ProtoNER

Few-shot classification in Named Entity Recognition Task

---

This is the code accompanying the paper "Few-shot classification in Named Entity Recognition Task" by Alexander Fritzler, Varvara Logacheva and Maksim Kretov.

Please make sure you read the paper before you try to use or analyse this code. This repository contains 4 folders that allow you to run a specific model using Allennlp (v0.6.0) framework.

1. simple_base

This is a model that trains on a small dataset from scratch.

2. warming

This model trains on big dataset that doesn't contain objects from the validation class.

3. warm_base

This model loads from initialization generated by the previous warming model and trains as "simple_base" model.

4. warm_protonet

This model loads from initialization generated by the warming model and trains by a procedure described in the paper

To launch the model run from the corresponding folder:

python global_experiment.py

If you have a question about this code, please contact me by [email protected]