A neural model for type classification of entities for text

Entity classification has become an increasingly crucial component in the development of knowledge graphs. Due to the incompleteness of the knowledge graph, the semantic relation features of entities in the knowledge graph are generally incomplete, leading to some entities cannot be complete classified. To overcome the weakness of existing research, in this study, we investigated the problem of classifying entities in knowledge graph from the text and proposed an end-to-end entity classification system based on the neural network model. To be specific, firstly, the mention model used long short-term memory to identify the types of each entity mention from the sentences that it contains. Secondly, we proposed a fusion model to fuse the types of multiple mentions to compensate for the existing systems of entity classification. The experimental results demonstrated the necessity and effectiveness of each module in the system. We believe that our proposed method posed a good complement for the existing systems of entity classification.