Conceptual Schema Inference for Tabular Datasets using Large Language Models

2026-05-21

AI Summary

This work addresses the challenge of automatically constructing coherent conceptual schemas from heterogeneous tabular data, where inconsistent representations impede schema inference. To overcome this, the authors propose two novel approaches: GeSI leverages large generative language models to infer a global hierarchical conceptual schema using only column names and cell values, while EmSI discovers shared structural attributes through clustering of table embeddings. By integrating semantic analysis with schema consolidation, both methods generate high-quality, concise, and scalable conceptual schemas without requiring manual annotations. Experimental results demonstrate that the inferred schemas exhibit strong structural coherence and parsimony, and their effectiveness is further validated in end-to-end downstream applications.

Abstract

Large collections of tabular data from data lakes, web tables and open data portals often originate from heterogeneous sources, leading to representational inconsistencies. Understanding and organizing such repositories therefore remains a major challenge. While prior work has primarily focused on dataset discovery and exploration, this paper addresses the complementary problem of conceptual schema inference: automatically deriving a conceptual schema that captures entity types, attributes and inter-type relationships directly from raw tables. We propose two large language model (LLM)-based approaches that use only column headers and cell values: GeSI uses generative LLMs to infer hierarchical types and their attributes from table- and column-level semantics, and to integrate them into a global schema that also captures relationships across types; EmSI employs LLM-based table embeddings to group tables by column-level semantics, infer attributes within each group, and construct hierarchical structures from shared attribute patterns. Finally, we report an experimental analysis demonstrating the effectiveness of our approaches in terms of the conciseness and structural quality of the inferred schema components, their scalability to large repositories, and a case study illustrating end-to-end schema inference.

Problem

• conceptual schema inference
• tabular datasets
• heterogeneous data sources
• entity types
• schema discovery

Innovation

• Conceptual Schema Inference
• Large Language Models
• Tabular Data
• Schema Integration
• Table Embeddings