Something's Fishy In The Data Lake: A Critical Re-evaluation of Table Union Search Benchmarks

TL;DR

This paper critically re-evaluates current benchmarks for table union search in data lakes, revealing their limitations and proposing criteria for more effective future evaluations of semantic understanding.

Contribution

It identifies key shortcomings in existing TUS benchmarks and offers essential criteria to improve their realism and reliability for future research.

Findings

Simple baselines outperform complex methods on current benchmarks.

Current benchmarks are heavily dataset-specific and do not isolate semantic understanding.

Proposes criteria to enhance future benchmark design.

Abstract

Recent table representation learning and data discovery methods tackle table union search (TUS) within data lakes, which involves identifying tables that can be unioned with a given query table to enrich its content. These methods are commonly evaluated using benchmarks that aim to assess semantic understanding in real-world TUS tasks. However, our analysis of prominent TUS benchmarks reveals several limitations that allow simple baselines to perform surprisingly well, often outperforming more sophisticated approaches. This suggests that current benchmark scores are heavily influenced by dataset-specific characteristics and fail to effectively isolate the gains from semantic understanding. To address this, we propose essential criteria for future benchmarks to enable a more realistic and reliable evaluation of progress in semantic table union search.