Reproduce “Relation-Aware Entity Alignment for Heterogeneous Knowledge Graphs” 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

Yuting Wu, Xiao Liu, Yansong Feng, Zheng Wang, Rui Yan, Dongyan Zhao

Peking University researchers developed Relation-aware Dual-Graph Convolutional Network (RDGCN), a model that learns entity representations by modeling attentive interactions between entities and a novel dual graph of relations. RDGCN achieves state-of-the-art entity alignment performance on cross-lingual knowledge graphs, demonstrating robustness even with limited training data.

Key insights

  • Modeling relationships between relations in a 'dual' graph, and iteratively refining entity and relation representations through attentive interactions, effectively integrates complex relation information into entity embeddings.
  • Approximating relation representations from aggregated entity features involved in a relation avoids the parameter explosion issue of R-GCNs while retaining relation-awareness.
  • Combining dual-primal attention with GCN layers and highway gates allows for comprehensive capture of both fine-grained relation semantics and broader structural context, leading to more discriminative entity representations.

Summary by alphaXiv.

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