Aspect Extraction Based on Weakly-Supervised Learning for Vietnamese

Aspect extraction is a crucial task in Natural Language Processing (NLP) and an integral part of sentiment analysis systems that rely on aspects. The objective of these systems is to identify the specific aspect being referenced in a comment or review sentence, enabling us to comprehend the customer's response. Recent studies have explored unsupervised and weakly supervised learning techniques to address this challenge. However, there is a lack of detailed assessments specifically for the Vietnamese language, primarily due to the absence of suitable Vietnamese datasets. This paper contributes in two key aspects. Firstly, we have created a dataset specifically tailored for aspect extraction in the Vietnamese domains of restaurant and hotel. Secondly, we explored and proposed an enhanced model based on the UCE approach. Our proposed model achieves higher accuracy and macro F1 scores compared to previous models on both English and Vietnamese datasets.