Abstract
The purpose of this study is to propose a Non-negative Tensor Factorization method for users' movement pattern extraction from a smart card data which is introduced Kotoden train in Kagawa prefecture, Japan. The data set have much information such as trip histories and passenger types of users. This study applied the Non-negative Tensor Factorization to the 594,880 movement patterns which are the combinations of the riding time, the passenger type, the origin station, and the destination station of each trip. As a result, this study succeeded in specifying 43 characteristic movement patterns and confirmed the effectiveness of the proposed approach.
