Enhancing the efficiency of sentiment analysis for Vietnamese service reviews

Abstract

Sentiment analysis has emerged as a powerful tool for understanding customer sentiment and enhancing service quality in the digital age. The service industry is becoming a significant driver of economic growth in Vietnam. However, customer sentiment analysis in the service industry for Vietnamese has not received due attention. Therefore, in this study, we propose three methods to improve the quality of product sentiment analysis in the service industry. First, we combine Title with Content rather than using Content as it is currently. In addition, emojis have become an indispensable part of the way users express emotions, helping to display emotional signals, solve semantic dreams, and improve model accuracy. However, famous models such as PhoBERT have not yet integrated emojis into sentiment analysis, except for ViSoBERT. In this study, we propose to integrate emojis into sentiment analysis. In VED_PhoBERT, we build a dictionary for Emoji. In E2V_PhoBERT, we present a refined version of PhoBERT that incorporates emoji understanding via Emoji2Vec vectors. Experimental results on various Machine Learning (ML) and Deep Learning (DL) models show promising outcomes. Combining Title and Content yields superior results when using Content. Likewise, VED_PhoBERT and E2V_PhoBERT yield significantly superior results compared to the current best model, ViSoBERT. In addition, this study has three papers that have been published or accepted in conferences whose proceedings are indexed by Scopus.