Table representation learning using heterogeneous graph embedding

Willy Carlos Tchuitcheu, Tan Lu, Ann Dooms

Abstract: Highlights•A Table Representation Learning (TRL) method using a Heterogeneous Graph Table (HGT).•Semantic Meta-Path (SMP) implements biased walks on HGT to extract meaningful paths.•NLP modeling on SMPs learns visual-semantic embeddings of table entries.•A TRL-augmented Transformer that uses a Semantic Matrix as an attention mask.•Our method shows competitive performance and reduced complexity versus existing ones.