Is table retrieval a solved problem?
Exploring Join-aware Multi-table Retrieval

Peter Baile Chen1, Yi Zhang2, Dan Roth3
1MIT, 2AWS AI, 3University of Pennsylvania
Paper Dataset and Code

In real-life database, identifying table-table relationships while conducting table retrieval is critical. Join-aware Multi-table Retrieval (JAR) adjusts rankings from query-table relevance based on table-table relevance and consider both components simultaneously.

Existing table retrieval methods are incomplete

  • Answering questions might need multiple tables
  • Question decomposition may not align with how tables are organized
  • Table relationships need to be inferred

JAR: join-aware multi-table retrieval

  • JAR adjusts rankings from query-table relevance based on table-table relevance and consider both components simultaneously.
  • JAR formulates the problem as an optimization problem and solves it using a mixed-integer linear program that maximizes both query-table relevance and table-table relevance.

BibTeX

@article{chen2024table,
  title={Is Table Retrieval a Solved Problem? Join-Aware Multi-Table Retrieval},
  author={Chen, Peter Baile and Zhang, Yi and Roth, Dan},
  journal={arXiv preprint arXiv:2404.09889},
  year={2024}
}