Enhancing Comparative Opinion Mining in Vietnamese Product Reviews: A Hybrid Generative Model Approach with Knowledge Base Integration

Abstract

Opinion mining holds numerous practical applications, especially as user-generated sentiment data become increasingly prevalent in the digital age. Comparative Opinion Mining (ComOM), a specific sub-field of opinion mining, focuses on extracting the elements involved in the comparison in a text and retrieving the corresponding tuples expressing comparative opinions. This research describes an improved version of our system that participated in the VLSP 2023 ComOM shared task. The baseline system’s performance was ranked second among 22 participating systems in the shared task through a combination of generative model and classification-based approaches, along with knowledge-based techniques with a 0.2300 F1-score. More experiments have been conducted to improve the performance of the system, reaching a final F1-macro score of 0.2391. This demonstrates the superiority of our proposed method compared to existing approaches in the task of comparative opinion mining.