Research On Fine-grained Entity Classification For Knowledge Graph Of Chinese Encyclopedia

Entity classification is a necessary step in building knowledge graphs.So far,plen-ty of efforts have been made in mining concept information for entities from knowledge graphs,but usually only coarse-grained concept information could be obtained for entities,which can hardly satisfy the purpose of knowledge graphs construction or application.This situation becomes even worse for classifying entity task for entities in Chinese.In this paper,we propose a fine-grained classifying entity algorithm for knowledge graph of Chinese encyclopedia.Regarding the encyclopedia entries as entities,we construct a knowledge extraction framework to extract entity information from encyclopedia pages,then obtain high-quality structured data through data cleaning,and finally store the data in knowledge graph in the form of triples.We work on mining high-quality fine-grained concept information for entities from not only the title-labels and info-boxes in the entity information,but also the abstracts and crowd-labels,which could provide more candidate fine-grained type information(with nois-es).In this paper,initially we only get reliable concept information from the title-labels and info-boxes.Then by putting entities,attributes,values and concepts into the information graph,some path information can be obtained between each candidate(entity-concept)pair.We finally rely on a proposed Path-CNN binary classification model based on convolution neural network to identify more(entity-concept)pairs with instance-of relation.Extensive experiments on real dataset demonstrate the superiority of our solution over existing representative approaches.