Entity Linking Based On Multiple Knowledge Base

With the proposed and rapid development of knowledge graph,as an important step in the construction of knowledge graph,more and more researchers from academic and industrial pay more and more attention on the entity linking task.In natural language texts,there are many ambiguous mentions that may affect many natural language processing tasks.The entity linking task can eliminate the ambiguity of the mentions and link mentions to the corresponding entity in the target knowledge base.This not only can solve the ambiguity of mentions,but also can complements the text with additional background information.So that more information can be used to improve the performance when further natural language processing tasks are processing.According to the realization process of entity linking,this paper divides the process of entity linking into two parts: candidate set construction and entity disambiguation.For these two parts,an entity linking method based on multiple knowledge base is proposed.The specific research content of this article includes:(1)Multiple knowledge base information extraction: The structure and content of multiple knowledge bases are analyzed.Then for specific information in each knowledge base,more useful information for entity linking is extracted.(2)Entity linking baseline construction: The definition of the entity linking is stated,and the basic flow is summed up.According to the basic flow,the method and features of the system is chosen.Finally,the entity linking baseline is constructed.(3)Multi-strategy entity candidate construction method based on multiple knowledge base: According to the baseline,the entity candidate set construction method is improved.A multi-strategy candidate set construction method based on multiple knowledge sources is proposed,and a higher quality candidate set is constructed.(4)Entity disambiguation based on CNN and multiple knowledge base: The entity disambiguation is divided into two parts.Then an improved method based on convolutional neural network,Ranking SVM and multiple knowledge base is proposed for these two parts.