Vietnamese Aspect-Based Sentiment Analysis

The Vietnamese Aspect-Based Sentiment Analysis (VABSa) aims to identify and extract specific aspects of opinions and sentiment tendencies from text to evaluate users' specific feedback on products or services. This task has been advanced by constructing a rigorous annotation system, leading to the formation of the UIT-ViSFD dataset, which includes 11,122 mobile e-commerce reviews, providing a benchmark for research. A method based on the Bi-LSTM architecture with fastText word embeddings achieved F1 scores of 84.48% and 63.06% in aspect identification and sentiment classification, respectively, outperforming traditional machine learning and deep learning systems. This achievement has facilitated the development of the SA2SL social listening system, offering strong support for intelligent decision-making by enterprises.