A Machine Learning Approach to Study Tourist Interests and Predict Tourism Demand on Bonaire Island from Social Media Data

Studying tourist interest on the Caribbean island Bonaire might be a good step to improving tourism management. Tourism brought Bonaire economic growth but also puts pressure on the island's natural ecosystem. Previous studies on tourist interest based on surveys are labour-intensive, time-consuming, and expensive. This paper explores whether the use of freely available social media data combined with automatic machine learning methods can function as a cheap and fast alternative to surveys. From 2003 to 2019, 13,706 geotagged Flickr data points assigned keywords, then weighted using TF-IDF (Term Frequency-Inverse Document Frequency), and finally clustered with DB-SCAN (Density-Based Spatial Clustering of Noise Applications). Two factors determine whether a cluster has an associated unique activity/interest: the most relevant and least relevant keywords. Eight identified clusters are useful for interpreting Bonaire tourists' interest: urban tourism; nature tourism around the lake; in-land natural tourism; conch shell and food; unique fishes; windsurf activity; cruise and ship; carnival, parade and singing. Tourism demand was forecasted using both Flickr and CBS (Centraal Bureau voor de Statistiek) data. Flickr data could show which continent the tourist came from in which seasons (Winter, Spring, Summer, Autumn) from 2015 to the end of 2021.