Research On Requirement Document Modeling Technology For Civil Aircraft Digital Design

The interdisciplinary and interdisciplinary nature of complex systems will lead to time-consuming and laborious requirements analysis process,ambiguity and insufficient acquisition of implicit requirements.Therefore,a simple and efficient automated tool is needed to analyze and process requirements documents,build professional domain knowledge spectrum,and help analysts build system models.However,building a knowledge graph in vertically segmented professional fields faces many difficulties,and entity relationship extraction,as a key task in building a knowledge graph,has become particularly important for research and exploration.In view of the complex types of entity relationships in professional fields,the lack of labeled corpora,and the difficulty in extraction,Thesis deeply studies the current research status of relationship extraction,explores the key technologies of relationship extraction,and proposes a method to build a map of professional domain knowledge,which mainly includes the following aspects:(1)A method for constructing professional domain datasets based on Chinese syntactic structure features and Bootstrapping method is proposed to solve the problem of missing labeled data in professional domains.By analyzing the characteristics of Chinese syntactic structure,constructing corresponding filtering rules,matching rules,and expanding rules,and extracting annotation seed sets with rich entity relationship types from text data.Then combined with the Bootstrapping method,more entity pairs with this relationship type in the text data are extracted to construct a high-quality labeled dataset.(2)A Chinese entity relationship extraction method based on probability graphs and expansion gates is proposed to solve the problems of complex entity relationships and overlapping relationships in professional fields.Firstly,the constructed professional field data set is input as the original corpus,and pre-processed such as word segmentation and word segmentation;then,the semantic features of the text are represented by combining word and word mixed vectors and position vectors,and input into the expansion gate attention network model for training;Finally,based on the idea of probability graph,the main entity is identified first,and then the sub-entity and entity relationship are identified.(3)Conducted self-comparison experiments and comparison experiments with other extraction models.The results of self-comparison experiments verify the influence of each module of the model on the performance improvement of relation extraction;the comparison experiments with other extraction models verify the effectiveness of the model.Finally,the extracted relational triples are used to build a knowledge map in the field of digital design of civil aircraft to realize the modeling of requirements documents.