Hybrid Multi-stage Decoding for Few-shot NER with Entity-aware Contrastive Learning

Anonymous

Abstract: Few-shot named entity recognition can identify new types of named entities based on a few labeled examples. Previous methods employing token-level or span-level metric learning suffer from the computational burden and a large number of negative sample entities. In this paper, we propose the Hybrid Multi-stage Decoding for Few-shot NER with Entity-aware Contrastive Learning (MsFNER). Specifically, we first detect named entities without type and then classify entity types by the entity classification module. We divide MsFNER into 3 stages: training stage, finetuning stage, and inference stage. In the training stage, we employ a contrastive learning module to enhance entity representations and train the modules on the source domain. During finetuning, we finetune the model on the target support domain. In the inference stage, we replace the contrastive learning module with a KNN module and the final entity type inference is jointly determined by KNN and entity classification module. We conduct experiments on the open FewNERD dataset and the results demonstrate the advance of MsFNER.

Paper Type: short

Research Area: Information Extraction

Contribution Types: Model analysis & interpretability, Approaches to low-resource settings, Approaches low compute settings-efficiency

Languages Studied: English