SEPT: Improving Scientific Named Entity Recognition with Span Representation

TL;DR

SEPT is a novel scientific named entity recognition model that leverages span representation with simplified architecture and under-sampling, achieving state-of-the-art results without relying on relation information.

Contribution

The paper introduces SEPT, a simplified span extractor model with under-sampling that outperforms previous models in scientific NER without using relation data.

Findings

SEPT achieves new state-of-the-art performance in scientific NER.

Simplified architecture maintains high performance.

Under-sampling balances positive and negative samples effectively.

Abstract

We introduce a new scientific named entity recognizer called SEPT, which stands for Span Extractor with Pre-trained Transformers. In recent papers, span extractors have been demonstrated to be a powerful model compared with sequence labeling models. However, we discover that with the development of pre-trained language models, the performance of span extractors appears to become similar to sequence labeling models. To keep the advantages of span representation, we modified the model by under-sampling to balance the positive and negative samples and reduce the search space. Furthermore, we simplify the origin network architecture to combine the span extractor with BERT. Experiments demonstrate that even simplified architecture achieves the same performance and SEPT achieves a new state of the art result in scientific named entity recognition even without relation information involved.