Reproduce “MetricPrompt: Prompting Model as a Relevance Metric for Few-shot Text Classification” 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
Hongyuan Dong, Weinan Zhang, Wanxiang Che
Researchers from the Harbin Institute of Technology propose MetricPrompt, a method that reframes few-shot text classification as a text pair relevance estimation task. This approach eliminates the need for task-specific verbalizers and consistently outperforms automatic verbalizer design methods while achieving competitive accuracy with manually crafted verbalizers on benchmark datasets.
Key insights
- Reformulating text classification as a text pair relevance estimation task eliminates the need for complex, task-specific verbalizer designs.
- This relevance estimation approach aligns better with pre-trained language models' (PLMs) original pre-training objectives, leading to more effective knowledge transfer.
- Aggregating relevance scores from multiple paired examples (e.g., via sum pooling) during inference provides robust and accurate classification for query texts.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.