved high performance in various problem-solving tasks, EI seems to remain an underdeveloped and understudied ability, and significantly inferior to that of humans (Sabour et al., 2024). The present project aims to improve LLM’s EI abilities by applying “Emotional Chain of Thought (ECoT)” (Li et al., 2024). ECOT prompts are characterized by step-by-step instructions highlighting specific aspects of emotions and breaking down their understanding and application into concrete components. They aim to guide the LLM’s reasoning and thereby yield more accurate, considerate results. Several LLMs will be selected and prompted, and their outputs will be evaluated with the Emo-Bench benchmark (Sabour et al., 2024) in order to assess whether certain prompts lead to significant performance differences.