Duet Representation Learning with Entity Multi-attribute Information in Knowledge Graphs

Yuanbo Xu, Yuanbo Zhang, Yongjian Yang, Hangtong Xu, Lin Yue

Abstract: Representation Learning (RL) of knowledge graphs aims to project both entities and relations into a continuous low-dimension space. Most methods concentrate on learning entities’ representations with structure information indicating the relations between entities (Trans- methods), while the utilization of entity multi-attribute information is insufficient for some scenarios, such as cold start issues or zero-shot problems. How to utilize the complex and diverse multi-attribute information for RL is still a challenging problem for enhancing knowledge graph embedding research. In this paper, we propose a novel RL model Duet Entity Representation Learning (DERL) for knowledge graphs, which takes advantage of entity multi-attribute information. Specifically, we devise a novel encoder Entity Attribute Encoder (EAE), which encodes both entity attribute types and values to generate the entities’ attribute-based representations. We further learn the entities’ representations with both structure information and multi-attribute information in DERL. We evaluate our method on two tasks: the knowledge graph completion task and the zero-shot task. Experimental results on real-world datasets show that our method outperforms other baselines on two downstream tasks by building effective representations for entities from their multi-attribute information. The source code of this paper can be obtained from https://anonymous.4open.science/r/DUET-adma2023/ .