A Multi-class Ride-hailing Service Subsidy System Utilizing Deep Causal Networks

TL;DR

The paper tackles subsidy allocation in multi-class ride-hailing services under confounding in observational data by introducing MulTeNet, a multi-branch neural network that jointly estimates treatment propensity, base conversion rate, and uplift with monotonicity constraints and orthogonal regularization. It integrates MulTeNet with a Model Predictive Control–based optimization to allocate subsidies across service classes and clusters within a fixed budget, using a primal-dual solver and periodic updates to handle modeling errors. Offline training on historical data and online inference with a fairness-aware clustering approach enable real-time, low-latency subsidy decisions delivered via a lookup dictionary. Empirical results on a large-scale dataset show improved uplift metrics (AUUC, QINI) and favorable business outcomes (revenue, orders, ROI) compared with baselines, highlighting practical impact for monetizing subsidy strategies in dynamic ride-hailing markets.

Abstract

In the ride-hailing industry, subsidies are predominantly employed to incentivize consumers to place more orders, thereby fostering market growth. Causal inference techniques are employed to estimate the consumer elasticity with different subsidy levels. However, the presence of confounding effects poses challenges in achieving an unbiased estimate of the uplift effect. We introduce a consumer subsidizing system to capture relationships between subsidy propensity and the treatment effect, which proves effective while maintaining a lightweight online environment.

Figures (2)

• Figure 1: The illustrated structure of the proposed MulTeNet
• Figure 2: The framework of our system