Data Completion of Ride-Hailing Service Based on Tensor Factorization

Abstract

This paper adopts the modified CPWOPT (CANDECOMP/PARAFAC Weighted Optimization) to recover the missing traffic speed data collected in the ride-hailing service. The data completion method based on the tensor decomposition is modified by adding factor tensor in the regular terms, which contains the characteristics of weekdays, time periods, and road segments. After evaluating the performance of the method in the ride-hailing data, the results indicate that the method not only increases the accuracy of data completion compared to CPWOPT, but also fills the missing data more reasonably catering to the temporal distribution of traffic speed data.