LCR-Rot-hop-ont++

Source code for Injecting Knowledge from a Domain Sentiment Ontology in a Neural Approach for Aspect-Based Sentiment Classification.

Installation

Data

First, create a data/raw directory and download the SemEval 2015, SemEval 2016 datasets, and the ontology. Then rename the SemEval datasets to end up with the following files:

• data/raw
  • ABSA15_Restaurants_Test.xml
  • ABSA15_Restaurants_Train.xml
  • ABSA16_Restaurants_Test.xml
  • ABSA16_Restaurants_Train.xml
  • ontology.owl-Extended.owl

Setup environment

Create a conda environment with Python version 3.10, the required packages and their versions are listed in requirements.txt, note that you may need to install some packages using conda install instead of pip install depending on your platform.

Usage

To view the available cli args for a program, run python [FILE] --help. These CLI args can for example be used to pick the year of the dataset.

• main_preprocess.py: remove opinions that contain implicit targets and generate embeddings, these embeddings are used by the other programs. To generate all embeddings for a given year, run python main_preprocess.py --all
• main_hyperparam.py: run hyperparameter optimization
• main_train.py: train the model for a given set of hyperparameters
• main_validate.py: validate a trained model. To do an ablation experiment, run python main_validate.py --ablation, this requires all embeddings to be created for a given year.

Acknowledgements

The model.bert_encoder module uses code from:

• Liu, W., Zhou, P., Zhao, Z., Wang, Z., Ju, Q., Deng, H., Wang, P.: K-BERT: Enabling language representation with knowledge graph. In: 34th AAAI Conference on Artificial Intelligence. vol. 34, pp. 2901–2908. AAAI Press (2020)
• https://github.com/Felix0161/KnowledgeEnhancedABSA