INT: Integrating Neighborhood and Textual Information for Knowledge Graph Entity Typing

Knowledge Graph Entity Typing (KGET) aims to infer missing types of entities in knowledge graphs, which plays an essential role for various downstream tasks. Existing methods for KGET model type instances, but overlook the information hidden in entity texts. In this paper, we propose a model called INT, which Integrates entity Neighborhood information and Textual type features. Specifically, the model takes into account complex relations and focuses on the interaction information among them. It is achieved by independently considering relation neighbors and type neighbors to infer missing entity types. Meanwhile, we extract type-specific information from entity text to strengthen the semantic representation of entity, further improving the performance of complex and fine-grained entity type prediction. We conduct extensive experiments on two public datasets, i.e., FB15kET and YAGO43kET. Experimental results demonstrate that our proposed model can outperform other methods. In addition, we verify the effectiveness and necessity of each module in the proposed model.