Character-level-CNN

Implementation of Character level CNN

Getting started

This repository consists of code for the implementation of Character Level CNN's proposed by Xiang Zhang,Junbo Zhao,Yann LeCun https://papers.nips.cc/paper/5782-character-level-convolutional-networks-for-text-classification.pdf. Character level CNN is useful when the word embeddings are not powerful enough to extract the relationship among words.In Character Level CNN a set of characters is considered as a dictionary each given an unique index.Each sentence is converted to a fixed size of 3D vector.Each character is one-hot encoded using the dictionary indices, if the character count is greater they are ignored. These are then trained using CNN.

Implementation

I have used a dictionary size of 26 and considered 50 as the maximum character size for each sentence.Dataset considered is rt-polarity 2.0 dataset which consists of 5331 positive reviews and 5331 negative reviews.

Usage

1. Install dependencies using

pip install -r requirements.txt

2. clone the repository using

git clone https://github.com/avinashsai/Character-level-CNN.git

3. Change the folder using

cd Models

4. Run the main file using

python character_level_cnn.py