Student exploration of Question Answering.
Our project aims to answer general knowledge questions by using Wikipedia data and the following BERT model pre-trained on SQuAD :
bert-large-uncased-whole-word-masking-finetuned-squad.
this model is available here : https://huggingface.co/bert-large-uncased-whole-word-masking-finetuned-squad
The pipeline can be summed-up in 3 essential steps :
- Subject extraction
- Wikipedia article retrieval
- Answer extraction
To run this project locally, you have download the WebApp file located in /Code/WebApp/
and run the file run_app.py with Python.
Our code is functional and we deployed a web app that can be accessed here : https://askme-app.azurewebsites.net/
The app can be launched locally with the following requirements :
flask==1.1.2
torch==1.6.0
transformers==3.3.1
scikit-learn==0.22.1
nltk==3.4.5
spacy==2.3.2
wikipedia==1.4.0
Wikipedia-API==0.5.4
langdetect==1.0.8
en_core_web_sm==2.3.1
The AzureAskme file with its Dockerfile can be used to create an image of our application. It will be ready to be deployed in a web service.
AzureAskme is available in /Results/
Pierre GONCALVES, Frederic ASSMUS, Axel DIDIER
M2 NLP - Université de Lorraine - 2020