Question answering

Motivation

The purpose of this repository is to explore question answering model in deep learning. We want to build a benchmark for question answering tasks by using the tensorflow API Dataset and Estimator which can help us import the dataset in parallel.

Models

• Deep Learning for Answer Sentence Selection

  • unigram
  • unigram+count

• Learning to Rank Short Text Pairs with Convolutional Deep Neural Networks

  • cnn
  • cnn+interaction

• Attentive Pooling Networks

• Inner Attention based Recurrent Neural Networks for Answer Selection

Requirements

• python3

• Tensorflow = 1.8

• jieba

• spacy

Processing

./run.py --task_type prepare

training

./run.py --task_type train

test

./run.py --task_type infer

Contributor

• @ZhanSu
• @Wabywang

Performance

• unigram

  data	map	mrr
  trec	0.62	0.67
  trec-all	0.658	0.67

• unigram + count

  data	map	mrr
  trec	0.72	0.76
  trec-all	0.73	0.77

• cnn

  data	map	mrr
  trec	0.75	0.79
  trec-all	0.73	0.78

##Model Detail

unigram

---