Semi-supervised Entity Alignment via Joint Knowledge Embedding Model and Cross-graph Model

Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing, pages 2723–2732, Hong Kong, China, November 3–7, 2019. c⃝2019 Association for Computational Linguistics 2723 Semi-supervised Entity Alignment via Joint Knowledge Embedding Model and Cross-graph Model Chengjiang Li1,2,3, Yixin Cao4∗, Lei Hou1,2,3, Jiaxin Shi1,2,3, Juanzi Li1,2,3, Tat-Seng Chua4 1Department of Computer Science and Technology, Tsinghua University 2KIRC, Institute for Artificial Intelligence, Tsinghua University 3Beijing National Research Center for Information Science and Technology 4School of Computing, National University of Singapore, Singapore {iamlockelightning,caoyixin2011,greener2009,shijx12}@gmail.com lijuanzi2008@gmail.com, dcscts@nus.edu.sg Abstract Entity alignment aims at integrating comple- mentary knowledge graphs (KGs) from dif- ferent sources or languages, which may ben- efit many knowledge-driven applications. is challenging due to the heterogeneity ofIt KGs and limited seed alignments. In this paper, we propose a semi-supervised en- tity alignment method by joint Knowledge Embedding model and Cross-Graph model (KECG). It can make better use of seed align- ments to propagate over the entire graphs with KG-based constraints. Specifically, as for the knowledge embedding model, we utilize TransE to implicitly complete two KGs to- wards consistency and learn relational con- straints between entities. As for the cross- graph model, we extend Graph Attention Net- work (GAT) with projection constraint to ro- bustly encode graphs, and two KGs share the same GAT to transfer structural knowledge as well as to ignore unimportant neighbors for alignment via attention mechanism. Results on publicly available datasets as well as fur- ther analysis demonstrate the effectiveness of KECG. Our codes can be found in https: //github.com/THU-KEG/KECG. 1 Introduction Recently, many Knowledge Graphs (KGs) (e.g., DBpedia (Lehmann et al., 2015), YAGO (Rebele et al., 2016) and BabelNet (Navigli and Ponzetto, 2012)) have emerged to provide structural knowl- edge for different applications. These separately constructed KGs contain heterogeneous but com- plementary contents; thus integrating KGs from different sources or languages into a unified KG becomes essential to better benefit knowledge- driven applications, ranging from information ex- traction (Han et al., 2018; Cao et al., 2018) to ques- tion answering (Cui et al., 2017). ∗ Corresponding author KG2 KG1 country New York City Hudson River New York Hudson River New York (state) U.S. state the U.S. adjacentTo locatedIn hasRiver adjoin country country Figure 1: Example of the New York City entity and its neighbors in two heterogeneous KGs (Dashed circles in the same color means prior alignments). Some efforts have been made to integrate KGs by aligning entities with their semantically same counterparts, namely entity alignment. Early en- tity alignment approaches either rely on human efforts (Mahdisoltani et al., 2013) or extra re- sources (Hu et al., 2011; Wang et al., 2013) to dis- cover new equivalent entity pairs. Recent embed- ding approaches show significant improvement on entity alignment. We roughly classify these em- bedding models into two groups: KG-based mod- els and graph-based models. KG-based mod- els (Hao et al., 2016; Chen et al., 2017; Sun et al., 2017; Zhu et al., 2017; Sun et al., 2018; …

Content truncated to 500 words