Enhanced Local and Global Context Focus Mechanism Using BART Model for Aspect Based Sentiment Analysis of Social Media Data
Sunitha C. Suresh, K. Sharmila Banu · IEEE Access · 2024
Sentiment Analysis(SA) is an integral part of Data Analytics(DA) in today’s social-media-driven world. Being one of the most sought-after tasks in Natural Language Processing, SA has branched out to a different level since its conception. In the place where it started with character-level, document-level, and aspect-level SA, now there are more specific and fine-grained tasks like target-based sentiment analysis (TBSA), intent-based sentiment analysis (IBSA), aspect-based sentiment analysis(ABSA) and other variants of SA. People quickly express their emotions and opinions on social media, often focusing on every aspect rather than overall opinions about a product or event. Determining the overall sentiment may be quite difficult in such cases. Many models have already been introduced, focusing on local and global context vectors. But, addressing the unstructured nature of social media data is still a challenge. In this paper, we propose Enhanced Local Global Context Focus using BART (Enhanced LGCF-BART) model, designed to accomplish the task of ABSA. Various experiments and related results that validate our model performance are also given. We were able to test our Enhanced LGCF-BART model on the ACL 2014 Twitter Dataset and Laptop Dataset of SemEval14 to see how it performs ABSA with datasets that involve more unstructured or missing texts and those with an unbalanced class distribution, respectively.