Advancements in Few-Shot Nested Named Entity Recognition: The Efficacy of Meta-Learning Convolutional Approaches

Anonymous

TL;DR: Meta-learning convolutional for Few-shot Nested named entity recognition

Abstract: Few-shot Named Entity Recognition (NER) involves the identification of new entities using a limited amount of labeled data, which may contain nested entities. Currently, mainstream fewshot NER methods are not designed to handle nested entities. This study introduces a novel span-based meta-learning framework that uses meta-learning convolution to address the challenges of few-shot nested NER. Our proposed method, called Meta-Learning Convolution for Few-Shot Nested NER (MCFSN), is the first to integrate meta-learning with convolutional neural networks, effectively handling nested entities with limited training examples. This study presents a two-stage processing approach: extracting span features using CNN combined with the Biaffine attention mechanism, followed by entity span classification utilizing ProtoNet and the Biaffine classifier. Our experiments demonstrate consistently superior performance across three diverse language datasets, outperforming several competing baseline models in terms of F1 scores. Specifically, our approach achieves 6.9% F1 score improvement on the Genia, 5.2% F1 value improvement on the GermEval, and 4.5% F1 value enhancement on the NEREL, thus validating the effectiveness of our proposed approach.

Paper Type: long

Research Area: Efficient/Low-Resource Methods for NLP

Contribution Types: Approaches to low-resource settings

Languages Studied: English,German,Russian