.png){kind=icon}
.png){kind=icon}
Return
A novel large-language-model-driven framework for named entity recognition
.png){kind=icon}
PRE
AI
DOI：10.1016/j.ipm.2024.104054.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Named entity recognition (NER) stands as the foundational pillar of knowledge graphs across multiple domains. Despite progress in NER using large language models (LLMs), challenges persist regarding the selection of LLMs, the retrieval of demonstrations, and the design of prompts. We introduce a novel framework for NER, termed LLMCC, which elucidates the synergistic interactions between different LLMs. Two new methods, SemnRank and InforLawthought, are proposed to address the issue of redundancy in demonstrations and to elevate prompt quality for boosting LLM's capabilities. Furthermore, LLMCC is trained through a new entity-aware contrastive learning. Extensive experiments across five domains confirm the competitiveness of LLMCC (surpassing ten recent studies by a margin of over 5% in F1 score), as well as the effectiveness of SemnRank and InforLaw-thought. We uncover a series of insights regarding information laws, prompting strategies, demonstration selections, and training designs. This research significantly advances the incorporation of LLMs into the construction of knowledge graphs.
Keywords:
Large language model
Named entity recognition
In-context learning
Contrastive learning
Knowledge graph
Journal
I
IF:
6.9
Papers:
5.2K
Citations:
1.4W