Beyond the Match: Predicting Bilateral-matching Satisfaction with Multi-view Heterogeneous Graph Neural Network

Bilateral matching is critical in platform economies, as matching quality directly influences participant satisfaction and the platform’s long-term success. While existing studies mainly optimize demand-supply matching for revenue or stability, few systematically examine matching quality from a bilateral satisfaction perspective. To address this gap, we propose a satisfaction prediction method based on a multi-view heterogeneous graph neural network. Guided by the Kano model and the anchoring effect, the model captures users' individual preferences and behavioral features across scenarios, distinguishing satisfaction with basic and additional services via an attention mechanism. A multi-task learning framework simultaneously predicts users’ preference features and bilateral satisfaction. We validate our approach through a collaboration with China’s largest ride-hailing platform, using three months of matching and evaluation data. Experimental results demonstrate that our model effectively predicts bilateral satisfaction, offering theoretical and practical insights for enhancing platform performance.