Development of an Effective Sentiment Classification Architecture for Texts Containing Negation Words and Emojis

Abstract

Abstract The widespread use of social media platforms like Facebook, Instagram, Reddit, and Twitter has profoundly influenced human interactions and decision-making. Twitter, in particular, has become a popular platform for sharing opinions on diverse topics, resulting in a substantial volume of user-generated content known as tweets. Emojis, which convey emotions and moods, have become integral to online communication. However, existing sentiment analysis methods struggle with the unique characteristics of tweets, such as negation words, contracted phrases, shortened text, language ambiguities, and emojis. To address these challenges, a novel word representational framework is proposed in this research, considering both negation words and emojis. Sentiment analysis is performed using Long Short-Term Memory (LSTM) neural networks to capture both syntactic and semantic features of tweets, enhancing sentiment classification efficiency. Extensive experiments on three benchmark datasets validate the approach, resulting in impressive accuracy rates of 95.24%, 87.13%, and 94.18% on different datasets. These results underscore the effectiveness of the proposed method in handling the complexities of sentiment analysis in the context of tweets, with applications in business insights, government policy formulation, decision-making processes, and brand identity monitoring.