RCNN for Text Classification in PyTorch

PyTorch implementation of "Recurrent Convolutional Neural Network for Text Classification (2015)"

Model

Requirements

PyTorch
sklearn
nltk
pandas

Dataset

AG NEWS Dataset [Download] : This link is from TORCHTEXT.DATASETS.

DATASET	COUNTS
TRAIN	110,000
VALID	10,000
TEST	7,600

Classes

Original classes are 1, 2, 3, 4 each, but changed them into 0, 1, 2, 3.

• 0: World

• 1: Sports

• 2: Business

• 3: Sci/Tech

Training

To train,

python main.py --epochs 10

To train and want to see test set result,

python main.py --epochs 10 --test_set

Result

For test set,

Accuracy	Precision	Recall	F1
91.5	0.9154	0.9150	0.9149

Confusion Matrix is like below,

[1712   47   63   78]
[  21 1852   18    9]
[  53   18 1660  169]
[  34   24  112 1730]

Reference

• Lai, S., Xu, L., Liu, K., & Zhao, J. (2015, February). Recurrent convolutional neural networks for text classification. In Twenty-ninth AAAI conference on artificial intelligence. [Paper]