Reproduce “Fine-Grained Emotion Recognition via In-Context Learning” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Zhaochun Ren, Zhou Yang, Chenglong Ye, Haizhou Sun, Chao Chen, Xiaofei Zhu, Xiangwen Liao
Researchers at Leiden University, Fuzhou University, and SmartMore Company investigated the decision-making mechanism of Large Language Models (LLMs) during In-Context Learning (ICL) for fine-grained emotion recognition, proposing it aligns with prototype theory. They developed Emotion In-Context Learning (EICL), a method that consistently and significantly improves emotion recognition accuracy and F1 score across various LLMs and datasets compared to traditional ICL and zero-shot baselines by integrating emotionally similar examples, dynamic soft labels, and a two-stage exclusion strategy.
Key insights
- ICL's decision-making mechanism for emotion recognition aligns with prototype theory, where LLMs classify emotions by matching query representations to internal emotional prototypes.
- Traditional ICL's reliance on purely semantic similarity can introduce
- emotional discrepancies
- in examples, leading to flawed query representations and inaccurate emotional classifications.
- Emotionally-aware examples, dynamic soft labeling, and a structured two-stage decision process collectively enhance both the reasoning and decision-making phases of ICL for nuanced emotion tasks.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.