Research On UML Use Case Digram Extraction Of Software Requirements Document

In the life cycle of software development,software requirements documents are often used to collate and integrate the user needs.These documents describe the needs of customers and related personnel for software systems in natural language.The interpretation and analysis of the documents are usually completed by requirements engineers and professionals.As software system grows in size continuously,the internal logic complexity also shows a rising trend.For the ever-increasing amount of software demand text,the workload of related processing personnel has also increased significantly.In order to shorten the modeling and analysis time and improve the efficiency of software development,Automatic modeling of software requirements documents has been paid more and more attention.With the update of related technologies in artificial intelligence and natural language processing,how to integrate these technologies with software requirement document knowledge extraction has always been a research hotspot in the field of automatic modeling.In order to solve the above problems,This paper propose an automatic modeling research plan for software requirements documents.By analyzing the organization and related characteristics of UML use case diagrams,the corresponding entities and the relationships between entities are classified and mapped,and the problem of extracting use case diagrams is transformed into the problem of extracting entity relationships from text sequences.Based on natural language processing and neural network technology,a model framework based on sentence-level attention-based joint entity relationship extraction is constructed to realize the extraction of entities and relationships from natural language software demand text.The model is divided into a bottom-level parameter extraction layer,an entity recognition layer,a relationship extraction layer,and a relationship filtering layer.The Encoder-Decoder framework is used as the basic architecture of the entity relationship model to ensure the completeness of sentence semantic extraction and the diversity of relationship extraction,as well as entity and The correlation between relationships.The cnostructed Encoder layer based on the cyclic neural network performs feature memory extraction at the word level,and integrates semantic feature information at the sentence level to ensure that key semantic features are not lost while removing redundant features.On the basis of the underlying parameters,the named entity recognition of all demand documents is carried out through the model of conditional random field.Use the copy mechanism in the Decoder layer to improve the extraction of relations.Aiming at the problem of easily losing key semantic features and the influence of noise in relation extraction tasks,propose a relation extraction module for sentence-level attention.At the sentence level,an attention mechanism is introduced to reduce the influence of noise data on the test results.Use a variety of currently popular joint entity relationship extraction methods and the model to conduct experimental comparisons on the same data set.Analyze the pros and cons of different models for the extraction capability and performance of the entity relationship triples of the software requirements documents.Subsequently,graph database technology is used to realize the transformation of entity relationship triples to UML use case diagrams.The program does not require the participation of domain experts in the process of the program,relying on the advantages of natural language processing and neural network high computing,high performance word processing to achieve automatic modeling of UML use case diagrams of software requirements documents.