Automatic Completion For Domain-Oriented Knowledge Graph Attributes

Domain knowledge graph can provide important semantic knowledge support for emergency management,effectively improve the accuracy and timeliness of emergency pre-warning,and provide scientific basis for in-process and in-process disposal.However,the current domain knowledge graph has the problem of incomplete knowledge,especially the very important attribute knowledge in the knowledge graph faced with the problem of missing attributes and attribute values,and the process of attribute completion is faced with the situation of attribute redundancy ambiguity and attribute value filling error.Therefore,this paper studies the automatic attribute completion technology of domain knowledge graph,which has important theoretical and application value.Firstly,this paper proposes a multi-dimensional feature multi-source knowledge base attribute fusion algorithm.The vector representation of attribute names is obtained by using the word vector representation model of fusing glyph features,and then the cosine similarity is used to measure the similarity between names,and then a standardized set of attribute names is constructed to provide knowledge support for attribute correction of domain knowledge graph.On this basis,aiming at the problem of incomplete attribute of domain knowledge graph,this paper proposes an attribute completion model,which is divided into two layers:the first layer uses the external text corpus of the graph to complete the attribute,and the second layer uses the internal knowledge of the graph to complete the attribute.For the first layer,an improved attribute extraction model of distant supervision is designed,and the twin neural network combined with dependency syntactic information is used to denoise the entity related corpus,and then the attribute information is extracted by relational classification method and supplemented into the atlas;For the second layer,the structure information of the graph,the text information of entity background and the type information are fully used to model the three tuples,and then the knowledge reasoning is used to complete and correct the knowledge graph attributes.Finally,based on the above core model,the automatic attribute completion system of knowledge map is designed and implemented.The experimental results show that the multi-dimensional feature multi-source knowledge graph attribute fusion model has a certain advantage in the accuracy of evaluation index,and can meet the needs of multi-source knowledge base attribute fusion in practical application;The attribute extraction model based on the improved remote monitoring algorithm can complete attributes based on multi-source external text,and its accuracy,recall rate and F1 value are better than the comparative experiment;The average completion rate of attribute completion task based on knowledge reasoning is better than that of control experiment.To sum up,the automatic attribute extraction model in this paper can meet the actual application requirements.