Entity alignment (EA) aims to find equivalent entities between two Knowledge Graphs. Existing embedding-based EA methods usually encode entities as embeddings, triples as embeddings' constraint and learn to align the embeddings. The structural and side information are usually utilized via embedding propagation, aggregation or interaction. However, the details of the underlying logical inference steps among the alignment process are usually omitted, resulting in inadequate inference process. In this paper, we introduce P-NAL, an entity alignment method that captures two types of logical inference paths with Non-Axiomatic Logic (NAL). Type 1 is the bridge-like inference path between to-be-aligned entity pairs, consisting of two relation/attribute triples and a similarity sentence between the other two entities. Type 2 links the entity pair by their embeddings. P-NAL iteratively aligns entities and relations by integrating the conclusions of the inference paths. Moreover, our method is logically interpretable and extensible due to the expressiveness of NAL. Our proposed method is suitable for various EA settings. Experimental results show that our method outperforms state-of-the-art methods in terms of Hits@1, achieving 0.98+ on all three datasets of DBP15K with both supervised and unsupervised settings. To our knowledge, we present the first in-depth analysis of entity alignment's basic principles from a unified logical perspective.

翻译：实体对齐旨在发现两个知识图谱间的等价实体。现有基于嵌入的实体对齐方法通常将实体编码为嵌入向量、三元组编码为嵌入约束，并通过学习实现嵌入对齐。结构信息与辅助信息通常通过嵌入传播、聚合或交互加以利用。然而，对齐过程中潜在的逻辑推理步骤细节常被忽略，导致推理过程不充分。本文提出P-NAL实体对齐方法，利用非公理逻辑捕获两类逻辑推理路径：第一类为待对齐实体对间的桥梁式推理路径，由两个关系/属性三元组及另两个实体间的相似度语句构成；第二类通过实体嵌入实现实体对关联。P-NAL通过整合推理路径结论，迭代式对齐实体与关系。此外，由于非公理逻辑的表达能力，该方法具有逻辑可解释性与可扩展性。本方法适用于多种实体对齐场景。实验结果表明，在有监督与无监督设置下，本方法在DBP15K三个数据集上的Hits@1均达到0.98+，优于现有最优方法。据我们所知，本文首次从统一逻辑视角对实体对齐的基本原理进行了深入分析。