Introduction

This repository implements the C-LSTM for text classification(SST-2) in Tensorflow2 and tested on binary classification and 5-class classification (SST-5) on Stanford Sentiment Treebank dataset.

Contribution

1. Implement with Tensorflow2 and Python3
2. Use pretrained Google-News-300 weights on the Embedding layer
3. Achieved performance very close to the original paper on SST dataset

Install

conda env create -f environment.yml
conda activate tensorflow

Train

Binary Classification

python train.py --dataset SST-2\
                --batch_size 48\
                --num_epochs 80\
                --num_class 2\
                --max_len 48\
                --learning_rate 1e-4

5-class classification

python train.py --dataset SST-5\
                --batch_size 48\
                --num_epochs 80\
                --num_class 5\
                --max_len 48\
                --learning_rate 1e-4

Result

Dataset	Accuracy of this implementation	Accuracy of original paper
SST-2	86.5%	87.8%
SST-5	48.6%	49.2%

Acknowledgement

• The SST-2 dataset is from clairett.
• The SST-5 dataset is from prrao87