Learning Explicit and Implicit Structures for Targeted Sentiment Analysis

This page contains the code used in the work "Learning Explicit and Implicit Structures for Targeted Sentiment Analysis" published at EMNLP 2019.

Contents

1. Usage
2. SourceCode
3. Data
4. Citation
5. Credits

Usage

Prerequisite: Python (3.5 or later), PyTorch (1.0 or later)

Run the following command to try out our model in the paper.

English

./exp_en.sh

Spanish

./exp_es.sh

After the training is complete, type the following command to display the result on test data. The performance outputed by conlleval.pl is shown as below.

tail -n 50 2019_en_ei.log
tail -n 50 2019_es_ei.log

SourceCode

The source code is written in PyTorch.

Data

The data is stored in "data/ts" folder for English and Spanish.

The embedding files need to be stored in the folder "embedding".

Download English Embedding (Glove 100) [https://drive.google.com/file/d/18ivfhXW0GpsBnDwvao-XNfm0P0d9592b/view?usp=sharing]

Download Spanish Embedding [https://drive.google.com/file/d/1uYCPv3_-ZwcYJ2Ga8DcmPoxi4SgIeLX2/view?usp=sharing]

Citation

If you use this software for research, please cite our paper as follows:

@InProceedings{learning19li, 
author = "Li, Hao and Lu, Wei", 
title = "Learning Explicit and Implicit Structures for Targeted Sentiment Analysis", 
booktitle = "Proc. of EMNLP2019", 
year = "2019", 
}

Credits

The code in this repository are based on Neural StatNLP framework: https://github.com/sutd-statnlp/statnlp-neural

Email to [email protected] if any inquery.