Information retrieval on SQuAD

IR + QA system on Wikipedia articles

This repository holds code for the Information Retrieval part of our project work for NLP2020 class by Paolo Torroni @unibo.

We tackle the IR problem with a classic tf-idf approach and with a contrastive Bi-Encoder model based on ELECTRA.
An in depth description of the work can be found here: QA-IR-report.pdf.

Experiment plots

https://wandb.ai/veri/IR/reports/IR--Vmlldzo1Mzk3MDc

Branches

• main: merged from the tfidf branch
• tfidf: tf-idf performances on SQuAD v1.1
• electra: neural model performances on SQuAD v1.1
• deploy: contains the end to end notebook that performs the IR + QA task and other files used to deploy the system, where the QA model is trained on SQuAD v2