This code is the program (RCNN) of the paper "Answer Sequence Learning with Neural Networks for Answer Selection in Community Question Answering" ACL-IJCNLP 2015: http://www.aclweb.org/anthology/P15-2117
%python Theano
The training and test for RCNN model: python model_train.py python model_test.py
The corresponding discription of each .py:
model_train.py - Model training (the model saved in ./WeightModel) model_test.py - Model test (the result saved in ./Result)
BidirectLstm.py - The Class of Bi-RNNs LeNetConvPoolLayer.py - The Class of CNN mlp.py - The Class of Hidden layer (fully connect layer) ReLU.py - The activation function ReLu (Rectified Linear Units)
parameters.py - Intilization of model parameters prepare.py - The Class of the preprocessing of QA data
Model Evaluation:
1 get-label.py - convert the result of RCNN model to the answer label for evaluation (saved in ./Label/)
2 CQA-QL-test-gold - the gold label of test data
3 scoreA.pl - the tool of computing the performance (provided by the answer selection task of semevel2015)
The Initialization of wordembedding is trained with the raw data in http://alt.qcri.org/semeval2015/task3/.