Multimodal-CNNs

Code of Text Classification with Topic-based Word Embedding and Convolutional Neural Networks (ACM BCB 2016).

In this paper, we propose a novel neural language model, Topic-based Skip-gram, to learn topic-based word embeddings for text classification with CNNs.

Please cite our paper if it helps your research:

@inproceedings{xu2016text,
  title={Text Classification with Topic-based Word Embedding and Convolutional Neural Networks.},
  author={Xu, Haotian and Dong, Ming and Zhu, Dongxiao and Kotov, Alexander and Carcone, April Idalski and Naar-King, Sylvie},
  booktitle={The 7th ACM Conference on Bioinformatics, Computational Biology, and Health Informatics (ACM BCB)},
  pages={88--97},
  year={2016}
}

Requirements

• Python==2.7
• Theano==0.7
• Lasagne==0.1
• Pre-trained word2vec vectors (CNN-gn in the paper) and Skip-gram model are available here.
• Latent Dirichlet Allocation model can be downloaded here.

Preprocessing

To process raw textual data,

python process_data_channel.py /path/to/wordvec1 /path/to/wordvec2 LengthofWordvec1 LengthofWordvec2 classnum /path/to/data_class0 /path/to/data_class1  

OR

python process_data_multi.py /path/to/wordvec1 /path/to/wordvec2 LengthofWordvec1 LengthofWordvec2 classnum /path/to/data_class0 /path/to/data_class1

Running CNN-channel or CNN-concat models

python conv_channel_w2v.py epochnum batchsize

OR

python conv_multi_w2v.py epochnum batchsize

You can choose epochnum=25 and batchsize=64.

Acknowledgment

The preprocessing code is adapted from Dr. Kim's work.