Char-CNN-for-Text-Classification

Character-level Convolutional Networks for Text Classification

Introduction

This code mainly implement the paper Character-level Convolutional Networks for Text Classification (Char CNN)

I find this image rather difficult to understand. So I redrawn the following image according to my own understanding. In actually, Char CNN is similar with conventional CNN (for image).

Dataset

Please click HERE to download the 'train.csv' and 'test.csv', then put them into the folder 'dataset'.

The original dataset AG's corpus of news articles

Data sample:

class	text content
world	Sister of man who died in Vancouver police custody slams chief (Canadian Press). Canadian Press - VANCOUVER (CP) - The sister of a man who died after a violent confrontation with police has demanded the city's chief constable resign for defending the officer involved.
sports	Johnson Back to His Best as D-Backs End Streak. NEW YORK (Reuters) - Randy Johnson struck out 14 batters in 8 1/3 innings to help the Arizona Diamondbacks end a nine-game losing streak with a 2-0 win over the host New York Mets in the National League Sunday.
Business	Dollar Briefly Hits 4-Wk Low Vs Euro. LONDON (Reuters) - The dollar dipped to a four-week low against the euro on Monday before rising slightly on profit-taking, but steep oil prices and weak U.S. data continued to fan worries about the health of the world's largest economy.
Sci/Tech	Search providers seek video, find challenges. Internet search providers are reacting to users #39; rising interest in finding video content on the Web, while acknowledging that there are steep challenges that need to be overcome.

Requirements

1. python3.5
2. tensorflow1.4.0
3. numpy
4. pandas

Results

Loss	Training accuracy

The testing accuracy (paper: 84.35%): 84.95% (trained about 25 000 iterations)

Acknowledgement

Thanks for mhjabreel' csv dataset

Reference

[1] Zhang X, Zhao J, LeCun Y. Character-level convolutional networks for text classification[C]//Advances in neural information processing systems. 2015: 649-657.