HAABSAStar

Code for "Adversarial Training for a Hybrid Approach to Aspect-Based Sentiment Analysis". This project builds on the code from https://github.com/ofwallaart/HAABSA and https://github.com/mtrusca/HAABSA_PLUS_PLUS.

All software is written in PYTHON3 (https://www.python.org/) and makes use of the TensorFlow framework (https://www.tensorflow.org/).

Installation Instructions (Windows):

Dowload required files and add them to data/externalData folder:

1. Download ontology: https://github.com/KSchouten/Heracles/tree/master/src/main/resources/externalData
2. Download SemEval2015 Datasets: http://alt.qcri.org/semeval2015/task12/index.php?id=data-and-tools
3. Download SemEval2016 Dataset: http://alt.qcri.org/semeval2016/task5/index.php?id=data-and-tools
4. Download Glove Embeddings: http://nlp.stanford.edu/data/glove.42B.300d.zip
5. Download Stanford CoreNLP parser: https://nlp.stanford.edu/software/stanford-parser-full-2018-02-27.zip
6. Download Stanford CoreNLP Language models: https://nlp.stanford.edu/software/stanford-english-corenlp-2018-02-27-models.jar

Setup Environment

1. Install chocolatey (a package manager for Windows): https://chocolatey.org/install
2. Open a command prompt.
3. Install python3 by running the following command: code(choco install python) (http://docs.python-guide.org/en/latest/starting/install3/win/).
4. Make sure that pip is installed and use pip to install the following packages: setuptools and virtualenv (http://docs.python-guide.org/en/latest/dev/virtualenvs/#virtualenvironments-ref).
5. Create a virtual environemnt in a desired location by running the following command: code(virtualenv ENV_NAME)
6. Direct to the virtual environment source directory.
7. Unzip the zip file of this GitHub repository in the virtual environment directrory.
8. Activate the virtual environment by the following command: 'code(Scripts\activate.bat)`.
9. Install the required packages from the requirements.txt file by running the following command: code(pip install -r requirements.txt).
10. Install the required space language pack by running the following command: code(python -m spacy download en)

Note: the files BERT768embedding2015.txt and BERT768embedding2016.txt are too large for GitHub. These can be generated using getBERTusingColab.py.

Configure paths

The following scripts contain file paths to adapt to your computer (this is done by adding the path to you virtual environment before the filename. For example "/path/to/venv"+"data/programGeneratedData/GloVetraindata"): main_cross.py, main_hyper.py, config.py, HyperDataMaker.py, adversarial.py.

Run Software

1. Configure one of the three main files to the required configuration (main.py, main_cross.py, main_hyper.py)
2. Run the program from the command line by the following command: code(python PROGRAM_TO_RUN.py) (where PROGRAM_TO_RUN is main/main_cross/main_hyper)

Software explanation:

The environment contains the following main files that can be run: main.py, main_cross.py, main_hyper.py

• main.py: program to run single in-sample and out-of-sample valdition runs. Each method can be activated by setting its corresponding boolean to True e.g. to run the Adversarial method set runAdversarial= True.

• main_cross.py: similar to main.py but runs a 10-fold cross validation procedure for each method.

• main_hyper.py: program that is able to do hyperparameter optimzation for a given space of hyperparamters for each method. To change a method change the objective and space parameters in the run_a_trial() function.

• config.py: contains parameter configurations that can be changed such as: dataset_year, batch_size, iterations.

• dataReader2016.py, loadData.py: files used to read in the raw data and transform them to the required formats to be used by one of the algorithms

• lcrModel.py: Tensorflow implementation for the LCR-Rot algorithm

• lcrModelAlt.py: Tensorflow implementation for the LCR-Rot-hop algorithm

• lcrModelInverse.py: Tensorflow implementation for the LCR-Rot-inv algorithm

• cabascModel.py: Tensorflow implementation for the CABASC algorithm

• OntologyReasoner.py: PYTHON implementation for the ontology reasoner

• svmModel.py: PYTHON implementation for a BoW model using a SVM.

• adversarial.py: Tensorflow implementation of adversarial training for LCR-Rot-hop

• att_layer.py, nn_layer.py, utils.py: programs that declare additional functions used by the machine learning algorithms.

Directory explanation:

The following directories are necessary for the virtual environment setup: __pycache, \Include, \Lib, \Scripts, \tcl, \venv

• cross_results_2015: Results for a k-fold cross validation process for the SemEval-2015 dataset
• cross_results_2016: Results for a k-fold cross validation process for the SemEval-2015 dataset
• Results_Run_Adversarial: If WriteFile = True, a csv with accuracies per iteration is saved here
• data:
  • externalData: Location for the external data required by the methods
  • programGeneratedData: Location for preprocessed data that is generated by the programs
• hyper_results: Contains the stored results for hyperparameter optimzation for each method
• results: temporary store location for the hyperopt package

Changed files with respect to https://github.com/mtrusca/HAABSA_PLUS_PLUS:

• main.py
• main_hyper.py
• main_cross.py
• config.py
• adversarial.py (added)