Mining Tourist Preferences with Twice-Learning

Abstract

Data mining techniques have been recognized as powerful tools for predictive modeling tourist decision-making process. However, two practical yet important problems have not been resolved by the data miners in empirical tourism research. Firstly, comprehensibility-the role of the data mining should not only generate accurate predictions, but also provide insights why certain prediction is made. But most widely used data mining methods that can generalize well are black-box in nature and can provide little information on the tourist decision-making facts. Secondly, the lack of training samples-it is usually rather difficult to collect enough training samples through surveying the tourist on site, especially for surveying the tourist’s decision-making facts. Many data mining methods may not achieve satisfactory performance if learned on small data set. In this paper, we show that these two problems can be addressed simultaneously using a twice-learning framework on the travel preference data. The results indicate that by addressing these two problems properly, we can predict tourist preferences accurately as well as extracting meaningful insights which would be useful for tourism marketing.