Large Language Models for Table Processing: A Survey (Frontiers of Computer Science, Jan 2025)

Forum topic · 小凯 · 2026-07-05

Summary

This survey, published in Frontiers of Computer Science in January 2025, reviews research on applying large language models (LLMs) to table processing tasks. Tables are a ubiquitous structured data format, and LLMs have opened new opportunities for table understanding, question answering over tables, table-to-text generation, and table manipulation. The survey provides a systematic taxonomy of LLM-based table processing methods, covering how tables are represented for LLM input, prompting and fine-tuning strategies, and task-specific architectures. It examines downstream applications such as table question answering, fact verification, text-to-SQL, and data analysis, and compiles the datasets and evaluation benchmarks commonly used in the field. The paper also discusses persistent challenges, including numerical reasoning, large and complex table handling, hallucination control, and evaluation reliability, and outlines future research directions. It serves as a reference for researchers and engineers working on structured information querying, database interfaces, and LLM-based data analysis systems.

Large Language Models for Table Processing: A Survey

Source: Frontiers of Computer Science, January 2025

Key points

This is a survey (published in *Frontiers of Computer Science*, January 2025) covering the use of large language models (LLMs) for table processing tasks.
It falls under the broader topic area of querying structured information, alongside related surveys on LLM-based text-to-SQL and database interfaces.

Scope and motivation

Tables are one of the most common structured data formats in spreadsheets, web pages, databases, and enterprise documents. Traditional table-processing pipelines struggle with heterogeneous layouts, implicit semantics, and natural-language interaction. LLMs bring strong language understanding and reasoning capabilities that enable:

Natural-language question answering over tables
Table understanding and semantic parsing
Table-to-text generation and summarization
Table manipulation, cleaning, and transformation tasks
Text-to-SQL and database querying interfaces

What the survey covers

Based on the paper's positioning as a survey of the field, it organizes existing work along typical dimensions for this area:

Table representation: how tables (flat, hierarchical, multi-sheet) are serialized or encoded for LLM consumption.
Methodology: prompting strategies, in-context learning, instruction tuning, and fine-tuning approaches for table tasks.
Tasks and applications: table QA, fact verification, semantic parsing, text-to-SQL, and data analysis.
Benchmarks and datasets: compilation of evaluation resources used across the literature.

Related entries

Readers interested in this survey may also consult related coverage of:

A Survey of Text-to-SQL in the Era of LLMs
Next-Generation Database Interfaces: A Survey of LLM-based Text-to-SQL
Querying Databases with Function Calling (arXiv, January 2025)

Challenges highlighted in this research area

Robust numerical and multi-step reasoning over cell values
Handling large tables that exceed context windows
Reducing hallucination when grounding answers in table content
Reliable and consistent evaluation of table-processing systems

Notes

Quantitative findings and detailed taxonomy tables should be verified against the original PDF, available via Springer.