Learning static embeddings from public transport networks modeled as heterogeneous bipartite graphs for zero-shot travel and dwell time prediction
