The implementation of our paper Modeling Sentence Pairs with Tree-structured Attentive Encoder(pdf) to appear at the COLING2016. It runs both training and evaluation. Note that we just test the projects on Mac OS X and Ubuntu 14.
You can download the preprocessed data (recommend) from here. Alternatively you can process them by yourself. The original links are:
The software can run on CPU or GPU, dependency requirements are following:
- torch7
- python
The package should be installed are the following. For example, you can install the nn package by luarocks install nn.
luarocks install nn
luarocks install nngraph
luarocks install penlight
luarocks install optim
luarocks install xlua
luarocks install sys
luarocks install cutorch
luarocks install cunnTo run our models, you can tap the command
th main.lua -<opt_name> opt_value -<opt_name> opt_val ...
For example, to run the model on the SICK dataset, you should tap
th main.lua -task MSRP -structure atreelstm -lr 0.05 -n_epoches 10
more details in the file main.lua.
We achieve the results as
@inproceedings{zhou2016sentpair,
title={Modelling Sentence Pairs with Tree-structured Attentive Encoder},
author={Yao Zhou, Cong Liu and Yan Pan},
booktitle={The International Conference on Computational Linguistics (COLING)},
year={2016}
}
MIT