A Machine Learning-Based Sentiment Analysis of Online International Tourist Reviews: A Study of Heritage Sites in Anuradhapura, Sri Lanka

Abstract

In the contemporary digital landscape, online reviews and other user-generated content significantly influence global travel behaviours by revealing first-hand visitor perspectives. However, these insights remain largely underexplored in tourism research and destination management. This study examines the application of text mining and machine learning-based sentiment classification to analyse tourist reviews of heritage sites in Anuradhapura, Sri Lanka, aiming to produce data-driven insights to support tourism planning. Reviews collected from Google and TripAdvisor between 2018 and 2024 were pre-processed and categorized into positive, neutral, or negative sentiments using a decision tree classifier. The model produced an overall accuracy of 80.85% with substantial inter-rater agreement (kappa = 0.634), effectively identifying prevailing sentiment patterns. Findings show that positive sentiments were driven by aesthetic value, architectural heritage, and spiritual enrichments, while negative sentiments emphasized operational deficiencies, unmet visitor expectations, and disappointing on-site conditions. The study highlights the value of automated sentiment analysis as a decision-support tool for heritage site management, offering actionable insights for targeted improvements and enhancing visitor satisfaction. The proposed approach is scalable and transferable to other heritage destinations worldwide.