Transformation Networks (TensorFlow)

This repo is the implementation of Transformation Networks for Target-Oriented Sentiment Classification (Xin Li, Lidong Bing, Wai Lam, Bei Shi, 2018) which is accepted by ACL 2018.

Before training the model, you need to get the embedding and dataset.

• Glove embedding: https://nlp.stanford.edu/projects/glove/ (glove.840B.300d)
• Dataset: https://github.com/lixin4ever/TNet

or you can use prepare_data.sh to download the laptop dataset and the embedding automately.

. ./prepare_data.sh

Changelog

• add the prepare_data.sh script.
• fix the performance problem (forget shuffle on each epoch.)

Requirements

• Python 3.6.x
• tensorflow 1.12.0
• numpy 1.15.3
• tqdm 4.23.4

pip install -r requirements.txt

Train Model

python train.py

Training Monitor

tensorboard --logdir=log --host=<host> --port=<port>

Reference

• Transformation Networks for Target-Oriented Sentiment Classification