Graph-based Learning for Multimodal Route Recommendation

Transportation recommendations are a vital feature of map services in navigation applications. Earlier transportation recommendation systems have struggled to deliver a satisfactory user experience because they focus exclusively on single-mode routes, such as cycling, taxis, or buses. In this paper, we represent the transportation network as a complex network (or graph). Modeling transportation as a network of nodes and edges has gained attention in the literature, generating numerous studies over the years. This approach requires a clear definition of what constitutes a node or an edge: nodes represent stops, while edges represent road segments connecting these stops. Based on this representation, we propose a framework that generates embeddings for each node and edge in the transportation network. These embeddings are created using GRU (Gated Recurrent Units) and GCN (Graph Convolutional Network) models to capture spatial and temporal patterns within the network, while incorporating centrality measures reflecting the influence of each stop. This vector representation facilitates multi-task learning for effective multi-modal transportation recommendations. The proposed framework is applied to the transportation network of Strasbourg, France. Experimental results demonstrate the framework’s efficiency in recommending suitable multimodal transportation routes, considering criteria such as meteorological conditions, safety, and passenger comfort.