Learning Travel Recommendations from User-Generated GPS Traces

The advance of GPS-enabled devices allows people to record their location histories with GPS traces, which imply human behaviors and preferences related to travel. In this article, we perform two types of travel recommendations by mining multiple users' GPS traces. The first is a generic one that recommends a user with top interesting locations and travel sequences in a given geospatial region. The second is a personalized recommendation that provides an individual with locations matching her travel preferences. To achieve the first recommendation, we model multiple users' location histories with a tree-based hierarchical graph (TBHG). Based on the TBHG, we propose a HITS (Hypertext Induced Topic Search)-based model to infer the interest level of a location and a user's travel experience (knowledge). In the personalized recommendation, we first understand the correlation between locations, and then incorporate this correlation into a collaborative filtering (CF)-based model, which predicts a user's interests in an unvisited location based on her locations histories and that of others. We evaluated our system based on a real-world GPS trace dataset collected by 107 users over a period of one year. As a result, our HITS-based inference model outperformed baseline approaches like rank-by-count and rank-by-frequency. Meanwhile, we achieved a better performance in recommending travel sequences beyond baselines like rank-by-count. Regarding the personalized recommendation, our approach is more effective than the weighted Slope One algorithm with a slightly additional computation, and is more efficient than the Pearson correlation-based CF model with the similar effectiveness.