Novel Table Search (NTS) Datasets

Benchmark Datasets for Novel Table Search This repository contains the datasets released alongside the paper “Novel Table Search.” We provide three benchmark datasets, each derived and adapted from existing benchmarks to suit the specific objectives of this work. Contents For each dataset, the following components are included: • Ground Truth Alignmnet ground truth files are provided: • One manually curated column alignment (for Ugen-V2 Small) • Two manually generated alignments based on column labels Unionable Tables • Precomputed Embeddings Vectors generated using the Starmie and TABBIE models are included to support reproducibility. • Query and Data Lake Tables All queries and associated data lake tables used in our experiments are provided. • Processed Tables Normalized table representations used in our experiments are included in the processed/ folder.