Scholarlink

The Semantic Table Annotation (STA) task involving Column Type Annotation (CTA) and Cell Entity Annotation (CEA) tasks, maps table contents to ontology entities, playing important roles in various semantic applications. However, complex tables often pose challenges such as semantic loss of column names or cell values, strict ontological hierarchy annotation, homonyms, spelling errors, abbreviations, which hinder the accuracy of annotation. To tackle these issues, this paper proposes an LLM-based agent approach for CTA and CEA tasks. We design and implement five external tools with tailored prompts based on the ReAct framework, enabling the STA agent to dynamically select suitable annotation strategies based on different table characteristics. The experiments are conducted on the Tough Tables and BiodivTab datasets related to the aforementioned challenges from the SemTab challenge, where it outperforms existing methods in various metrics. Furthermore, by using Levenshtein distance to reduce redundant annotations, we achieve a 70% reduction in time costs and a 60% reduction in LLM token usage, providing an efficient, cost-effective solution for STA task.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

An LLM Agent-Based Complex Semantic Table Annotation Approach

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Yilin Geng,
  • Shujing Wang,
  • Chuan Wang,
  • Keqing He,
  • Yanfei Lv,
  • Ying Wang,
  • Zaiwen Feng,
  • Xiaoying Bai

摘要

The Semantic Table Annotation (STA) task involving Column Type Annotation (CTA) and Cell Entity Annotation (CEA) tasks, maps table contents to ontology entities, playing important roles in various semantic applications. However, complex tables often pose challenges such as semantic loss of column names or cell values, strict ontological hierarchy annotation, homonyms, spelling errors, abbreviations, which hinder the accuracy of annotation. To tackle these issues, this paper proposes an LLM-based agent approach for CTA and CEA tasks. We design and implement five external tools with tailored prompts based on the ReAct framework, enabling the STA agent to dynamically select suitable annotation strategies based on different table characteristics. The experiments are conducted on the Tough Tables and BiodivTab datasets related to the aforementioned challenges from the SemTab challenge, where it outperforms existing methods in various metrics. Furthermore, by using Levenshtein distance to reduce redundant annotations, we achieve a 70% reduction in time costs and a 60% reduction in LLM token usage, providing an efficient, cost-effective solution for STA task.