A convolutional stacked bidirectional LSTM with a multiplicative attention mechanism for aspect category and sentiment detection

• Main Authors: Kumar, Ashok J., Trueman, Tina Esther, Cambria, Erik
• Other Authors: School of Computer Science and Engineering
• Format: Article
• Language: English
• Published: 2022
• Subjects: Engineering::Computer science and engineering; Aspect-Based Sentiment Analysis; Attention Mechanism
• Online Access: https://hdl.handle.net/10356/160176
• Institution: Nanyang Technological University

Traditionally, sentiment analysis is a binary classification task that aims to categorize a piece of text as positive or negative. This approach, however, can be too simplistic when the text under scrutiny contains more than one opinion target. Hence, aspect-based sentiment analysis provides fine-grained sentiment understanding of the product, service, or policy. Machine learning and deep learning algorithms play an important role in this kind of task. Also, attention mechanism has shown breakthrough in the field of natural language processing. Therefore, we propose a convolutional stacked bidirectional long short-term memory with a multiplicative attention mechanism for aspect category and sentiment polarity detection. More specifically, we treat the proposed model as a multiclass classification problem. The proposed model is evaluated using SemEval-2015 and SemEval-2016 dataset. Our proposed model outperforms state-of-the-art results in aspect-based sentiment analysis.