A Web-Based Gold Price Prediction Model Using Geopolitical Sentiment from Social Media and Gated Recurrent Unit (GRU)

Abstract

Gold is widely recognized as a safe-haven asset whose price is sensitive to economic and geopolitical uncertainty. This study develops a web-based international gold price prediction system using the Gated Recurrent Unit (GRU) algorithm and evaluates the contribution of geopolitical sentiment from the X social media platform. The study uses historical XAUUSD data and English-language posts related to the Iran–United States–Israel conflict. Text data were processed through cleaning, case folding, tokenization, stopword removal, lemmatization, TF-IDF transformation, and sentiment analysis using VADER. The resulting daily sentiment scores were integrated with historical gold price features and used as an additional input to the GRU model. Two experimental scenarios were evaluated: GRU without sentiment and GRU with sentiment. The results show that the GRU model without sentiment achieved an MAE of 55.07, RMSE of 69.81, MAPE of 1.17%, and R² of 0.9629, while the model with sentiment achieved an MAE of 80.87, RMSE of 92.93, MAPE of 1.72%, and R² of 0.9342. These findings indicate that incorporating daily aggregated social media sentiment did not improve prediction performance for the dataset used. The developed Flask-based web application provides prediction, sentiment analysis, visualization, and model evaluation features, demonstrating the practical implementation of the proposed approach.