A Question Answering (QA) model to predict answers for given pairs of context paragraph and question.
-
Colab Notebook
The notebook serves to train a Question Answering (QA) model to predict answers for given pairs of context paragraph and question using the The Standford Question Answering Dataset (SQuAD). The answers are confined to be either a span of text in the context paragraph or no answer, in case the model classifies the question to be unanswerable. The QA model trained is Bidirectional Encoder Representations from Transformers (BERT) (Devlin et al., 2018) with an additional Gating mechanism proposed by Xue & Li (2018) in the classification layer. The codes are modified based on Google's TensorFlow code and pre-trained models for BERT. -
This repository also comes with my academic paper and presentation of findings throughout the project in my part-time study in Master of Statistics in the University of Hong Kong.