A Survey of Surveys (SoS) on Sentiment Analysis using Machine Learning

Abstract

This paper presents a comprehensive Survey of Surveys (SoS) on sentiment analysis using machine and deep techniques. Although several survey studies have been conducted in this domain, the absence of a centralized and structured synthesis makes it difficult for researchers to identify trends challenges, and research opportunities. To address this gap, this study systematically reviewed 70 papers categorizing them based on application domain such as fake news detection, hate speech detection, sarcasm analysis, context based, and content-based sentiment analysis. Unlike other surveys, the work provides a meta level analysis by comparing existing surveys, identifying methodological patterns, and highlighting their limitation. We present a structured taxonomy of sentiment analysis approaches, alongside an examination of the evolution from classical machine learning to deep learning techniques. In addition, recent advances such as transformer-based architecture and large language models are discussed to provide updated context. In addition, this study identifies key challenges, including data imbalance, domain adoption, multimodal complexity, and lack of standardized evaluation frameworks. The SoS serves as a comprehensive reference for researchers by consolidating existing knowledge, identifying research gaps and suggesting future research directions.