This page contains the code used in the work "Learning Latent Sentiment Scopes for Entity-Level Sentiment Analysis" published at AAAI 2017.
Prerequisite: JRE (1.8 or later) and Ant
- Run "make" to compile the source code.
make- Run "make init" to initialize files and folders.
make init- Run "make " for experiments mentioned in the paper.
make EXPERIMENTNote that the EXPERIMENT refers to anyone of the following experiments mentioned in the paper.
sentimentscope_latent_english
sentimentscope_latent_nohiddeninfo_english
sentimentscope_latent_withpostag_english
sentimentscope_latent_wordembedding_english
sentimentscope_nonlatent_english
sentimentscope_semimarkov_nonlatent_english
sentimentscope_semimarkov_latent_english
baseline_collapse_english
baseline_pipeline_english
sentimentscope_latent_spanish
sentimentscope_latent_nohiddeninfo_spanish
sentimentscope_latent_withpostag_spanish
sentimentscope_latent_wordembedding_spanish
sentimentscope_nonlatent_spanish
sentimentscope_semimarkov_nonlatent_spanish
sentimentscope_semimarkov_latent_spanish
baseline_collapse_spanish
baseline_pipeline_spanish
After executing one command, the program will create a unique folder for storing experiment results. You can type "make check" to see the progress logs and evaluation results in the end. The experiment results store in the following path.
experiments/sentiment/model/<modelname>/<lang>
Notice that pipeline will involve two steps of training and evaluation automatically.
The source code is written in Java, which can be found under the "src" folder.
Note that this is a Maven project. Therefore you can import this project as a Maven project in eclipse.
If you use this software for research, please cite our paper as follows:
@inproceedings{li2017learning,
title={Learning Latent Sentiment Scopes for Entity-Level Sentiment Analysis.},
author={Li, Hao and Lu, Wei},
booktitle={AAAI},
pages={3482--3489},
year={2017}
}
Email to hao_li@mymail.sutd.edu.sg if any inquery.