Evaluation of classifying geotagged photos aiming at analysis of tourist behavior

Abstract

In this research, we evaluate classification method in order to understand preference of tourists from geotagged photos and to build a system for recommendation of tourism spots aimed at spatial dispersion of tourists. Three clustering method is applied to collected photos and their labels: 1) principal component analysis and nonhierarchical clustering, 2) ClustOfVar, an R package and hierarchical clustering, and 3) topic model by Latent Dirichlet Allocation. By focused on spatial distribution of photo and tf-idf, which is used to evaluate importance of words in documents, we evaluate each method. Classification by ClustOfVar is able to clusify photos which match contents of label. On the other hand, classification by topic model shows different result from other methods, thus there is room for improvement.