Using Active Learning to Expand Training Data for Implicit Discourse Relation Recognition

It is slightly simplified implementation of our Using Active Learning to Expand Training Data for Implicit Discourse Relation Recognition paper in Tensorflow.

Requirements

Python 3.5
Tensorflow 1.4
Numpy
sklearn

Preprocess:

training a basic model: python train.py 
Run by excuting the command of train.py. The default parameters are used.

active learning step: python active_learning_2.py

Data sets:

PDTB2.0
Sec 00~01 implicit: dev set
Sec 02~20 implicit: train set
Sec 21~22 implicit: test set
Sec 00~24 explicit: unlabeled set