VED_PhoBERT: Enhancing Vietnamese Sentiment Analysis with Emoji Descriptions Integration

Abstract

Sentiment analysis incorporating emojis has gained increasing attention in recent years, as emojis serve as visual symbols that convey emotional nuances and contextual information, helping to fill the gap left by the absence of non-verbal signals in digital communication. While integrating emojis has significantly enhanced sentiment analysis performance, this aspect remains underexplored in Vietnamese. This study proposes integrating emojis into sentiment analysis through the creation of an emoji description dictionary (called the Vietnamese emoji dictionary). During preprocessing, emojis are replaced with corresponding descriptions to preserve the original emotional intent of the author. Furthermore, our method leverages PhoBERT, a state-of-the-art pre-trained model for Vietnamese text processing. Experimental evaluations on two benchmark datasets demonstrate that the proposed approach (VED_PhoBERT (https://github.com/hqvjet/VivelAI/tree/VED_PhoBERT)) outperforms the previous best-performing model, ViSoBERT, in sentiment analysis task.