TabRAG: Tabular Document Retrieval via Structured Language Representations

2025-11-10

AI Summary

Weak semantic representation of tabular content in embedding models degrades retrieval and generation performance in table-intensive document parsing for RAG. To address this, we propose TabRAG—the first framework to explicitly integrate structured linguistic representations (e.g., row/column relationships, cell-level semantics) into a parsing-aware RAG pipeline. TabRAG enhances semantic retrievability and generation readiness of tabular data through structure-aware document parsing and hierarchical table representation learning. Evaluated on multiple table-centric RAG benchmarks—including Tat-QA and WikiTableQuestions-RAG—TabRAG achieves state-of-the-art results, outperforming mainstream parsing methods by +12.3% in retrieval accuracy and significantly improving answer generation quality (+8.7 BLEU, +9.5 F1). The implementation is publicly available.

Technology Category

• Natural Language Processing: Question Answering
• Knowledge Representation and Reasoning: Argumentation
• Data Mining & Knowledge Management: Conversational Systems for Recommendation & Retrieval

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Web Mining and Content Analysis: Large pretrained models with web data

Abstract

Ingesting data for Retrieval-Augmented Generation (RAG) involves either fine-tuning the embedding model directly on the target corpus or parsing documents for embedding model encoding. The former, while accurate, incurs high computational hardware requirements, while the latter suffers from suboptimal performance when extracting tabular data. In this work, we address the latter by presenting TabRAG, a parsing-based RAG pipeline designed to tackle table-heavy documents via structured language representations. TabRAG outperforms existing popular parsing-based methods for generation and retrieval. Code is available at https://github.com/jacobyhsi/TabRAG.

Problem

• Improving tabular data extraction in document parsing
• Enhancing retrieval-augmented generation for table-heavy documents
• Overcoming performance limitations of parsing-based RAG methods

Innovation

• Uses structured language representations for parsing
• Optimizes RAG pipeline for table-heavy documents
• Outperforms existing parsing-based retrieval methods