Neighborhood-Aware Attentional Representation for Multilingual Knowledge Graphs Qiannan Zhu 1,2 , Xiaofei Zhou *1,2 , Jia Wu 3 , Jianlong Tan 1,2 and Li Guo 1,2 1 Institute of Information Engineering, Chinese Academy of Sciences, Beijing, China 2 School of Cyber Security, University of Chinese Academy of Sciences, Beijing, China 3 Department of Computing, Macquarie University, Sydney, Australia {zhouxiaofei,zhuqiannan}@iie.ac.cn,Jia.Wu@mq.edu.au Abstract Multilingual knowledge graphs constructed by en- tity alignment are the indispensable resources for numerous AI-related applications. Most existing entity alignment methods only use the triplet-based knowledge to find the aligned entities across mul- tilingual knowledge graphs, they usually ignore the neighborhood subgraph knowledge of entities that implies more richer alignment information for aligning entities. In this paper, we incorporate neighborhood subgraph-level information of enti- ties, and propose a neighborhood-aware attention- al representation method NAEA for multilingual knowledge graphs. NAEA devises an attention mechanism to learn neighbor-level representation by aggregating neighbors’ representations with a weighted combination. The attention mechanism enables entities not only capture different impact- s of their neighbors on themselves, but also attend over their neighbors’ feature representations with different importance. We evaluate our model on t- wo real-world datasets DBP15K and DWY100K, and the experimental results show that the proposed model NAEA significantly and consistently outper- forms state-of-the-art entity alignment models. 1 Introduction The multilingual knowledge graphs (KGs) like YAGO [Suchanek et al., 2008] and DBpedia [Bizer et al., 2009] increasingly play an significant role in supporting vari- ous knowledge-driven tasks. Those multilingual knowledge graphs consist of monolingual knowledge, in forms of direct- ed graphs, where entities are represented as nodes and re- lations as edges. The monolingual knowledge are stored as triplets (e h , r, e t ), representing that the head entity e h and tail entity e t are linked by relation r. Besides monolingual knowl- edge, the multilingual KGs also embody cross-lingual knowl- edge (e h , align(),e ′ h ) that matches the same real-world en- tities e h and e ′ h among different human languages L and L ′ by alignment operation align(), see Figure 1. A great deal of methods focus on exploiting monolingual knowledge ∗ Corresponding Author KG KG’ align align h e h e' t e r Figure 1: Multilingual Knowledge Graphs. KG and KG ′ are the knowledge graphs of languages L and L ′ . graphs in recent years. Particularly, the embedding-based methods that encode entities and relations into continue low- dimensional vector spaces, have achieved promising perfor- mance. Exemplarily, given a triplet (e h , r, e t ), TransE [Bor- des et al., 2013] regards the relation embedding r as the trans- lation vector between the head and tail entity embedding e h and e t , and expects e h + r ≈ e t when (e h , r, e t ) holds. Other extended works such as TransH [Wang et al., 2014], TransR [Lin et al., 2015a] and TransD [Ji et al., 2015] also emerged with different translation forms in characterizing relation r. However, a few methods have been done for modeling multi- lingual knowledge graphs. Entity alignment is an effective way to integrate the mul- tilingual KGs, which is the task of finding the same real- world entities in different KGs. The traditional multilingual entity alignment methods mainly based on machine trans- lation, have low accuracy due to the poor performance in translation between multiple languages. Most recently, fol- lowing above popular embedding-based models, MTransE [Chen et al., 2017] provides the cross-lingual transitions for both entities and relations across different knowledge graph embeddings. IPTransE [Zhu et al., 2017] jointly encodes both entities and relations of various KGs into an unified low-dimensional semantic space via sharing parameters on a seed set of aligned entities, JAPE [Sun et al., 2017] fur- ther incorporates attribute triplets as additional information for learning KGs’ embeddings in an unified space. BootEA [Sun et al., 2018] adopts bootstrapping [Yarowsky, 1995; Abney, 2004] approach to iteratively label likely entity align- ment as training data and leverage it for learning alignment- oriented embeddings. Existing entity alignment methods on- ly use the triplet-based information, but ignore the inheren- t neighborhood information of entities for aligning entities. Proceedings of the Twenty-Eighth International Joint Conference on Artificial Intelligence (IJCAI-19) 1943