Coreference Approach

We call coreference resolution service when the following criteria is met:

1. We already have a Key Topic in cache, which is a medical condition resolved previously in ER.
2. We are in a search intent
3. We do not resolve anything in ER for the current query
4. We found a relation in Coreference between current query and previous topic

If the above conditions are met, we send the user a message that Zana understands that current query is related to the previous topic and if we have an article for the topic and current intent, display article.

Library

We are using NeuralCoref library to resolve Coreference (Anaphora) Resolution

 Link in github: https://github.com/huggingface/neuralcoref/blob/master/readme.rst

To try it online use the following link: https://huggingface.co/coref/

Service is in the new server nlu-dev.zana.com, port 8000

Installation

git clone https://github.com/huggingface/neuralcoref.git

cd neuralcoref
pip install -e .

Download the model:

pip install https://github.com/huggingface/neuralcoref-models/releases/download/en_coref_lg-3.0.0/en_coref_lg-3.0.0.tar.gz --no-cache-dir

If everything goes fine you should see a message: Successfully installed en-coref-lg-3.0.0

Run the service:

cd examples

python server.py &

service in: http://nlu-dev.zana.com:8000/?text=