/SQuAD

To implement a state-of-the-art deep learning solution for SQuAD challenge, attempting to integrate statistical approaches for semantic inference.

Primary LanguagePythonMIT LicenseMIT

Hybrid learning for SQuAD Challenge

This project is inspired by [3] which accounts for the Machine Reading Comprehension task [1]. It is first a Tensorflow[2] implementation of [3], then a further exploration combing End-to-End MRC model with statistical inference model. The major intention of this project is to validate the power of integrating statistical inference into deep learning model on MRC task.

Methodology

Dependencies

  • Stanford CoreNLP[4]
    • Java 8
    • py-corenlp[5]
  • Stanford Glove[6]
  • Tensorflow[2] 1.4.0 for Python 3.6.3
    • CUDA 9

Implementaion

Experiment

Reference

[1] Stanford Question Answering Dataset (SQuAD)

[2] Tensorflow.

[3] R-NET: Machine Reading Comprehension with Self-matching Networks

[4] The Stanford CoreNLP Natural Language Processing Toolkit

[5] py-corenlp

[6] GloVe: Global Vectors for Word Representation