Multi-view metro station clustering based on passenger flows: A functional data-edged network community detection approach

Publication Type

Journal Article

Publication Date

2-2023

Abstract

This paper aims at metro station clustering based on passenger flow data. Compared with existing clustering methods that only use boarding or alighting data of each station separately, we focus on higher granularity origin-destination (O-D) path flow data, and provide more flexible and insightful clustering results. In particular, we regard the metro system as a network, with each station as a node. The real-time passenger flows over time between different O-D paths serve as directed edges between nodes. Compared with traditional networks, our edges are temporal curves, and can be regarded as functional data. For this functional data-edged graph, we are the first to develop a novel community detection approach for node clustering. Our method is based on functional factorization. First a dual time-warped sparse nonnegative functional factorization is proposed for extracting patterns of the functional edges. Then the passenger flow of each O-D path can be regarded as a linear combination of different extracted passenger flow patterns. Based on it, we construct a multi-view directed and weighted network, where each view represents one particular pattern, and the factorization coefficient of each O-D path on this pattern is treated as the weight of this directed edge in this particular view. Then a novel community detection algorithm based on nonnegative matrix tri-factorization is constructed according to the topological structure of the multi-view network. The fusion of different views can be either subjectively determined or objectively learnt in a data-driven way, which gives flexibility of the clustering algorithm to emphasize on different travel patterns. Two real datasets of Singapore and Hong Kong metro systems are used to validate the proposed method.

Keywords

Functional data-edged network, Multi-view network community detection, Nonnegative functional factorization, Passenger flow pattern extraction, Station clustering, Smart card data

Discipline

Databases and Information Systems | Numerical Analysis and Scientific Computing | Transportation

Research Areas

Data Science and Engineering

Publication

Data Mining and Knowledge Discovery

Volume

37

Issue

3

First Page

1154

Last Page

1208

ISSN

1384-5810

Identifier

10.1007/s10618-023-00916-w

Publisher

Springer

Copyright Owner and License

Authors

Additional URL

https://doi.org/10.1007/s10618-023-00916-w

This document is currently not available here.

Share

COinS