A tensorflow implementation of Convolutional Neural Networks for Sentence Classification. The original paper can be found at https://arxiv.org/abs/1408.5882
The implementation differs from the original paper in the following ways :
- Pretrained word vectors are not used
- Two seperate channels of word vectors are not used, all vectors are learned
- L2 norm constraints are not enforced
Edit the config file to change the configuration of the model
The model produces a test accuracy of 75.33 % within 2 epochs. The results produced in the paper for the given architecture is 76.1 %
Comparison with other models (taken from the paper). The model could perform much better with pretrained word vectors
| Architecture | Test accuracy |
|---|---|
| CNN-rand | 76.1 |
| CNN-static | 81.0 |
| CNN-non-static | 81.5 |
| CNN-multichannel | 81.1 |
| RAE (Socher et al., 2011) | 77.7 |
| MV-RNN (Socher et al., 2012) | 79.0 |
| CCAE (Hermann and Blunsom, 2013) | 77.8 |
| Sent-Parser (Dong et al., 2014) | 79.5 |
| NBSVM (Wang and Manning, 2012) | 79.4 |
| MNB (Wang and Manning, 2012) | 79.0 |
| G-Dropout (Wang and Manning, 2013) | 79.0 |
| F-Dropout (Wang and Manning, 2013) | 79.1 |
| Tree-CRF (Nakagawa et al., 2010) | 77.3 |