Enhancing museum visitor forecasting using deep learning and sentiment analysis: A transformer-based approach for sustainable management

Abstract

This study aims to develop a forecasting model that predicts the annual number of museum visitors by integrating structured museum-related data and unstructured sentiment data. While prior research has often relied on a single data type or traditional regression techniques, this study incorporates sentiment scores extracted from museum-related news articles and user comments to empirically assess the influence of external public opinion. Seven predictive algorithms including traditional models (Linear Regression and Random Forest Regressor) and deep learning models (RNN, GAN, CNN, LSTM, and Transformer) were evaluated for performance. Among these, the Transformer model demonstrated the highest predictive accuracy across all evaluation metrics (RMSE, MSLE, and MAPE) and was adopted as the final forecasting model. The results show that incorporating sentiment data significantly enhances forecasting precision, highlighting the substantial impact of media narratives and public sentiment on visitor behavior. This study offers a robust forecasting framework that integrates both structured and unstructured data, providing practical implications for sustainable museum planning and strategic decision-making.