Research On Knowledge Graph Construction And Retrieval For Copper-based Composite Materials Literatur

In recent years,with the accumulation of materials data and the popularization of the Materials Genome Project,more and more materials experts and scholars are aware of the value and importance of materials computing in materials development.However,because a large amount of knowledge in this field exists in materials science literature,key entity data such as composition-property and relationship between entities cannot be extracted effectively,and there are very few data of composite materials literature with entity or relation labels,which leads to the low accuracy of entities and relations obtained by using deep learning or other common entity-relation extraction methods.Meanwhile,the retrieval through knowledge graphs can only query the related information of a certain material,and it is difficult to obtain a large number of composite materials with high similarity.In this regard,this paper takes the literature on copper-based composites,a typical representative in the field of composite materials,as the research object,and completes the construction and retrieval of the corresponding knowledge graphs,and the specific research work is as follows.(1)Knowledge graph construction for copper-based composites literatureFirstly,documents related to copper-based composites from 2011-2022 were collected as the original data for the construction of the knowledge graph,taking the field of copper-based composites as the research background.After analyzing the characteristics of copper-based composites data,we proposed the regularization method based on rule matching for entity relation extraction to convert unstructured data into structured data;then we calculated the similarity between semantics for knowledge fusion to realize the construction of the knowledge graph in the copper-based materials domain.As a result,a total of 6,154 entities and 15,561 pairs of relations were extracted,and their accuracy and precision rates reached over 80%.The established material knowledge graph can form a "matrix-enhanced body-composite material-performance-performance value" connection path,which can be visually queried and analyzed on the Neo4j graph database.(2)Material knowledge retrieval based on meta-path similarity calculationIn order to quickly and accurately find materials similar to a certain composite material,the material knowledge graph is viewed as a heterogeneous information network,and a weighted material similarity calculation model WM-PathSim that can measure different types of nodes is proposed.Firstly,Metapath2vec is used to learn the embedding representation of composite materials,properties,performance values,and literature sources.Secondly,the TFIDF-CBOW model is introduced to learn the existence probability of material path instances,and then calculate the weights of different metapaths.Finally,the final similarity metric is obtained by weighting and fusing the eligible meta-paths to predict the degree of similarity between different copper-based composites.The results on the real dataset show that the proposed model in this paper has a greater performance improvement compared with the baseline method in different path relations,and its AUC and Precision metrics are improved by 0.37%-5.02%and 1%-7.33%,respectively,indicating that this model is more accurate and effective in obtaining the degree of similarity between materials.