Visual analytics of spatio-temporal urban mobility patterns via network representation learning

Abstract

Bicycle-sharing systems play an essential role in the transportation system of urban cities due to their outstanding advantages such as more convenience and less pollution. The visualization of large-scale and complex spatio-temporal characteristics of bicycle borrowing and returning data can help analyze citizens’ travel patterns and explore work-life patterns. However, as the scale of data increases and complex spatio-temporal features emerge, analyzing the spatio-temporal patterns of urban transport based on bike-sharing data remains a daunting task. In this paper, we present a novel approach to analyze spatio-temporal patterns and simultaneously construct a visual analysis system. First, node2vec is employed to learn the vectorized representation of spatial correlation characteristics to build a bicycle-sharing network. The t-SNE is adopted to transform high-dimensional vectors into two-dimensional space. Then, the human mobility patterns are extracted by k-means based on the constructed network. A set of visualization and interaction interfaces are provided to explore the pattern evolutions over time with multiple-view collaboration, enabling users to perceive apparent differences between patterns in detail. Case studies based on real-world datasets and interviews with domain experts demonstrate the effectiveness of our system in providing insight into co-occurrence and facilitating various analytical tasks.