Global entity alignment with Gated Latent Space Neighborhood Aggregation

Global entity alignment with Gated Latent Space Neighborhood Aggregation Wei Chen School of Computer, Wuhan University of Technology Wuhan, China chenw720@foxmail.com Xiaoying Chen∗ School of Computer, Wuhan University of Technology Hubei Credit Information Center Wuhan, China 13971520707@139.com Shengwu Xiong School of Computer, Wuhan University of Technology Wuhan, China Sanya Science and Education Innovation Park of Wuhan University of Technology Sanya xiongsw@whut.edu.cn Abstract Existing entity alignment models mainly use the topology structure of the original knowledge graph and have achieved promising performance. However, they are still challenged by the heterogeneous topological neighborhood structures, which could cause the models to produce different representations of counterpart entities. In the paper, we propose a global entity align- ment model with gated latent space neighborhood aggregation (LatsEA) to address this chal- lenge. Latent space neighborhood is formed by calculating the similarity between the entity embeddings, it can introduce long-range neighbors to expand the topological neighborhood and reconcile the heterogeneous neighborhood structures. Meanwhile, it uses vanilla GCN to aggre- gate the topological neighborhood and latent space neighborhood respectively. Then, it uses an average gating mechanism to aggregate topological neighborhood information and latent space neighborhood information of the central entity. In order to further consider the interdependence between entity alignment decisions, we propose a global entity alignment strategy, i.e., formu- late entity alignment as the maximum bipartite matching problem, which is effectively solved by Hungarian algorithm. Our experiments with ablation studies on three real-world entity alignment datasets prove the effectiveness of the proposed model. Latent space neighborhood information and global entity alignment decisions both contributes to the entity alignment performance im- provement. 1 Introduction Knowledge graph (KG) is an important tool to store knowledge, which provides support for many appli- cations, such as question answering systems, recommender systems. Many KGs have been constructed for particular applications. Sometimes single KG cannot meet the needs of certain applications, so it is necessary to integrate multiple complementary KGs. Entity alignment is non-trivial to integrating dif- ferent KGs. Traditional entity alignment methods are mainly divided into two categories, one is based on the entity’s label information (Ngomo and Auer, 2011; Pershina et al., 2015), the other is based on manually defining features. Recently, the embedding-based representation learning has attracted more and more attention. Given a KG, embedding-based representation learning can map entities to a low- dimensional vector space, then entity alignment model can find counterpart entities according to the similarity of the embeddings. The embedding-based entity alignment approaches show its superiority. Such methods are all built based on KG embedding models (such as TransE (Bordes et al., 2013)). Then they use a transformation matrix to convert the embedding space and complete the entity alignment by calculating the similarity between entity embeddings, such as MTransE (Chen et al., 2017). In recent researches, graph convolutional neural networks (GCN) have shown amazing results when processing graph structure data. GCN-based Entity alignment models all hope that counterpart entities have similar neighborhood structures and have similar vector representations. Several recent GCN-based alignment models (such as GCN-Align …

Content truncated to 500 words