Exploring Activity Patterns and Trip Purposes of Public Transport Passengers from Smart Card Data

Abstract

With the development of technology, there is an increasing number of automatically collected data sources applied in human mobility research. Although these data sets can record travel spatiotemporal information, the semantical information of travel cannot be reflected, e.g., activity pattern or trip purpose. In this paper, we proposed a methodological framework to explore the activity patterns and trip purposes of public transit riders using smart card data in an unsupervised way. First, the heuristic rules are proposed to identify home/working activity based on the multidays travel regularity of passengers. Second, we use a modified latent Dirichlet allocation (LDA) model to explore the activity patterns for the remaining activities based on four activity attributes (including arrival time, duration, day of the week, and destination station functional attribute). In this model, the trip attributes of each passenger are considered as a word in the document with specific topics that correspond to different activity characteristics, based on which the trip purpose of each topic is inferred to interpret travel behavior. The proposed methodology is demonstrated using transit smart card data from Beijing. The performance of our model is compared with two baselines based on perplexity and the result shows that our model achieved the best. Besides, the proportions of inferred trip purposes are compared to the values from travel survey in 2020 Beijing Transport Development Annual Report. The reliability of the results is further confirmed. This work can be extended to other automated travel datasets without ground-truth labels and used to understand and predict travel demand. This paper proposes an integrated methodological framework to using trip records to explore activity behavior and infer the trip purpose of public transport passenger from smart card data in an unsupervised way. The methodology is implemented on the smart card data in Beijing and demonstrated the availability. The results show that the performance of our model is superior to the other two baselines, moreover, the proportions of inferred different trip purposes are approximate to the ground-truth data from travel survey in the 2020 Beijing Transport Development Annual Report. The methodology can be extended to other automated travel datasets without ground-truth labels and the result can be used to understand and predict travel demand. The study makes it possible to enrich human mobility data, which eventually would be meaningful for comprehensive city transport planning and management and can initiate a new wave of innovative applications in the public transit network, such as passenger portrayal and targeted advertising.