Few-Shot Named Entity Recognition in Natural Language Processing
Decoding NLP Library — Concise Concepts
Named-entity recognition is the task of identifying key information from the text and tagging them under the pre-defined set of elaborate and meaningful categories for a given domain. The usual way of going about thinking of such systems is to train a many-to-many model under supervised setting, where the input is word sequence and output is tag sequence. The process starts with labelling large set of sentences under BIO scheme with relevant entities. The tagging process is a time-taking and expensive exercise. Recent adoption of zero-shot and few-shot learning paradigm in natural language processing has produced decent performing first cut models and also using them to bootstrap the labelling process with significantly less turnaround time and eventually training a high-performing supervised model.
In this blog, I’ll be going through this very interesting Python library called “Concise Concepts” that will help you build a named entity recognition system under a few-shot setting i.e having an only handful of examples to learn and extrapolate from.
Here is how easy it is to make it work —
