From Alignment to Assignment: Frustratingly Simple Unsupervised Entity Alignment

Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 2843–2853 November 7–11, 2021. c⃝2021 Association for Computational Linguistics 2843 From Alignment to Assignment: Frustratingly Simple Unsupervised Entity Alignment Xin Mao1, Wenting Wang2, Yuanbin Wu1, Man Lan1∗ 1 School of Computer Science and Technology, East China Normal University Lazada, Alibaba Group xmao@stu.ecnu.edu.cn,wenting.wang@lazada.com {ybwu,mlan}@cs.ecnu.edu.cn Abstract Cross-lingual entity alignment (EA) aims to find the equivalent entities between cross- lingual KGs (Knowledge Graphs), which is a crucial step for integrating KGs. Recently, many GNN-based EA methods are proposed and show decent performance improvements on several public datasets. However, exist- ing GNN-based EA methods inevitably inherit poor interpretability and low efficiency from neural networks. Motivated by the isomorphic assumption of GNN-based methods, we suc- cessfully transform the cross-lingual EA prob- lem into an assignment problem. Based on this re-definition, we propose a frustratingly Sim- ple but Effective Unsupervised entity align- ment method (SEU) without neural networks. Extensive experiments have been conducted to show that our proposed unsupervised ap- proach even beats advanced supervised meth- ods across all public datasets while having high efficiency, interpretability, and stability. 1 Introduction The knowledge graph (KG) represents a collection of interlinked descriptions of real-world objects and events, or abstract concepts (e.g., documents), which has facilitated many downstream applica- tions, such as recommendation systems (Cao et al., 2019b; Wang et al., 2019) and question-answering (Zhao et al., 2020; Qiu et al., 2020). Over re- cent years, a large number of KGs are constructed from different domains and languages by different organizations. These cross-lingual KGs usually hold unique information individually but also share some overlappings. Integrating these cross-lingual KGs could provide a broader view for users, espe- cially for the minority language users who usually suffer from lacking language resources. Therefore, how to fuse the knowledge from cross-lingual KGs has attracted increasing attentions. As shown in Figure 1, cross-lingual entity align- ment (EA) aims to find the equivalent entities Marvel USA Robert Downey Jr. Iron Man Spider Man [own by] [born in] [friend] 漫威 美国 钢铁侠 美国队长 [own by] [born in] [friend] [act] SAME? KGCN KGEN Figure 1: An example of cross-lingual knowledge graph entity alignment. across multi-lingual KGs, which is a crucial step for integrating KGs. Conventional methods (Suchanek et al., 2011; Jiménez-Ruiz and Grau, 2011) usually solely rely on lexical matching and probability rea- soning, which requires machine translation systems to solve cross-lingual tasks. However, existing ma- chine translation systems are not able to achieve high accuracy with limited contextual information, especially for language pairs that are not alike, such as Chinese-English and Japanese-English. Recently, Graph Convolutional Network (GCN) (Kipf and Welling, 2017) and subsequent Graph Neural Network (GNN) variants have achieved state-of-the-art results in various graph application. Intuitively, GNN is better in capturing structural in- formation of KGs to compensate for the shortcom- ing of conventional methods. Specifically, several GNN-based EA methods (Xu et al., 2019; Wu et al., 2019a; Wang et al., 2020) indeed demonstrate de- cent performance improvements on public datasets. All …

Content truncated to 500 words