AutoETER: Automated Entity Type Representation for Knowledge Graph Embedding

Findings of the Association for Computational Linguistics: EMNLP 2020, pages 1172–1181 November 16 - 20, 2020. c⃝2020 Association for Computational Linguistics 1172 AutoETER: Automated Entity Type Representation for Knowledge Graph Embedding Guanglin Niu1, Bo Li1,2, Yongfei Zhang1,2,3∗, Shiliang Pu4, Jingyang Li1 1Beijing Key Laboratory of Digital Media, School of Computer Science and Engineering, Beihang University, Beijing 100191, China 2State Key Laboratory of Virtual Reality Technology and Systems, Beihang University, Beijing 100191, China 3Pengcheng Laboratory, Shenzhen 518055, China 4Hikvision Research Institute, Hangzhou 311500, China {beihangngl, boli, yfzhang, lijingyang}@buaa.edu.cn, pushiliang.hri@hikvision.com Abstract Recent advances in Knowledge Graph Embed- ding (KGE) allow for representing entities and relations in continuous vector spaces. Some traditional KGE models leveraging additional type information can improve the representa- tion of entities which however totally rely on the explicit types or neglect the diverse type representations specific to various relations. Besides, none of the existing methods is ca- pable of inferring all the relation patterns of symmetry, inversion and composition as well as the complex properties of 1-N, N-1 and N-N relations, simultaneously. To explore the type information for any KG, we develop a novel KGE framework with Automated Entity TypE Representation (AutoETER), which learns the latent type embedding of each entity by regard- ing each relation as a translation operation be- tween the types of two entities with a relation- aware projection mechanism. Particularly, our designed automated type representation learn- ing mechanism is a pluggable module which can be easily incorporated with any KGE model. Besides, our approach could model and infer all the relation patterns and com- plex relations. Experiments on four datasets demonstrate the superior performance of our model compared to state-of-the-art baselines on link prediction tasks, and the visualization of type clustering provides clearly the explana- tion of type embeddings and verifies the effec- tiveness of our model. 1 Introduction In recent years, knowledge graph (KG) has been viewed as a powerful technique for recognition systems and prevalent in many fields such as E- commerce, intelligent healthcare, and public secu- rity. Knowledge graphs collect and store a great deal of commonsense or domain knowledge in fac- tual triples composed of entity pairs with their relations. The existing large scale KGs such as ∗Corresponding Author Will Smith The Pursuit of Happiness ActedInFilm Live It Up SangSong Entity Level Type Level SangSong Attention ActedInFilm Attention Song Singer Actor Film Figure 1: An actual example of the entity-specific triples and the type-specific triples with relation-aware projection mechanism. Will Smith has multiple types such as Singer and Actor, but only the type Singer should be focused on for the relation SangSong. Freebase (Bollacker et al., 2008), WordNet (Miller, 1995), YAGO (Suchanek et al., 2007) have shown their validity in various applications, including question answering (Diefenbach et al., 2018), di- alogue generation (He et al., 2017) and recom- mender systems (Wang et al., 2019). However, the existing KGs are inevitably in- complete whether they are constructed manually or automatically, limiting the effectiveness when exploited for downstream applications. Some ex- isting KG …

Content truncated to 500 words