Copyright © 2026 Authors retain the copyright of this article. This article is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

@article{200957,
        author = {PASAM ANIL and PEDDAPUDI AJAY KUMAR REDDY and PATTEM SAI and PERUBOINA SIVA MANI and Ms.SHALINI.L},
        title = {Stock Price Prediction Using Graph Neural Network with Sentiment- Based Attention},
        journal = {International Journal of Innovative Research in Technology},
        year = {2026},
        volume = {12},
        number = {12},
        pages = {1961-1969},
        issn = {2349-6002},
        url = {https://ijirt.org/article?manuscript=200957},
        abstract = {Stock price prediction is a challenging and consequential task owing to the highly dynamic, nonlinear, and interdependent nature of financial markets. Traditional statistical time-series models such as ARIMA fail to capture complex inter-stock relationships, while deep learning models such as LSTM treat each stock as an isolated entity. This paper proposes a novel end-to-end framework for stock price prediction that integrates a Graph Neural Network (GNN) with a Sentiment-Based Attention mechanism. Stocks are represented as nodes in a multi-relational financial graph, where edges encode sector similarity, rolling price correlation, and supply-chain dependencies derived from regulatory filings. Sentiment signals are extracted from financial news articles and social media posts using a pre-trained FinBERT model. A learned additive attention layer dynamically weights each document by its market relevance for a given stock and time step, producing a compact sentiment context vector. A multi-modal fusion layer combines the GNN graph embeddings, attention-weighted sentiment context, and LSTM-encoded temporal price features for final regression and directional classification. Experiments on 50 S&P 500 constituent stocks spanning five GICS sectors demonstrate a Mean Absolute Error (MAE) of 1.93, Root Mean Square Error (RMSE) of 3.02, classification accuracy of 89.7%, and F1-score of 0.89, outperforming ARIMA, LSTM, GCN, and baseline GNN approaches on all metrics. An ablation study confirms the independent and additive value of each architectural component},
        keywords = {Stock Price Prediction; Graph Neural Network; Graph Attention Network; Sentiment Analysis; Attention Mechanism; Deep Learning; Financial Forecasting.},
        month = {May},
        }