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

Area of Science:

• Data Science
• Computational Social Science
• Cultural Heritage Management

Background:

• Traditional museum visitor forecasting often uses limited data types.
• External factors like public opinion are increasingly recognized as influential.
• Integrating diverse data sources can improve predictive accuracy.

Purpose of the Study:

• To develop an advanced forecasting model for annual museum visitors.
• To assess the impact of unstructured sentiment data on visitor numbers.
• To compare the performance of various predictive algorithms.

Main Methods:

• Collected and integrated structured museum data with unstructured sentiment data from news and comments.
• Extracted sentiment scores to quantify public opinion.
• Evaluated seven algorithms: Linear Regression, Random Forest Regressor, RNN, GAN, CNN, LSTM, and Transformer.
• Selected the Transformer model for its superior predictive performance.

Main Results:

• The Transformer model achieved the highest accuracy across RMSE, MSLE, and MAPE metrics.
• Incorporating sentiment data significantly enhanced the precision of visitor forecasts.
• Public sentiment and media narratives demonstrably influence museum visitor behavior.

Conclusions:

• A robust forecasting framework integrating structured and unstructured data was established.
• Sentiment analysis is a valuable component for accurate visitor prediction.
• Findings provide practical implications for sustainable museum planning and strategic decision-making.