Aspect-based Sentiment Analysis via Knowledge Enhancement

Aspect-based sentiment Analysis (ABSA), a crucial task in Natural Language Processing (NLP), delves into texts to extract nuanced and detailed sentiment insights, focusing particularly on emotional expressions pertaining to specific aspects. While traditional methods have shown commendable results by employing deep learning techniques for semantic learning and incorporating syntactic information, this study introduces a novel knowledge fusion model. This model integrates resources from Wiktionary and SenticNet, enhancing the accuracy in determining sentiment polarity related to specific text aspects. Our approach, characterized by a unique structural integration, leverages entity definitions from Wiktionary and the sentiment lexicon from SenticNet as external knowledge sources. These are seamlessly incorporated into semantic and syntactic models, substantially improving the model’s capability to discern sentiment polarity. Empirical evaluations across various datasets have demonstrated notable improvements in both accuracy and F1 scores, underscoring the efficacy of amalgamating diverse knowledge sources to augment ABSA performance.