A Robust Distance-Weighted Collaborative Filtering Method for Route Recommendation Based on Historical Trajectory Data

Modern navigation systems try to provide users with the optimal route considering multiple factors, such as travel time and user preference. However, the sampling of historical route data could be discrete and intermittent, which brings a great challenge to learning user preference correctly. Meanwhile, route recommendation methods often face the problem of insufficient robustness and stability, which deteriorates the user experience. This paper introduces an effective approach to reconstructing routes from historical trajectory data with a low sampling rate using an entropy-based method for GPS trajectory segmentation and a hidden Markov model to match trajectories to road networks. We further introduce a distance-weighted collaborative filtering (DWCF) method for route recommendation, considering both driver preference and route distance. Experiments based on massive driving trajectory data show that the DWCF method can effectively learn driver preference from historical data and show superior robustness and stability over the other benchmark methods.