ReInceptionE: Relation-Aware Inception Network with Joint Local-Global Structural Information for Knowledge Graph Embedding

Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 5929–5939 July 5 - 10, 2020. c⃝2020 Association for Computational Linguistics 5929 ReInceptionE: Relation-Aware Inception Network with Joint Local-Global Structural Information for Knowledge Graph Embedding Zhiwen Xie1, Guangyou Zhou2∗, Jin Liu1∗, Jimmy Xiangji Huang3 1School of Computer Science, Wuhan University 2School of Computer Science, Central China Normal University 3School of Information Technology, York University xiezhiwen@whu.edu.cn, gyzhou@mail.ccnu.edu.cn jinliu@whu.edu.cn, jhuang@yorku.ca Abstract The goal of Knowledge graph embedding (KGE) is to learn how to represent the low- dimensional vectors for entities and relations based on the observed triples. The conven- tional shallow models are limited to their ex- pressiveness. ConvE (Dettmers et al., 2018) takes advantage of CNN and improves the ex- pressive power with parameter efficient oper- ators by increasing the interactions between head and relation embeddings. However, there is no structural information in the embed- ding space of ConvE, and the performance is still limited by the number of interac- tions. The recent KBGAT (Nathani et al., 2019) provides another way to learn embed- dings by adaptively utilizing structural infor- mation. In this paper, we take the bene- fits of ConvE and KBGAT together and pro- pose a Relation-aware Inception network with joint local-global structural information for knowledge graph Embedding (ReInceptionE). Specifically, we first explore the Inception net- work to learn query embedding, which aims to further increase the interactions between head and relation embeddings. Then, we propose to use a relation-aware attention mechanism to enrich the query embedding with the local neighborhood and global entity information. Experimental results on both WN18RR and FB15k-237 datasets demonstrate that ReIncep- tionE achieves competitive performance com- pared with state-of-the-art methods. 1 Introduction Knowledge graphs (KGs) are at the core of most state-of-the-art natural language processing solu- tions and have been spotlighted in many real-world applications, including question answering (Hao et al., 2017), dialogue generation (He et al., 2017; Madotto et al., 2018) and machine reading compre- hension (Yang and Mitchell, 2017). Typically, KGs ∗ Corresponding author. are directed graphs whose nodes denote the entities and edges represent the different relations between entities. The structured knowledge in KGs is orga- nized in the form of triples (h, r, t), where h and t stand for the head and tail entities respectively, and r represents the relation from h to t. Although large-scale KGs (e.g., Freebase (Bollacker et al., 2008), DBpedia (Lehmann et al., 2015)) have al- ready contained millions or even billions of triples, they are still far from complete since the emerging new knowledge appears. Knowledge graph embed- ding (KGE) is an effective solution to solve the incompletion problem. KGE aims to learn the low-dimensional vectors (embeddings) for entities and relations based on the observed triples in KGs. Conventional models including TransE (Bordes et al., 2013) and its nu- merous extensions (e.g., TransD (Ji et al., 2015), TransR (Lin et al., 2015), DistMul (Yang et al., 2015), ComplEx (Trouillon et al., 2016), etc.) have been proposed. These shallow models are limited …

Content truncated to 500 words