A method of latent semantic information mining for trajectory data

To explore the potential characteristics of trajectory data, this paper presents a method of latent semantic information mining for trajectory data (T-LSD. We aim to solve the problem of structuring the spatio-temporal data and discovering potential patterns from the trajectory data. First, a vector space model is proposed for trajectory data. Then, by the singular value decomposition of the trajectory matrix, we extract its low-dimensional semantic subspace to further mining its latent semantic information. Finally, we evaluate this research method with a mass of practical trajectory data, which has a total mileage of more than three million kilometers. The experimental results showed that this approach can be employed to analyze the potential characteristics of road network with trajectory data, and can be further used to uncover the driving behavior patterns.