Social networks now function as primary conduits for real-time financial data, yet the mechanisms by which they induce stock market volatility remain complex and often non-linear. This study interrogates the current state of predictive modeling by systematically reviewing literature that bridges sentiment analysis with financial econometrics. Unlike previous surveys restricted to dominant platforms like Twitter, this research broadens the scope to include diverse digital channels and high-variance periods, such as the global COVID-19 crisis. Evaluating methodologies retrieved via comprehensive keyword indexing across major academic databases reveals that hybrid models—pairing Naïve Bayes or Tree-based algorithms (XGBoost, Random Forest) with sentiment lexicons and GARCH variants—offer superior predictive accuracy for near-term horizons. Despite these successes, we identify a persistent methodological flaw in extant studies: a tendency toward short observation windows and single-source data dependency. Consequently, we propose a research agenda prioritizing the development of cross-platform, longitudinally robust models capable of capturing the nuanced, non-linear dependencies inherent in modern market dynamics.