MetaRTL improves efficiency and scalability in relational table learning
MetaRTL: Meta-path Attention Enhanced Relational Table Learning
Artificial Intelligence
Summary
Relational databases store information in tables that are linked in complex ways. Learning meaningful patterns from these tables usually requires complex and slow methods. The authors created MetaRTL, a two-step approach that first prepares simple table representations, then combines relational connections more efficiently using a special attention method called MetaAttn. Their approach works faster and better on many real-world datasets.
What this means in practice
- •For database engineers: Improve performance and scalability when extracting insights from large relational databases by integrating MetaRTL into existing data pipelines.
- •For enterprise data teams: Accelerate analytics tasks on complex multi-table corporate databases by using MetaRTL to get richer relational features efficiently.
Authors
Ken Zhong, Weichen Li, Zheng Wang
Abstract
Relational table learning has gained increasing attention with the widespread use of relational databases. Existing methods typically rely on deep GNN or HGNN stacks, leading to high computational costs and limited performance on large real-world databases. We propose MetaRTL, a two-stage framework for scalable and expressive relational table learning. In the first stage, MetaRTL obtains initial table embeddings via lightweight pre-training. In the second stage, it performs non-parametric message passing to derive meta-path features, which are then aggregated by an attention module, MetaAttn. By shifting computation from deep message passing to efficient meta-path aggregation, MetaRTL captures rich relational semantics while maintaining high efficiency. Experiments on 10 real-world datasets across 24 tasks demonstrate the effectiveness of the proposed method.