Double-Checker: Large Language Model as a Checker for Few-shot Named Entity Recognition
Abstract: Recently, few-shot Named Entity Recognition (NER) has attracted significant attention due to the high cost of obtaining high-quality labeled data. Decomposition-based methods have demonstrated remarkable performance on this task, which initially train a type-independent span detector and subsequently classify the detected spans based on their types. However, this framework has an evident drawback as a domain-agnostic detector cannot ensure the identification of only those entity spans that are specific to the target domain. To address this issue, we propose Double-Checker, which leverages collaboration between Large Language Models (LLMs) and small models. Specifically, we employ LLMs to verify candidate spans predicted by the small model and eliminate any spans that fall outside the scope of the target domain. Extensive experiments validate the effectiveness of our method, consistently yielding improvements over two baseline approaches.
Paper Type: Short
Research Area: Information Extraction
Research Area Keywords: named entity recognition, few-shot extraction
Contribution Types: Approaches to low-resource settings
Languages Studied: English
Submission Number: 5351
Loading