Research On Task Data Semantic Knowledge Annotation Method For Academic Atlas Text

Semantic knowledge of task data is the basis of task-driven remote sensing data retrieval,which consists of six elements: time,space,task,remote sensing data product,method and effect entity.Due to the text of academic atlas contains rich semantic knowledge of task data,the annotation is a prerequisite for automating the acquisition of task data semantic knowledge from the text.Therefore,it is of great significance for obtain the semantic knowledge of large-scale task data to designing the markup language of task data semantic knowledge text for academic atlas,formulating its annotation specification and improving the annotation rate.This paper uses the text of academic graph as the data source,constructs the ontology model of six entities of task data semantic knowledge,designs the task data semantic knowledge text markup language,establishes the task data semantic knowledge text labeling specification,and design a semi-automatic labeling algorithm used the deep active learning method,which realizes the rapid labeling of task data semantic knowledge.The main work of this paper includes the following three parts:(1)Design of task data semantic knowledge text markup languageReferring to the relevant knowledge in the field of remote sensing,this paper constructs the ontology framework of the entity composed of task data semantic knowledge,and designs the task data semantic knowledge text markup language by analyzing the structure of academic graph text and the description features of task data semantic knowledge.(2)Text annotation specification of task data semantic knowledgeThe annotation specification is formulated for the constituent entities in task data semantic knowledge and a relatively complete annotation system is formed according to the text expression characteristics of named entities of task data semantic knowledge to solve problems of diverse text descriptions and inconsistent entity expressions of task data semantic knowledge.(3)Semi-automatic annotation of semantic knowledge text of task data based on deep active learningTo address the problem of low efficiency of manual annotation,this paper utilizes deep active learning method to semi-automatically annotate the semantic knowledge text of task data.Firstly,samples are selected by sample selection strategy,then the BERT-NER model is used for annotation,and after manual review,the data is expanded to the training set to select samples again from and annotate again until the model achieves the expected effect.Finally,the effectiveness of the proposed method is verified by experiments.