The Semi-Automatic Sentiment Ontology Building Using Word embeddings (SASOBUW) framework for developing (semi-automatic sentiment domain) ontologies for ABSA with the use of word embeddings.
- Optional synset extraction step
- 3 different hierarchy building methods
- Download WordNet and put it into SASOBUW\src\main\resources\externalData
- Download the DataSASOBUW.zip folder from https://1drv.ms/u/s!Ao_FuxrvQWwphb5r5w3_Jny0PbHhnw?e=bePwpB
- Put the contents of DataSASOBUW.zip into SASOBUW\src\main\resources\data
- SASOBUW package edu.eur.absa.Yelp
- YelpReader.java - a class that finds the relevant (restaurant-related) reviews from the Yelp dataset
- TxtForEmbeddings.java - a class which performs the necessary NLP steps (sentence splitting, tokenization, POS-tagging, lemmatization etc.) for the Yelp restaurant reviews and tips
- SASOBUW package edu.eur.absa.OntBuilding
- MainOntBuilder.java - a class that creates an ontology and where the main features and approaches for the building process can be chosen (e.g. GloVe or fastText embeddings, the hierarchy building method etc.)
- GridSearchParametersMain.java - a class where the parameter optimization can be done (e.g. the range and step size of the grid search can be chosen)
- OntHelper.java - a class with methods to find terms, create concepts, build a hierarchy etc.
- Ontology.java - a class with methods for the ontology (e.g. getting lexicalizations, adding classes or concepts etc.)