Multi-Source Sentiment Analysis for Cryptocurrency Prediction Using Machine and Deep Learning Models

Cryptocurrency markets are highly volatile and driven not only by technical indicators but also by rapidly shifting public sentiment expressed across digital platforms. Traditional sentiment analysis approaches using a single data source—either social media or news—are often insufficient for capturing the multidimensional dynamics influencing price movements. This study proposes an enhanced multi-modal sentiment analysis framework that integrates Twitter discussions, online news headlines, and market data to provide a deeper and more reliable understanding of cryptocurrency sentiment. By employing a fusion of machine learning and deep learning techniques such as Logistic Regression, Random Forest, Support Vector Machine, and Long Short-Term Memory networks, the research examines cross-platform sentiment patterns and their predictive power on cryptocurrency price fluctuations. Advanced pre-processing techniques, including tokenization, stemming, stop-word removal, class-imbalance handling, and noise filtering, ensure data quality and improve model performance. The system further incorporates sentiment drift detection and feature importance analysis to explain sentiment variations across platforms and over time. Experimental results demonstrate that multi-source sentiment fusion significantly outperforms singlesource models, achieving improved accuracy and stronger correlation with real market movements. The findings confirm that integrating diverse sentiment streams provides traders, analysts, and investors with more actionable insights, enabling informed decision-making in highly uncertain markets. This study contributes to the evolving field of financial sentiment analysis by demonstrating the benefits of multidimensional data integration and suggesting pathways for future enhancements using transformer-based models and real-time processing.