CharSCNN

About

This is implementation of CharSCNN and SCNN[Ćıcero Nogueira dos Santos 2014].
This repository is heavily based on satwantrana/CharSCNN.
Experimental results are shown in this Japanese post.

Requirement

• Python 2 >= 2.6
• NumPy >= 1.7.1
• SciPy >= 0.11
• Theano == 0.9.0
• scikit-learn == 0.17
  • for dividing dataset into training set and test set

Usage

git clone https://github.com/hogefugabar/CharSCNN-theano

Download this tweets file from satwantrana/CharSCNN and add to cloned repository directory.
e.g.,

cd CharSCNN-theano
wget https://raw.githubusercontent.com/satwantrana/CharSCNN/master/tweets_clean.txt

To run SCNN;

python scnn.py

To run CharSCNNl;

python charscnn.py

Note: default dataset size (i.e., num_sent) is set to 20,000 in load.py or char_load.py

Reference

Ćıcero Nogueira dos Santos and Máıra Gatti. 2014. Deep convolutional neural networks for sentiment analysis of short texts. In Proceedings of the 25th International Conference on Computational Linguistics (COLING), Dublin, Ireland.