Aspect-Category-Opinion-Sentiment Extraction Using Generative Transformer Model

Sentiment analysis is one of Natural Language Processing's applications that aims to process and extract sentiment information quickly and effectively. To expand upon the previous triplet extraction, that being aspect-opinion-sentiment triplets, Aspect-Category-Opinion-Sentiment (ACOS) quadruple extraction was created. There are several methods to extract quadruples, albeit with several limitations, such as their effectiveness towards implicit information and their overall low-performance score. This paper proposes a method of using BART-Aspect-Based-Sentiment-Analysis (BARTABSA), a sentiment analysis model that aims to unify the previous Aspect Based Sentiment Analysis subtask - namely Aspect-Opinion pair extraction and Aspect-Opinion-Sentiment triplet extraction, and solve them without changing the core algorithm or adding other models to it - to solve the ACOS subtask. After some modification to the data and the model's outer layer, the result shows significant and promising improvements over previous results.