Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Match, Compare, or Select? An Investigation of Large Language Models for Entity Matching

Tianshu Wang, Xiaoyang Chen, Hongyu Lin, Xuanang Chen, Xianpei Han, Hao Wang, Zhenyu Zeng, Le Sun

TL;DR

This work investigates how large language models can be used for entity matching in entity resolution by moving beyond independent pairwise decisions to leverage inter-record interactions. It analyzes three strategies—matcher, comparator, and selector—and introduces ComEM, a compound framework that combines local filtering with global selecting to balance accuracy and cost. Across eight ER datasets and ten LLMs, the study shows that interaction-based strategies substantially outperform independent matching, with the selector offering the strongest gains and ComEM delivering notable improvements while reducing computational cost. The findings highlight the importance of task decomposition and LLM composition for practical, cost-efficient EM in real-world data integration scenarios.

Abstract

Entity matching (EM) is a critical step in entity resolution (ER). Recently, entity matching based on large language models (LLMs) has shown great promise. However, current LLM-based entity matching approaches typically follow a binary matching paradigm that ignores the global consistency among record relationships. In this paper, we investigate various methodologies for LLM-based entity matching that incorporate record interactions from different perspectives. Specifically, we comprehensively compare three representative strategies: matching, comparing, and selecting, and analyze their respective advantages and challenges in diverse scenarios. Based on our findings, we further design a compound entity matching framework (ComEM) that leverages the composition of multiple strategies and LLMs. ComEM benefits from the advantages of different sides and achieves improvements in both effectiveness and efficiency. Experimental results on 8 ER datasets and 10 LLMs verify the superiority of incorporating record interactions through the selecting strategy, as well as the further cost-effectiveness brought by ComEM.

Match, Compare, or Select? An Investigation of Large Language Models for Entity Matching

TL;DR

This work investigates how large language models can be used for entity matching in entity resolution by moving beyond independent pairwise decisions to leverage inter-record interactions. It analyzes three strategies—matcher, comparator, and selector—and introduces ComEM, a compound framework that combines local filtering with global selecting to balance accuracy and cost. Across eight ER datasets and ten LLMs, the study shows that interaction-based strategies substantially outperform independent matching, with the selector offering the strongest gains and ComEM delivering notable improvements while reducing computational cost. The findings highlight the importance of task decomposition and LLM composition for practical, cost-efficient EM in real-world data integration scenarios.

Abstract

Entity matching (EM) is a critical step in entity resolution (ER). Recently, entity matching based on large language models (LLMs) has shown great promise. However, current LLM-based entity matching approaches typically follow a binary matching paradigm that ignores the global consistency among record relationships. In this paper, we investigate various methodologies for LLM-based entity matching that incorporate record interactions from different perspectives. Specifically, we comprehensively compare three representative strategies: matching, comparing, and selecting, and analyze their respective advantages and challenges in diverse scenarios. Based on our findings, we further design a compound entity matching framework (ComEM) that leverages the composition of multiple strategies and LLMs. ComEM benefits from the advantages of different sides and achieves improvements in both effectiveness and efficiency. Experimental results on 8 ER datasets and 10 LLMs verify the superiority of incorporating record interactions through the selecting strategy, as well as the further cost-effectiveness brought by ComEM.
Paper Structure (20 sections, 7 equations, 6 figures, 6 tables)

This paper contains 20 sections, 7 equations, 6 figures, 6 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Entity Resolution
  4. Large Language Model
  5. Entity Matching with LLMs
  6. Problem Formulation
  7. LLM as a Matcher
  8. LLM as a Comparator
  9. LLM as a Selector
  10. Compound Entity Matching Framework
  11. Experiments
  12. Experimental Setup
  13. Main Results
  14. Analysis of Strategies
  15. Effect of LLMs
  16. ...and 5 more sections

Figures (6)

  • Figure 1: Three strategies for LLM-based entity matching. We omit other attributes of records for simplicity.
  • Figure 2: Illustration of ComEM. It first filters candidate records by matching or comparing strategies and then identifies the match via the selecting strategy.
  • Figure 3: F1 score w.r.t. matching record positions.
  • Figure 4: Effect of open-source LLMs on different strategies and ComEM.
  • Figure 5: Ranking recall@1 w.r.t. model parameters.
  • ...and 1 more figures