Fine-Grained Entity Recognition via Large Language Models.

Area of Science:

• Natural Language Processing
• Information Extraction
• Artificial Intelligence

Background:

• Fine-grained entity recognition (FGER) is crucial for information extraction but hindered by a lack of domain-specific labeled data.
• Large language models (LLMs), like generative pretrained transformers (GPT), show potential for data-scarce FGER tasks.
• LLMs can exhibit 'hallucination' when processing extensive or complex input, impacting reliability.

Purpose of the Study:

• To propose a novel method, FGER-GPT, for effective fine-grained entity recognition in data-scarce domains.
• To address the hallucination issue in LLMs when applied to FGER.
• To develop an approach that bypasses the need for costly labeled data.

Main Methods:

• Leveraging multiple inference chains within the FGER-GPT framework.
• Implementing a hierarchical strategy for fine-grained entity recognition.
• Utilizing generative pretrained transformers (GPT) without labeled entity annotations.

Main Results:

• FGER-GPT demonstrates significant performance improvements in fine-grained entity recognition.
• The method achieves competitive results compared to state-of-the-art approaches in low-resource scenarios.
• The approach successfully mitigates LLM hallucination in the context of FGER.

Conclusions:

• FGER-GPT offers a viable solution for fine-grained entity recognition, especially in domains with limited labeled data.
• The method's ability to perform without annotations makes it practical for real-world applications.
• This work highlights the potential of LLMs for advancing information extraction tasks under resource constraints.