Daily Arxiv

This is a page that curates AI-related papers published worldwide.
All content here is summarized using Google Gemini and operated on a non-profit basis.
Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.

EmoBench-M: Benchmarking Emotional Intelligence for Multimodal Large Language Models

Created by
  • Haebom

Author

He Hu, Yucheng Zhou, Lianzhong You, Hongbo Xu, Qianning Wang, Zheng Lian, Fei Richard Yu, Fei Ma, Laizhong Cui

Outline

This paper presents EmoBench-M, a new benchmark for evaluating the emotional intelligence (EI) of multimodal large-scale language models (MLLMs). To overcome the limitations of existing benchmarks, we evaluate the EI capabilities of MLLMs across three key dimensions: basic emotion recognition, conversational emotion understanding, and socially complex emotion analysis, across 13 scenarios, reflecting the multimodal complexity and dynamics of real-world interactions. EmoBench-M evaluations of open-source and closed-source MLLMs reveal significant performance gaps between MLLMs and humans, highlighting the need for further improvement in EI capabilities. All benchmark data are publicly available.

Takeaways, Limitations

•
Takeaways:
◦
Introducing EmoBench-M, a new benchmark for assessing emotional intelligence (EI) in MLLMs.
◦
Evaluation that considers multimodal and dynamic emotional expressions in real-world situations
◦
Raising the need to improve EI capabilities of MLLMs and providing publicly available benchmark data.
◦
Analysis of performance differences between open-source and closed-source MLLMs
•
Limitations:
◦
The 13 scenarios in EmoBench-M may not cover all real-world emotional situations.
◦
There is a possibility that the benchmark results may not fully reflect the overall level of MLLMs' EI capabilities.
◦
Difficulty in generalizing due to datasets biased towards specific languages or cultural backgrounds.
👍
Made with Slashpage