PromptNER : Prompting For FewShot Named Entity Recognition
Primary Area: general machine learning (i.e., none of the above)
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Keywords: Named Entity Recognition, Large Language Models, Prompting, FewShot Learning, NER
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.
TL;DR: We introduce PromptNER, a new state-of-the-art algorithm which uses prompting with an LLM to achieve start-of-the-art performance on few-shot NER and cross-domain NER tasks.
Abstract: In a surprising turn, Large Language Models (LLMs), together with a growing arsenal of prompt-based heuristics, provide powerful few-shot solutions to myriad classic NLP problems. However, despite promising early results, current LLM based few-shot methods remain far from the state of the art in Named Entity Recognition (NER), where prevailing methods include learning representations via end-to-end structural understanding and fine-tuning on standard labeled corpora. In this paper, we introduce PromptNER, an algorithm for few-shot and cross-domain NER. To adapt to a new NER task, PromptNER requires a set of entity definitions, and a set of few-shot examples, along with explanatory text justifying the applicability of each entity tag. Given a sentence, PromptNER prompts an LLM to produce a list of potential entities along with corresponding explanations justifying their compatibility with the provided entity type definitions. PromptNER achieves state-of-the-art performance on few-shot NER, achieving improvements in F1 score (absolute) of 4% on the ConLL dataset, 9% on GENIA, 4% on FewNERD, 5% on FaBNER and 24% on TweetNER. PromptNER also achieves state-of-the-art performance on Cross Domain NER beating even methods not restricted to the few-shot setting on 3/5 CrossNER target domains, with an average F1 gain of 3%, despite using less than 2% of the available data.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 7538
Loading