Research On Construction Method Of Domain-specific Knowledge Graph

Knowledge Graph is an efficient knowledge representation model proposed by Google in 2012.It maps symbols to entities or concepts existing in the real world,and then links different types of information together to form a huge semantic network graph using the association between these entities or concepts as connectors.Compared with traditional information management methods,the logical relationship between knowledge can be effectively acquired by knowledge graph,which is conducive to the intelligent reasoning among knowledge.The domain-specific knowledge graph can be applied to search engine,intelligent question and answer system,knowledge mining and decision support.Therefore,the research of the method of its construction is of great significance.This thesis research on the construction of Chinese domain-specific Knowledge Graph based on the current research achievement,and puts forward some improvement schemes for some key technologies.The main contributions of this thesis include:1.Based on the character-level Bi-LSTM-CRF named entity recognition model,this thesis proposes some innovative improvement schemes according to the characteristics of Chinese domain-specific.Among them,the ACCW model can incorporate the word embedding in proportion to the contribution of the meaning of the candicate words to the Chinese character by the attention-based match model.In addition,we procese a method which can incorporate the grapheme feature of Chinese character to the character-level Bi-LSTM-CRF model to enhance the recognition effect of cold boot characters.Then,our proposed method is applied to the Chinese electronic medical record named entity recognition task.The results show our methods has achieved greater recognition accuracy than the baseline.2.This thesis process a knowledge graph representation learning method incorporate the Description Embedding built by Doc2Vec into the translation-based representation model through a single-layer neural network.Then,the method is applied to a medical knowledge graph,and the effective of our method is verified by comparative experiments.3.This thesis designs the framework of the construction of domain-specific knowledge graph,and takes the medical knowledge graph as an example,designs and implements a series of concrete methods for knowledge extraction,knowledge fusion,knowledge storage and knowledge visualization.These methods have high feasibility and high universality for other vertical fields...