SEEK: Segmented Embedding of Knowledge Graphs

Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 3888–3897 July 5 - 10, 2020. c⃝2020 Association for Computational Linguistics 3888 SEEK: Segmented Embedding of Knowledge Graphs Wentao Xu1, Shun Zheng2, Liang He2, Bin Shao2, Jian Yin1∗, and Tie-Yan Liu2 1 School of Data and Computer Science, Sun Yat-sen University, Guangzhou, China; Guangdong Key Laboratory of Big Data Analysis and Processing, Guangzhou, China 2 Microsoft Research Asia, Beijing, China {xuwt6@mail2,issjyin@mail}.sysu.edu.cn {Shun.Zheng,Liang.He,binshao,tyliu}@microsoft.com Abstract In recent years, knowledge graph embedding becomes a pretty hot research topic of arti- ficial intelligence and plays increasingly vi- tal roles in various downstream applications, such as recommendation and question answer- ing. However, existing methods for knowl- edge graph embedding can not make a proper trade-off between the model complexity and the model expressiveness, which makes them still far from satisfactory. To mitigate this problem, we propose a lightweight modeling framework that can achieve highly competitive relational expressiveness without increasing the model complexity. Our framework focuses on the design of scoring functions and high- lights two critical characteristics: 1) facilitat- ing sufficient feature interactions; 2) preserv- ing both symmetry and antisymmetry proper- ties of relations. It is noteworthy that owing to the general and elegant design of scoring func- tions, our framework can incorporate many fa- mous existing methods as special cases. More- over, extensive experiments on public bench- marks demonstrate the efficiency and effec- tiveness of our framework. Source codes and data can be found at https://github.com/ Wentao-Xu/SEEK. 1 Introduction Learning embeddings for a knowledge graph (KG) is a vital task in artificial intelligence (AI) and can benefit many downstream applications, such as personalized recommendation (Zhang et al., 2016; Wang et al., 2018) and question answering (Huang et al., 2019). In general, a KG stores a large collec- tion of entities and inter-entity relations in a triple format, (h, r, t), where h denotes the head entity, t represents the tail entity, and r corresponds to the relationship between h and t. The goal of knowl- edge graph embedding (KGE) is to project massive ∗Corresponding author. interconnected triples into a low-dimensional space and preserve the initial semantic information at the same time. Although recent years witnessed tremendous re- search efforts on the KGE problem, existing re- search did not make a proper trade-off between the model complexity (the number of parameters) and the model expressiveness (the performance in capturing semantic information). To illustrate this issue, we categorize existing research into two cat- egories. The first category of methods prefers the simple model but suffers from poor expressiveness. Some early KGE methods, such as TransE (Bordes et al., 2013) and DistMult (Yang et al., 2015), fell into this category. It is easy to apply these methods to large-scale real-world KGs, but their performance in capturing semantic information (such as link prediction) is far from satisfactory. In contrast, the second category pursues the excellent expressiveness but introduces much more model parameters and tensor computations. Typical examples include TransH (Wang et al., 2014), TransR (Lin …

Content truncated to 500 words