Tag
This paper introduces AutoGrable, a method that scores candidate graph constructions for tabular data without training a GNN, using a label-alignment risk based on 1-WL color refinement, enabling cheap search for effective table-to-graph mappings.