MBSE-Net: multi-view attributed graph model for predicting and evaluating incentive impacts on individual-level behaviour status evolution of multimodal transit users

Document Type

Journal Article

Publication Date

2025

Subject Area

place - asia, place - urban, ridership - behaviour, ridership - forecasting, operations - coordination, policy - fares

Keywords

Public transport, Mobility-as-a-Service (MaaS), Incentives, Behaviour dynamics, Deep learning

Abstract

Quantifying the incremental effect of incentive strategies on individual Mobility-as-a-Service (Maas) riders’ travel behaviour is vital for developing effective operation policies. Despite the existing effort in rider engagement promotion, the incremental effect is still not quantified clearly since it has not only an immediate impact but also long-term influences on Maas riders’ lifecycle. To address this challenge, this study proposed the MBSE-Net to estimate the incremental effect of incentives in MaaS platforms by designing a dual-channel multi-modal behaviour status evolution path prediction structure, forecasting the evolution paths on counterfactual (incentivised) and factual (non-incentivised) scenarios in coordination. Since the unobservable behaviour dynamics in the counterfactual and factual scenarios in the same rider, this study designed a multi-view attributed graph model in the proposed MBSE-Net to estimate travel behaviour similarities between incentivised and non-incentivised riders for matching to estimate the incremental effect. Our empirical analysis on two kinds of incentive data, i.e., the Weekly-pass discount incentive and the Random post-trip discount incentive, from Shanghai’s Suishenxing MaaS platform has demonstrated that the proposed MBSE-Net achieves high accuracy in identifying status evolution paths and anticipating churn events with an 85.03% churn recall and 80.30% behaviour status evolution path accuracy. Results have revealed that the Weekly-pass discount incentives yield significantly greater uplifts than the random post-trip discount incentives in both short-term (within the incentive week) and long-term (multi-week status evolution path) contexts. Medium-frequency and low-regularity riders exhibit the strongest long-term engagement response to incentives. Moreover, cumulative status evolution path incremental effects (about 0.31) substantially exceed the immediate one-week effects (about 0.10), underscoring the strategic importance of modelling extended behaviour status evolution. This study has further offered actionable view for the MaaS platform based on the findings on targeted and personalised incentive design, showing the benefits of sustained incentive strategies and inventive mixes to improve retention.

Rights

Permission to publish the abstract has been given by Elsevier, copyright remains with them.

Comments

Transportation Research Part C Home Page:

http://www.sciencedirect.com/science/journal/0968090X

Link to Full Text

Share

COinS