Abstract
Accurate forecasting of stock market trends is crucial for investors and financial analysts, as it enables informed decision-making and risk management. Our research introduces SentiStockPredictor, a novel framework that integrates sentiment analysis with historical stock price data to predict market movements with high precision. By leveraging a Transformer-based model, specifically DistilBERT, our approach processes sequential data to capture the complex time-based dependencies between market sentiment and stock prices. We employ multi-headed selfattention mechanisms, which allow the model to focus on different aspects of the input data simultaneously, and feedforward networks to analyze and synthesize this information. The data is standardized using advanced scaling techniques to ensure consistency and improve model performance. Our extensive experiments demonstrate that SentiStockPredictor achieves an accuracy of over 90%, significantly outperforming traditional models and current benchmarks in predicting stock trends. This superior performance underscores the effectiveness of integrating transformer techniques with social media analytics. The study not only advances the state of stock market prediction but also illustrates the broader potential of using advanced machine learning models to analyze and predict complex, dynamic systems within societal contexts. Our findings suggest that this approach can be extended to other domains where sentiment plays a pivotal role in market behavior.
Keywords: Feed-forward Networks, Multi-Headed Self- Attention, Sentiment Analysis (SA), Senti Stock Predictor, Social Media, Transformer.