From Pictures to Travel Characteristics: Deep Learning-Based Profiling of Tourists and Tourism Destinations

Tourism products are complex and strongly tied to emotions. Thus, it is not easy for consumers to explicitly communicate their travel preferences, needs, and interest, especially in the early phase of travel decision making. In the spirit of the idiom "A picture is worth a thousand words" we utilize pictures to characterize tourists as well as tourism destinations in order to build the foundations of a recommender system (RS). In this work all entities (i.e., users and items) are characterized using the Seven-Factor Model. Pre-labelled pictures are used in order to train convolutional neural networks (CNN) in a transfer learning manner with the goal to extract the Seven-Factors of a given picture. We demonstrate that touristic characteristics can be extracted out of pictures. Furthermore, we show that those characteristics can be aggregated for a collection of pictures, such that a representation of a user or a destination can be determined respectively.