Named Entity Recognition with Flair-based Embeddings

Overview

Flair provides a development framework for Natural Language Processing (NLP) that can be combined with Label Studio. Together, the integration allows users the ability to train a custom-named entity recognition (NER) model using the high-performance embeddings of the Flair library.

After connecting the Flair backend to Label Studio, annotators can conduct active learning by sending annotations to update the Flair model. Predictions can then be made against new data, including an additional filter to account for when two named entities are predicted in a sentence.

Benefits

• Accuracy: Flair is a high-quality NLP library supporting disambiguation and classification.
• Multilingual: Flair supports a rapidly growing number of languages, allowing for increased support.
• Adaptable: Flair builds directly on PyTorch, making it easy to train new models and experiment with new approaches using Flair embeddings and classes.