ALORE improves table overlap estimation with alignment-guided hypergraph encoding

Researchers propose ALORE, a scalable and domain-robust estimator for the size of the largest overlap between tables in large repositories. It addresses limitations of the state-of-the-art Armadillo model by explicitly representing row-column structure and exposing inter-table alignment signals during training.

• Uses a Two-View Row-Column Hypergraph encoder to preserve structural membership.
• Employs alignment-guided objectives with inexpensive interaction signals.
• Implements domain-robust value mapping to reduce sensitivity to corpus-specific distributions.
• Reduces Mean Absolute Error (MAE) by up to 55% overall and 69% in zero-shot transfer.
• Achieves up to 89x speedup compared to previous methods.

ALORE demonstrates superior performance across diverse domains and scales, validating its effectiveness for query-by-table retrieval tasks.