Entity matching (EM) is a critical step in entity resolution. 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 between different
records. 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 compositional entity matching (ComEM) framework that leverages
the composition of multiple strategies and LLMs. In this way, ComEM can benefit
from the advantages of different sides and achieve improvements in both
effectiveness and efficiency. Experimental results show that ComEM not only
achieves significant performance gains on various datasets but also reduces the
cost of LLM-based entity matching in real-world application.

基于大型语言模型的实体匹配方法通常忽略不同记录之间的整体一致性。本文研究了多种基于大型语言模型的实体匹配方法，将不同视角的记录交互纳入考量，并综合比较了三种具有代表性的策略：匹配、比较和选择。基于实验结果，我们设计了一个组合式的实体匹配框架（ComEM），该框架借鉴了多种策略和大型语言模型的优势，提高了实体匹配的效果和效率，在不同数据集上均取得了显著的性能提升，并降低了基于大型语言模型实体匹配的成本。