Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

STAR: Semantic Table Representation with Header-Aware Clustering and Adaptive Weighted Fusion

Shui-Hsiang Hsu, Tsung-Hsiang Chou, Chen-Jui Yu, Yao-Chung Fan

TL;DR

STAR tackles the semantic gap in table retrieval by introducing a header-aware semantic clustering stage that yields diverse, representative table content and cluster-specific synthetic queries, followed by a weighted fusion stage that separately encodes and combines table and query signals. The framework replaces heuristic top-k sampling and simple concatenation with a principled two-stage process (SCQG and WF), using header information and adaptive weights to achieve finer-grained semantic alignment. Empirical results across five benchmarks show STAR consistently outperforms the QGpT baseline, with notable gains in Recall@1 and Recall@5, and demonstrate the benefits of dynamic weighting and cluster-guided query generation. The work highlights the importance of semantic coverage and source-aware fusion for robust, expressive table representations, with practical implications for scalable, query-driven table retrieval.

Abstract

Table retrieval is the task of retrieving the most relevant tables from large-scale corpora given natural language queries. However, structural and semantic discrepancies between unstructured text and structured tables make embedding alignment particularly challenging. Recent methods such as QGpT attempt to enrich table semantics by generating synthetic queries, yet they still rely on coarse partial-table sampling and simple fusion strategies, which limit semantic diversity and hinder effective query-table alignment. We propose STAR (Semantic Table Representation), a lightweight framework that improves semantic table representation through semantic clustering and weighted fusion. STAR first applies header-aware K-means clustering to group semantically similar rows and selects representative centroid instances to construct a diverse partial table. It then generates cluster-specific synthetic queries to comprehensively cover the table's semantic space. Finally, STAR employs weighted fusion strategies to integrate table and query embeddings, enabling fine-grained semantic alignment. This design enables STAR to capture complementary information from structured and textual sources, improving the expressiveness of table representations. Experiments on five benchmarks show that STAR achieves consistently higher Recall than QGpT on all datasets, demonstrating the effectiveness of semantic clustering and adaptive weighted fusion for robust table representation. Our code is available at https://github.com/adsl135789/STAR.

STAR: Semantic Table Representation with Header-Aware Clustering and Adaptive Weighted Fusion

TL;DR

STAR tackles the semantic gap in table retrieval by introducing a header-aware semantic clustering stage that yields diverse, representative table content and cluster-specific synthetic queries, followed by a weighted fusion stage that separately encodes and combines table and query signals. The framework replaces heuristic top-k sampling and simple concatenation with a principled two-stage process (SCQG and WF), using header information and adaptive weights to achieve finer-grained semantic alignment. Empirical results across five benchmarks show STAR consistently outperforms the QGpT baseline, with notable gains in Recall@1 and Recall@5, and demonstrate the benefits of dynamic weighting and cluster-guided query generation. The work highlights the importance of semantic coverage and source-aware fusion for robust, expressive table representations, with practical implications for scalable, query-driven table retrieval.

Abstract

Table retrieval is the task of retrieving the most relevant tables from large-scale corpora given natural language queries. However, structural and semantic discrepancies between unstructured text and structured tables make embedding alignment particularly challenging. Recent methods such as QGpT attempt to enrich table semantics by generating synthetic queries, yet they still rely on coarse partial-table sampling and simple fusion strategies, which limit semantic diversity and hinder effective query-table alignment. We propose STAR (Semantic Table Representation), a lightweight framework that improves semantic table representation through semantic clustering and weighted fusion. STAR first applies header-aware K-means clustering to group semantically similar rows and selects representative centroid instances to construct a diverse partial table. It then generates cluster-specific synthetic queries to comprehensively cover the table's semantic space. Finally, STAR employs weighted fusion strategies to integrate table and query embeddings, enabling fine-grained semantic alignment. This design enables STAR to capture complementary information from structured and textual sources, improving the expressiveness of table representations. Experiments on five benchmarks show that STAR achieves consistently higher Recall than QGpT on all datasets, demonstrating the effectiveness of semantic clustering and adaptive weighted fusion for robust table representation. Our code is available at https://github.com/adsl135789/STAR.
Paper Structure (25 sections, 8 equations, 1 figure, 2 tables)

This paper contains 25 sections, 8 equations, 1 figure, 2 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Methodology
  4. Semantic Clustering and Query Generation
  5. Header-aware K-means Clustering.
  6. Clustering-Guided Query Generation.
  7. Weighted Fusion
  8. Fixed Weight Fusion (FWF)
  9. Dynamic Weight Fusion (DWF)
  10. Experiments
  11. Experimental Setup
  12. Datasets
  13. Baselines
  14. Implementation
  15. Main Experimental Results
  16. ...and 10 more sections

Figures (1)

  • Figure 1: Overview of the STAR framework compared to the QGpT baseline. STAR improves representation via two stages: (1) replacing top-$k$ sampling with Header-aware Clustering for diverse instance selection and query generation, and (2) replacing simple concatenation with Weighted Fusion to explicitly model the importance of structured data and synthetic queries. Different colors denote different semantic row clusters.