Discrete Emotions Detection on Social Media: A Systematic Review and Comparative Analysis
Meng-Jie Wang, XiaoCui Zhang · 2024
Despite a growing corpus of work delving into social media communication, gauging discrete emotions on such sites remains challenging, especially for researchers with constrained computational expertise. By synthesizing insights from nearly 2,000 studies, we seek to address this gap through an extensive review and comparative analysis that pinpoints the most prevalent and user-friendly tools for emotional detection and verifies their effectiveness against a validated human-annotated benchmark. While each identified tool captures a spectrum of affective expressions in line with established psychological frameworks, considerable disparity emerges when examining their efficacy. Lexicon-based approaches, such as NRC EmoLex and Text2Emotion, though frequently utilized in the reviewed studies, yield merely adequate precision, with LIWC-22 providing a marginally enhanced outcome. IBM NLU, however, leveraging advanced deep learning techniques, demonstrates superior performance across both sentimental and nuanced emotional assessments. ChatGPT-4, notably, unlike some studies praising its extensive applicability, may falter in this context, particularly noted for its reduced recall. Amid the rapid advancements in social media communication, such findings may provide a foundational exploration for discerning discrete emotions online, paving the way for the development of refined navigational tools in the complex emotional landscape of the digital realm.