Research And Implementation Of Mineral Knowledge Graph Construction Technology

The advent of Big Data makes the Internet information that people come into contact with more and more complex,so the searchability of the Internet is required to be higher and higher.The information retrieval mode of keyword matching supported by the traditional knowledge storage method cannot meet the demand.At present,the industry and academia agree that the application of Search-Question Answering based on knowledge graphs will become the research frontier.Knowledge graph includes general knowledge graph and vertical domain knowledge graph.The construction of vertical domain knowledge graph is a crucial step for the development of traditional industries enabled by advanced technology.Whether it is for the basic research of mineralogy,or the applications in metallurgy,material preservation,unique materials development,and others,an extensive and complete knowledge databases,and the application based on the knowledge database are critical.This paper extracted knowledge triple from the systematic mineralogical database and professional mineral literature,constructed a mineral knowledge graph,and presented it visually.The visualization system realized the minimum feasibility verification of minerals knowledge graph,which lays a solid foundation for building a comprehensive and rich mineral knowledge database and realizing the application of mineral knowledge in the whole field.The main work of this paper is as follows:The first step is the data acquisition and preprocessing in the mineral domain.This paper mainly obtains the structured data of professional mineralogy databases,unstructured data such as mineral literature.These can be used as data sources for knowledge acquisition.The second step is the knowledge graph construction stage.This paper mainly studies how to extract and store knowledge triples from mineral corpus.For structured mineral data,this paper kept it in the Neo4 j database after data cleaning and coding processing.In the aspect of entity recognition,by comparing the recognition efficiency of three deep learning methods,Bi LSTMCRF\Albert-Bi LSTM-CRF and Albert-Bi GRU-CRF,and analyzing the results,we find that Albert-Bi GRU-CRF,has the best efficiency.In relation extraction,we study the relation extraction of dependency parsing method and deep learning method.In the dependency parsing module,LTP is used to process the mineral corpus.Finally,the mineral triples are obtained by dependency parsing In the deep learning module,we use the Albert-Bi GRU-Attention method and AlbertBi LSTM-Attention method to classify the mineral corpus and get the relationship between entities.The knowledge representation and storage stage is completed by using the Neo4 j graph database.Finally,Web-based visualization of knowledge graph.The system is constructed by using the Flask framework,using graph database storage and query technology,and using Echarts to achieve data visualization display.