Style-Oriented Personalized Landmark Recommendation

Abstract

Personalized travel recommendation has attracted lots of research attention in both academic and industry communities. Although a great progress has been achieved so far, existing travel recommender systems have not well-exploited users’ style-oriented preference on landmarks and local preference on the targeted city. Typically, users have their own preferences on the styles of landmarks (e.g., natural scenes or historic sites). When visiting a city, their preferences will be affected by the characteristics of this city (e.g., historic or scenic) or the “must-go” landmarks, as well as local contexts such as distance and time constraints. In this paper, we propose a novel style-oriented recommender system, which considers all the above factors to facilitate personalized landmark recommendation. Specifically, we first propose a unified classifier to detect landmark styles based on domain adaptation by leveraging web-photos in the source domain and landmark-image in the target domain. The detected landmark styles are then utilized to learn users’ style-oriented preferences based on users’ travel records in the past. Next, given a targeted city, the influence of users’ landmark style preferences and the characteristics of the must-go landmarks of this city are simultaneously considered by a proposed style-oriented recommender system to make optimal recommendations. In addition, we further study the effects of local contexts, such as landmark popularity or location, on the performance of landmark recommendation. Extensive experiments on the real-world travel data of six cities demonstrate the effectiveness of the proposed style-oriented landmark recommendation strategy.