Title: Fine-grained Entity Typing for Relation-Sparsity Entities
Authors: Lei Niu, Binbin Gu, Zhixu Li, Wei Chen, Ying He, Zhaoyin Zhang and Zhigang Chen
Abstract: This paper works on fine-grained entity typing without using external knowledge for Knowledge Graphs (KGs). Aiming at identifying the semantic type of an entity, this task has been studied predominantly in KGs. Provided with dense enough relations among entities, the existing mainstream KG embedding based approaches could achieve great performance on the task. However, many entities are sparse in their relations with other entities in KGs, which fails the existing KG embedding models in fine-grained entity typing. In this paper, we propose a novel KG embedding model for relation-sparsity entities in KGs. In our model, we map all attributes and types into the same vector sapce, where attributes could be granted with different weights according to an employed attention mechanism, while attribute values could be trained as bias vecotrs from attribute vectors pointing to type vectors. Based on this KG embedding model, we perform entity typing from coarse-grained level to more fine-grained level hierarchically. Besides, we also propose ways to utilize zero-shot attribute values that never appear in the training set. Our experiments performed on real world KGs show that our approach is superior to the most advanced models in most cases.
