ContextRank: Personalized Tourism Recommendation by Exploiting Context Information of Geotagged Web Photos

In this paper, we propose a method: Context Rank, which utilizes the vast quantity of geo tagged photos in photo sharing website to recommend travel locations. To enhance the personalized recommendation performance, our method exploits different context information of photos, such as textual tags, geo tags, visual information, and user similarity. Context Rank first detects landmarks from photos' GPS locations, and estimates the popularity of each landmark. Within each landmark, representative photos and tags are extracted. Furthermore, Context Rank calculates the user similarity based on users' travel history. When a user's geo tagged photos are given, the landmark popularity, representative photos and tags, and the user similarity are used to predict the user preference of a landmark from different aspects. Finally a learning to rank algorithm is introduced to combine different preference predictions to give the final recommendation. Experiments performed on a dataset collected from Panoramio show that the Context Rank can obtain a better result than the state-of-the-art method.