Few-shot Learning for Named Entity Recognition in Medical Text

TL;DR

This paper explores methods to enhance medical text named entity recognition with only ten annotated examples, achieving significant performance improvements through five targeted strategies.

Contribution

It introduces and evaluates five specific techniques to improve few-shot NER performance in medical texts, demonstrating substantial F1 score gains.

Findings

F1 score improved from 69.3% to 78.87%

Layer-wise initialization and hyperparameter tuning are effective

Combining pre-training data and custom embeddings boosts accuracy

Abstract

Deep neural network models have recently achieved state-of-the-art performance gains in a variety of natural language processing (NLP) tasks (Young, Hazarika, Poria, & Cambria, 2017). However, these gains rely on the availability of large amounts of annotated examples, without which state-of-the-art performance is rarely achievable. This is especially inconvenient for the many NLP fields where annotated examples are scarce, such as medical text. To improve NLP models in this situation, we evaluate five improvements on named entity recognition (NER) tasks when only ten annotated examples are available: (1) layer-wise initialization with pre-trained weights, (2) hyperparameter tuning, (3) combining pre-training data, (4) custom word embeddings, and (5) optimizing out-of-vocabulary (OOV) words. Experimental results show that the F1 score of 69.3% achievable by state-of-the-art models can be improved to 78.87%.