Optimizing Aspect Term Extraction and Sentiment Classification through Attention Mechanism and Sparse Attention Techniques

Aspect-based sentiment analysis (ABSA) has become an essential field in Natural Language Processing (NLP) in recent years.ABSA not only categorizes sentiment as positive, negative, or neutral but also understands the specific aspects or topics discussed in the text review.This study focuses on two important elements of ABSA: Aspect Term Extraction (ATE) and Aspect Sentiment Classification (ASC).This study uses a combination of Sentence Embedding (SBERT) techniques, Part-of-Speech tagging, cosine-similarity calculation to assess words with their respective aspect labels, and the sparse attention mechanism (BIGBIRD) method, which has been proven to increase accuracy effectively and is effective in terms of time and memory usage.By applying this method to two hotel review datasets, Traveloka Review and Semeval 2016 dataset, it is proven to work well on two ABSA tasks, namely ATE and ASC.The results of the ATE test obtained an accuracy of 0.99, and the ASC test obtained an accuracy of 0.89.This study contributes to the advancement of ABSA by introducing a new methodology that improves the accuracy of aspect term extraction and sentiment classification.Additionally, it identifies avenues for future research, including exploring additional techniques to improve model performance and address potential limitations.