A Matrix-Decomposition-Based Context Tensor Approach for Personalized Travel Time Estimation

Xiaopei Li, Fanzhang Li, Xin Dai, Helan Liang

Abstract: In this paper, we use the matrix-decomposition-based context-aware tensor approach to estimate the user's personalized travel time. The model mainly has the following steps: firstly, by dealing with the GPS trajectories received in current time period to filter the appropriate data and the filtered data is expanded into a third-order tensor, corresponding to the driver, the road segment and the period of time. Sorting the GPS data of the past period into a third-order tensor of the same form. Each data in the tensor represents the travel time of a driver on a certain road during a certain period of time. Secondly, based on the definition of the third-order tensor, the third-order tensor can be considered as a combination of multiple second-order tensors (matrices). According to the slicing operation, the matrix of the invisible position is taken from the tensor, and the pre-processing is performed by using matrix decomposition to fill the vacancy. Thirdly, the feature matrix of different time space and geographic location, i.e. the context information, is extracted by the known GPS trajectory dataset to decompose the travel-time tensor, and then the time cost of a certain driver on a certain road in a certain period is obtained. Personalized travel time forecasts provide users with more choices of travel options and travel routes.