Knowing the No-match: Entity Alignment with Dangling Cases

Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing, pages 3582–3593 August 1–6, 2021. ©2021 Association for Computational Linguistics 3582 Knowing the No-match: Entity Alignment with Dangling Cases Zequn Sun1 Muhao Chen2,3 Wei Hu1 1State Key Laboratory for Novel Software Technology, Nanjing University, China 2Department of Computer Science, University of Southern California, USA 3Information Sciences Institute, University of Southern California, USA zqsun.nju@gmail.com, muhaoche@usc.edu, whu@nju.edu.cn Abstract This paper studies a new problem setting of entity alignment for knowledge graphs (KGs). Since KGs possess different sets of entities, there could be entities that cannot find align- ment across them, leading to the problem of dangling entities. As the first attempt to this problem, we construct a new dataset and de- sign a multi-task learning framework for both entity alignment and dangling entity detection. The framework can opt to abstain from pre- dicting alignment for the detected dangling en- tities. We propose three techniques for dan- gling entity detection that are based on the distribution of nearest-neighbor distances, i.e., nearest neighbor classification, marginal rank- ing and background ranking. After detecting and removing dangling entities, an incorpo- rated entity alignment model in our framework can provide more robust alignment for remain- ing entities. Comprehensive experiments and analyses demonstrate the effectiveness of our framework. We further discover that the dan- gling entity detection module can, in turn, im- prove alignment learning and the final perfor- mance. The contributed resource is publicly available to foster further research. 1 Introduction Knowledge graphs (KGs) have evolved to be the building blocks of many intelligent systems (Ji et al., 2020). Despite the importance, KGs are usually costly to construct (Paulheim, 2018) and naturally suffer from incompleteness (Gal´arraga et al., 2017). Hence, merging multiple KGs through entity alignment can lead to mutual enrichment of their knowledge (Chen et al., 2020), and provide downstream applications with more comprehensive knowledge representations (Trivedi et al., 2018; Chen et al., 2020). Entity alignment seeks to dis- cover identical entities in different KGs, such as En- glish entity Thailand and its French counterpart source KG entities target KG entities dangling entities in the target KG dangling entities in the source KG Figure 1: Illustration of entity alignment between two KGs with dangling cases. Paired red and black squares in the overlap region denote entity alignment while oth- ers are dangling entities without counterparts. Tha¨ılande. To tackle this important problem, literature has attempted with the embedding-based entity alignment methods (Chen et al., 2017; Wang et al., 2018; Cao et al., 2019; Fey et al., 2020; Wu et al., 2020a; Liu et al., 2020; Sun et al., 2020a). These methods jointly embed different KGs and put similar entities at close positions in a vector space, where the nearest neighbor search can re- trieve entity alignment. Due to its effectiveness, embedding-based entity alignment has drawn ex- tensive attention in recent years (Sun et al., 2020c). Nonetheless, to practically support the alignment of KGs as a …

Content truncated to 500 words