Research and Analysis on the Relationship between Social Media Emotions and Stock Market Fluctuations

Contemporary financial markets are driven by more than traditional economic data; the proliferation of social media has introduced a powerful, real-time stream of public sentiment. To enhance the robustness of financial forecasting models, the effective utilization of this data is of great importance. This paper explores the extent to which public sentiment, gathered from social media, can predict stock market fluctuations. The research puts forward a hybrid FinBERT-GRU model, which first applies the domain-specific FinBERT to calculate a daily sentiment score from raw social media posts. Following this, the sentiment score is integrated with historical price data into a Gated Recurrent Unit (GRU) network to forecast the subsequent day’s volatility. Experiments conducted for this study, based on a public collection of tweets about Apple Inc. (AAPL), reveal that the FinBERT-GRU model substantially outperforms a traditional GARCH model and a baseline GRU network lacking sentiment input, achieving lower prediction errors across all metrics. These results confirm that quantified public sentiment offers significant predictive insights beyond what historical price data alone can provide, presenting a solid framework for improving volatility forecasts.