Stock Price Prediction Using Twitter Data Set

Abstract

Stock price prediction has long been a challenging problem in the domain of financial analytics due to the highly volatile, non-linear, and interdependent nature of market data. Traditional machine learning (ML) and deep learning approaches such as Convolutional Neural Networks (CNNs) have achieved moderate success by analyzing historical price patterns and technical indicators. However, these models often fail to capture the complex relationships among multiple financial entities such as stocks, sectors, and macroeconomic variables. To address these limitations, this paper proposes a Graph Neural Network (GNN)-based model combined with Multiple Instance Learning (MIL) for stock price prediction. The GNN framework models relationships between different stocks as a graph structure, where nodes represent stocks and edges represent correlations, sector similarities, or trading relationships. Meanwhile, MIL enables the system to process grouped data (bags of instances), making it suitable for handling temporal sequences and multi-source financial data. The proposed system enhances prediction accuracy by integrating relational learning and instance-level aggregation. It considers not only individual stock behavior but also interdependencies within the financial ecosystem. This approach is expected to outperform traditional ML and CNN-based systems by providing more robust, scalable, and context-aware predictions.