In this article, we will go through how to easily fine-tune any pretrained Natural Language Processing (=NLP) transformer for Named-Entity Recognition (=NER) in any language.
Why should you care? Well, NER is a powerful NLP task with many applications, as has been thoroughly described on Towards Data Science. However, effectively using NER often requires language or domain specific fine-tuning of your NER model based on the pretrained transformers that are available and realistic to use given your compute budget.
Many organizations have invested in developing services based on machine learning but are still fighting to wrap their heads around the best way to transition machine learning models from development to production.
As of today the approaches for this are often rigorous and imply all sorts of manual processes (with all of the operational risks such manual processes entail). This is hugely inefficient and counter-productive from both an organizational and model developer perspective.
When you develop a real-life machine learning service, you are in fact developing a piece of software. …
Data Scientist @ Ekstra Bladet, PIN. Author of a couple of R and Python Packages.