zainabzf/deep-text-classification-pytorch

Deep neural network text classification papers in PyTorch

Deep Text Classification in PyTorch

PyTorch implementation of deep text classification models including:

• WordCNN : Convolutional Neural Networks for Sentence Classification
• CharCNN : Character-level Convolutional Networks for Text Classification
• VDCNN : Very Deep Convolutional Networks for Text Classification
• QRNN : Quasi-Recurrent Neural Networks

Requirements

• Python 3.5+
• PyTorch 0.3
• gensim 3.2
• tqdm
• requests

Usage

To begin, you will need to download datasets as follows:

$ python download_dataset.py all

You can also download a specific dataset by specifying its name instead of all. Available datasets are MR, SST-1, SST-2, ag_news, sogou_news, dbpedia, yelp_review_full, yelp_review_polarity, yahoo_answers, amazon_review_full, and amazon_review_polarity

To download word vectors, run the following:

$ python download_wordvector.py word2vec
$ python download_wordvector.py glove

WordCNN

To train WordCNN with rand mode:

$ python main.py --dataset MR WordCNN --mode rand --vector_size 128 --epochs 300

To train WordCNN with multichannel mode:

$ python main.py --dataset MR WordCNN --mode multichannel --wordvec_mode word2vec --epochs 300

Available modes are rand, static, non-static, and multichannel

CharCNN

To train CharCNN with small mode:

$ python main.py --dataset MR CharCNN --mode small --epochs 300

To train CharCNN with large mode:

$ python main.py --dataset MR CharCNN --mode large --epochs 300

VDCNN

To train VDCNN with depth = 29:

$ python main.py --dataset MR VDCNN --depth 29

QRNN

To train QRNN with four layers:

$ python main.py --dataset MR QRNN --wordvec_mode glove --num_layers 4 --epochs 300

TF-IDF (benchmark)

You can train a multinomial logistic regression with TF-IDF features as a benchmark.

$ python tf-idf.py --dataset MR

Help

Refer to python main.py --help and python main.py {WordCNN, CharCNN, VDCNN, QRNN} --help for full description of how to use.

Experiments

Results are reported as follows: Test accuracy reproduced here (Test accuracy reported by the paper)

To find the settings for experiments, refer to experiments.sh.

	MR	SST_1	SST_2	ag_news	sogu_news	db_pedia	yelp_review_full	yelp_review_polarity	yahoo_answer	amazon_review_full	amazon_review_polarity
WordCNN (rand)	69.4 (76.1)	(45.0)	(82.7)	88.3			92.5
WordCNN (static)	(81.0)	(45.5)	(86.8)
WordCNN (non-static)	(81.5)	(48.0)	(87.2)
WordCNN (multichannel)	(81.1)	(47.4)	(88.1)
CharCNN (small)
CharCNN (large)
VDCNN (29-layers)
QRNN (k=2)	(91.4)
QRNN (k=4)	(91.1)

References

• Shawn1993's CNNs for Sentence Classification in PyTorch
• ArdalanM's nlp-benchmarks
• salesforce's Quasi-Recurrent Neural Network (QRNN) for PyTorch