A Comprehensive Review of Deformable Graph Convolutional Networks with NLP Based Social Sentimental Data for Enhanced Stock Price Predictions

Abstract

Financial markets are highly complex systems characterized by dynamic interactions among economic, behavioral, and informational factors, making stock price prediction a persistent challenge in computational finance. Traditional forecasting methods, relying on historical prices and technical indicators, often fail to capture nonlinear dependencies and evolving inter-asset relationships. Recent advances in deep learning, particularly graph neural networks and transformer-based natural language processing (NLP), offer promising alternatives for modeling such complexities.  This paper presents a comprehensive review of hybrid approaches that integrate Deformable Graph Convolutional Networks (DGCNs) with NLP-based sentiment analysis for enhanced stock prediction. DGCNs extend conventional graph models by enabling adaptive receptive fields and dynamic relationship modeling, allowing better representation of evolving financial networks. Concurrently, transformer-based models such as BERT and FinBERT extract contextual sentiment from unstructured data sources including news, social media, and financial reports. The reviewed framework combines structured financial time-series data with sentiment embeddings to construct dynamic graphs, where nodes represent assets and edges capture correlations and sentiment-driven dependencies. Empirical studies across global markets demonstrate that such hybrid models outperform traditional and single-modality approaches. Despite promising results, challenges remain in scalability, data reliability, and real-world deployment. This survey provides insights into future directions for intelligent, data-driven financial forecasting systems.