Naive_SQuAD

This repo contains the data pre-processing and model training code for the following two methods:

Naive DrQA

We utilize the Document Reader architecture in DrQA to predict answer text span in the context for the given questions. But for the context encoding, we only use fixed word embedding and aligned question embedding as inputs.

Finetune BERT-base

We add a linear classifer on top of BERT output and finetune them together on the given question answering task. The BERT model is Huggingface Transformer's BERT-base-uncased.