Reproduce “Unlocking the Power of Large Language Models for Entity Alignment” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Xuhui Jiang, Yinghan Shen, Zhichao Shi, Chengjin Xu, Wei Li, Zixuan Li, Jian Guo, Huawei Shen and 1 more

ChatEA integrates large language models with knowledge representation learning for entity alignment, demonstrating improved accuracy, particularly on highly heterogeneous knowledge graphs like ICEWS-WIKI and ICEWS-YAGO, where it outperformed previous methods by up to 16% Hits@1. The framework enhances entity understanding by leveraging LLMs' background knowledge and providing explainable alignment decisions.

Key insights

  • The integration of LLMs' extensive background knowledge and advanced reasoning capabilities effectively addresses the limitations of traditional KRL methods, especially for complex, heterogeneous knowledge graphs.
  • Translating structured knowledge graph data into an executable 'code format' is an effective approach for enabling LLMs to understand and reason over graph-structured information.
  • An adaptive, iterative two-stage strategy balances the computational demands of LLMs with the need for high accuracy, allowing for efficient yet thorough reasoning across diverse datasets.
  • The framework exhibits robustness to noisy initial KRL embeddings, demonstrating that LLM reasoning can mitigate imperfections from underlying embedding methods.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.