Cross-lingual Knowledge Graph Alignment via Graph Matching Neural Network

Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 3156–3161 Florence, Italy, July 28 - August 2, 2019. c⃝2019 Association for Computational Linguistics 3156 Cross-lingual Knowledge Graph Alignment via Graph Matching Neural Network Kun Xu1, Liwei Wang1, Mo Yu2, Yansong Feng3, Yan Song1, Zhiguo Wang4, Dong Yu1 1Tencent AI Lab 2IBM T.J. Watson Research 3Peking University 4Amazon AWS {syxu828,wlwsjtu1989,zgw.tomorrow}@gmail.com yum@us.ibm.com, fengyansong@pku.edu.cn, {clksong,dyu}@tencent.com Abstract Previous cross-lingual knowledge graph (KG) alignment studies rely on entity embeddings derived only from monolingual KG structural information, which may fail at matching enti- ties that have different facts in two KGs. In this paper, we introduce the topic entity graph, a local sub-graph of an entity, to represent en- tities with their contextual information in KG. From this view, the KB-alignment task can be formulated as a graph matching problem; and we further propose a graph-attention based so- lution, which first matches all entities in two topic entity graphs, and then jointly model the local matching information to derive a graph- level matching vector. Experiments show that our model outperforms previous state-of-the- art methods by a large margin. 1 Introduction Multilingual knowledge graphs (KGs), such as DBpedia (Auer et al., 2007) and Yago (Suchanek et al., 2007), represent human knowledge in the structured format and have been successfully used in many natural language processing applications. These KGs encode rich monolingual knowledge but lack the cross-lingual links to bridge the lan- guage gap. Therefore, the cross-lingual KG align- ment task, which automatically matches entities in a multilingual KG, is proposed to address this problem. Most recently, several entity matching based approaches (Hao et al., 2016; Chen et al., 2016; Sun et al., 2017; Wang et al., 2018) have been proposed for this task. Generally, these ap- proaches first project entities of each KG into low- dimensional vector spaces by encoding monolin- gual KG facts, and then learn a similarity score function to match entities based on their vec- tor representations. However, since some enti- ties in different languages may have different KG KG-1 KG-2 Matched e0 e0’ Figure 1: A challenging entity matching example. facts, the information encoded in entity embed- dings may be diverse across languages, making it difficult for these approaches to match these enti- ties. Figure 1 illustrates such an example where we aim to align e0 with e′ 0, but there is only one aligned neighbor in their surrounding neighbors. In addition, these methods do not encode the en- tity surface form into the entity embedding, also making it difficult to match entities that have few neighbors in the KG that lacks sufficient structural information. To address these drawbacks, we propose a topic entity graph to represent the KG context informa- tion of an entity. Unlike previous methods that uti- lize entity embeddings to match entities, we for- mulate this task as a graph matching problem be- tween the topic entity graphs. To achieve this, we propose a novel graph matching method to esti- mate the similarity of two …

Content truncated to 500 words