Quantifying Challenges in the Application of Graph Representation Learning

Graph Representation Learning (GRL) has experienced significant progress as a\nmeans to extract structural information in a meaningful way for subsequent\nlearning tasks. Current approaches including shallow embeddings and Graph\nNeural Networks have mostly been tested with node classification and link\nprediction tasks. In this work, we provide an application oriented perspective\nto a set of popular embedding approaches and evaluate their representational\npower with respect to real-world graph properties. We implement an extensive\nempirical data-driven framework to challenge existing norms regarding the\nexpressive power of embedding approaches in graphs with varying patterns along\nwith a theoretical analysis of the limitations we discovered in this process.\nOur results suggest that "one-to-fit-all" GRL approaches are hard to define in\nreal-world scenarios and as new methods are being introduced they should be\nexplicit about their ability to capture graph properties and their\napplicability in datasets with non-trivial structural differences.\n