Deep Learning and Optimization Approaches in Deformable Graph Convolutional Networks with NLP Based Social Sentimental Data for Enhanced Stock Price Predictions: A Review

Abstract

Financial markets are inherently complex, nonlinear, and influenced by both quantitative factors and human sentiment, making stock price prediction a challenging task. Traditional statistical models such as ARIMA and GARCH rely on assumptions of linearity and stationarity, limiting their effectiveness in real-world financial environments. The emergence of deep learning has enabled the modeling of complex temporal dependencies and integration of heterogeneous data sources for improved forecasting accuracy. This review paper examines advanced deep learning approaches, with a focus on Deformable Graph Convolutional Networks (DGCNs) integrated with Natural Language Processing (NLP)-based sentiment analysis. DGCNs enhance conventional graph neural networks by introducing adaptive graph structures that dynamically capture evolving inter-stock relationships and market dependencies. Simultaneously, transformer-based NLP models such as BERT and FinBERT extract meaningful sentiment signals from unstructured data sources including financial news, social media, and analyst reports. These sentiment features are combined with structured financial data to form a hybrid predictive framework capable of capturing both market dynamics and investor behavior. The study also reviews optimization strategies, datasets, and evaluation metrics used in recent research, highlighting improvements in predictive performance. Despite advancements, challenges such as data alignment, model complexity, and generalization persist. This work provides a comprehensive overview of emerging methodologies and future directions for intelligent financial forecasting systems.