Improve the translational distance models for knowledge graph embedding
- PDF / 821,644 Bytes
- 23 Pages / 439.642 x 666.49 pts Page_size
- 4 Downloads / 290 Views
Improve the translational distance models for knowledge graph embedding Siheng Zhang1,2 · Zhengya Sun1 · Wensheng Zhang1,3 Received: 20 August 2019 / Revised: 23 December 2019 / Accepted: 27 December 2019 / © Springer Science+Business Media, LLC, part of Springer Nature 2020
Abstract Knowledge graph embedding techniques can be roughly divided into two mainstream, translational distance models and semantic matching models. Though intuitive, translational distance models fail to deal with the circle structure and hierarchical structure in knowledge graphs. In this paper, we propose a general learning framework named TransX-pa, which takes various models (TransE, TransR, TransH and TransD) into consideration. From this unified viewpoint, we analyse the learning bottlenecks are: (i) the common assumption that the inverse of a relation r is modelled as its opposite −r; and (ii) the failure to capture the rich interactions between entities and relations. Correspondingly, we introduce position-aware embeddings and self-attention blocks, and show that they can be adapted to various translational distance models. Experiments are conducted on different datasets extracted from real-world knowledge graphs Freebase and WordNet in the tasks of both triplet classification and link prediction. The results show that our approach makes a great improvement, showing a better, or comparable, performance with state-of-the-art methods. Keywords Knowledge graph embedding · Translational distance model · Positional encoding · Self-attention Wensheng Zhang
[email protected] Siheng Zhang [email protected] Zhengya Sun [email protected] 1
Research Center of Precision Sensing and Control, Institute of Automation, Chinese Academy of Sciences, Beijing, China
2
University of Chinese Academy of Sciences, Beijing, China
3
School of Mathematics and Big Data, Foshan University, Foshan, China
Journal of Intelligent Information Systems
1 Introduction A typical knowledge graph (KG) consists of a set of interconnected typed entities and their attributes. Usually, entities are modelled as nodes, and relations are modelled as different types of edges, linking from a head entity to a tail entity, denoted as (head, relation, tail) or (h, r, t). Although it is well defined and structured, KGs retain the underlying symbolic nature, which makes it difficult to automatically construct or inference on it. To tackle this issue, lots of work has been carried out on knowledge graph embedding. The key idea is to use distributed representation, i.e., embed entities and relations into continuous lowdimensional space, so that manipulation on KG can be simplified as algebraic operations (Nickel et al. 2016). Roughly speaking, embedding techniques in this sort can be divided as two groups: translational distance models and semantic matching models (Wang et al. 2017). Our work follows the route of the first one, which measures the plausibility of a fact as the distance between the head and tail entities after a translation. Note that some other m
Data Loading...
