MFGCN: A Multimodal Fusion Graph Convolutional Network for Online Car-Hailing Demand Prediction

Abstract

<italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">The rapid growth of online car-hailing provides an excellent opportunity to meet convenient travel services. However, with the tremendous increase of users and online taxis, online car-hailing prediction systems face several challenges: (1) the hardness of modeling nonlinear spatio-temporal interactions between users and vehicles; (2) the difficulty of incorporating context information and multimodal attribute enhancement data; (3) the problems of data sparsity. To cope with these challenges, we propose a novel Multimodal Fusion Graph Convolutioal Network (MFGCN) for online car-hailing prediction. The model consists of a Multimodal Origin Destination Graph Convolutional Network (MODGCN) module that contains three graph convolutional networks to extract spatial patterns from Geography, Semantics, and Functional Correlation, a Multimodal Attribute Enhancement (MAE) module incorporates weather and temporal activity pattern, and a Temporal Attention Skip-LSTM (TAS-LSTM) module captures the periodic variations. Extensive experiments conducted on the real-world taxi demand datasets show that MFGCN outperforms the state-of-the-art methods.</i>