MosaicJoin: Efficient Semantic Join Discovery for Data Lakes
A new method called MosaicJoin addresses the trade-off between accuracy and scalability in semantic join discovery. Traditional equi-join approaches fail in data lakes where values referring to the same entity use different syntactic representations. Recent semantic methods either perform slow value-level comparisons or use column-level embeddings that miss fine-grained alignment. MosaicJoin uses a compact sketching strategy to approximate joinability at value level while maintaining scalability. The paper is published on arXiv with ID 2607.21781.
Key facts
- MosaicJoin is a value-level semantic join discovery method.
- It balances accuracy and scalability via a novel sketching strategy.
- The paper is on arXiv, ID 2607.21781.
- It targets data lakes and open-data repositories.
- It addresses the challenge of different syntactic representations for same entities.
- Value-level methods are accurate but scale poorly with high cardinality.
- Column-level methods are efficient but miss fine-grained alignment.
- MosaicJoin approximates joinability of columns.
Entities
Institutions
- arXiv