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

Forum topic · 小凯 · 2026-07-05

Summary

This forum post summarizes a survey titled "Large Language Models for Table Processing," published in January 2025 in Frontiers of Computer Science (Springer). The survey systematically reviews how large language models (LLMs) handle structured tabular data, a domain that differs sharply from free-form text because tables demand precise cell-level grounding, schema understanding, and multi-hop reasoning across rows and columns. The post situates the work within the broader evolution of neural information retrieval and LLM integration, covering paradigms such as dense retrieval, late interaction (e.g., ColBERT), generative information retrieval, and agentic search, along with retrieval-augmented generation (RAG) and tool-use patterns. It outlines evaluation practices (MS MARCO, BEIR, Natural Questions; nDCG, MRR, LLM-as-judge) and highlights open problems: benchmark reproducibility, hallucination and safety, latency and cost constraints, cross-lingual and multimodal extension, and the gap between offline metrics and real user satisfaction. The post also provides an engineering checklist covering data privacy, latency budgets, quality validation, security against retrieval poisoning, and token/GPU cost control. Related cross-references include several LLM-based Text-to-SQL surveys and function-calling database querying work. Readers are advised to verify quantitative claims against the original PDF.

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

Overview

| Field | Content | |------|------| | Title | Large language model for table processing: a survey | | Venue | Frontiers of Computer Science, January 2025 | | Link | https://link.springer.com/article/10.1007/s11704-024-40763-6 | | Type | Survey | | Section | Querying Structured Information |

Context and Motivation

Tables are ubiquitous structured data, but they pose unique challenges for LLMs: precise cell-level grounding, schema understanding, heterogeneous layouts, and multi-hop reasoning across rows and columns. Traditional pipelines separate retrieval, ranking, and generation, which struggles in the LLM era where users expect natural-language interaction and integrated reasoning over structured knowledge. This survey systematically organizes research on applying LLMs to table processing.

Core Contributions

Provides a unified taxonomy that organizes scattered related work into a comparable framework.
Decomposes method components (representation learning, retrievers, rerankers, planners, generators, feedback mechanisms) for engineering adoption.
Offers reproducible benchmarks, datasets, and classification tables, lowering the entry cost for new researchers.
Discusses interfaces with LLM tool calling, reinforcement learning, and multi-agent collaboration.
Lists open problems: evaluation reliability, latency and cost, hallucination and safety, cross-lingual and multimodal extension.

Landscape of Approaches

The survey situates table processing within four main lines of LLM-era information access:

Dense retrieval: high recall and low latency, suited for first-stage retrieval;
Late interaction (e.g., ColBERT): higher precision but larger indexes;
Generative IR: directly "generates" documents via tokens or docids, simplifying cascades;
Agentic search: models search as sequential decision-making, supporting multi-hop and self-reflection.

Timeline: 2019–2021 BERT reranking and DPR established neural retrieval; 2022–2023 RAG pushed retrieval–generation fusion; from 2024 onward, conversational/agentic search and generative recommendation exploded; 2025–2026 trends include RL-trained search agents, Deep Research, and GraphRAG.

Evaluation Practices

Datasets: MS MARCO, BEIR, Natural Questions, domain corpora;
Metrics: nDCG@10, MRR, Recall@k, Hit@k, human preference, task success rate, latency and token cost;
Baselines: BM25, dense retrieval, cross-encoder reranking, retrieval-free LLMs, commercial search APIs.

Key Takeaways

1. Architecture: cascaded retrieve–rerank–generate remains mainstream, but agentic paradigms make retrieval strategy itself learnable. 2. Data: high-quality instruction data and interaction logs are critical; synthetic data risks leakage and distribution shift. 3. Evaluation: offline metrics diverge from online satisfaction; LLM-as-judge needs human cross-validation. 4. Productization: latency, cost, explainability, and safety are hard constraints in deployment.

Engineering Checklist

| Item | Question | Recommendation | |--------|------|------| | Data | PII in training/index? Versioning? | Partitioned indexes, anonymization, rollback-capable embeddings | | Latency | p99 budget? Retrieval steps? | Cascade + early exit, query caching, async reranking | | Quality | Do offline gains translate online? | Interleaving experiments, human audits, citation verification | | Security | Poisoning/bias via open retrieval? | Source whitelists, adversarial detection, output filtering | | Cost | Token and GPU cost per query? | Small-model routing, distillation, hybrid sparse+dense |

Related Entries

A Survey of Text-to-SQL in the Era of LLMs (arXiv 2408.05109)
A Survey on Employing Large Language Models for Text-to-SQL Tasks (ACM, 10.1145/3737873)
Next-Generation Database Interfaces: A Survey of LLM-based Text-to-SQL (arXiv 2406.08426)
Querying Databases with Function Calling, Jan 2025 (arXiv 2502.00032)

References

Original paper: Large language model for table processing: a survey. Frontiers of Computer Science, Jan 2025. https://link.springer.com/article/10.1007/s11704-024-40763-6