Improving Table Understanding with LLMs and Entity-Oriented Search

2025-08-23

AI Summary

Existing table understanding methods rely heavily on manual preprocessing and keyword-based matching, suffering from inefficient large language model (LLM) reasoning due to the absence of structured contextual representations. To address this, we propose an entity-centric table understanding framework: first constructing a table entity graph to explicitly encode semantic entities and their interrelations; then designing an entity-guided semantic search mechanism that jointly models semantic similarity and implicit cell-level relationships. Notably, we introduce Graph Query Language (GQL) into table understanding for the first time, enabling interpretable, context-aware reasoning over tabular structures. Our approach significantly reduces dependence on heuristic preprocessing and surface-level pattern matching. Evaluated on WikiTableQuestions and TabFact benchmarks, it achieves state-of-the-art performance. This work establishes a novel paradigm for table semantic understanding—synergistically leveraging graph-structured representations and fine-grained entity semantics—thereby advancing robust, explainable, and context-sensitive tabular reasoning.

Technology Category

• Natural Language Processing: Question Answering
• Knowledge Representation and Reasoning: Computational Complexity of Reasoning
• Data Mining & Knowledge Management: Intelligent Query Processing

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Search and Retrieval-Augmented AI: Web query analysis, representation and understanding

Abstract

Our work addresses the challenges of understanding tables. Existing methods often struggle with the unpredictable nature of table content, leading to a reliance on preprocessing and keyword matching. They also face limitations due to the lack of contextual information, which complicates the reasoning processes of large language models (LLMs). To overcome these challenges, we introduce an entity-oriented search method to improve table understanding with LLMs. This approach effectively leverages the semantic similarities between questions and table data, as well as the implicit relationships between table cells, minimizing the need for data preprocessing and keyword matching. Additionally, it focuses on table entities, ensuring that table cells are semantically tightly bound, thereby enhancing contextual clarity. Furthermore, we pioneer the use of a graph query language for table understanding, establishing a new research direction. Experiments show that our approach achieves new state-of-the-art performances on standard benchmarks WikiTableQuestions and TabFact.

Problem

• Addresses challenges in understanding unpredictable table content
• Overcomes limitations from lack of contextual information in tables
• Reduces reliance on preprocessing and keyword matching methods

Innovation

• Entity-oriented search method for table understanding
• Leveraging semantic similarities and implicit relationships
• Pioneering graph query language for tables