Automatic Generation Method Of Data Flow Diagram Based On Text Diagram Element Extraction

Data flow diagrams,as an indispensable part of software engineering requirements analysis,play an essential role in analyzing software and assuring software quality.Among the classic software engineering methods,manual design as the primary method of generating data flow diagrams has the shortcoming of low efficiency.Therefore,it has important research significance and application value to improve the efficiency and quality of drawing by automatically extracting the graphic elements in the software requirement text.This paper designs a method for automatically extracting data flow graph elements.The method first filters out key sentences containing graph elements from software requirements documents through text classification,and then uses named entity recognition to extract data flow graph elements from sentences.The research work of this study includes two parts.First,aiming at the problem that irrelevant information in the software requirements documents will interfere with the extraction of diagram elements,this paper proposes a text classification method based on the weight matrix of the requirements document.According to the semistructured characteristics of the software requirements document,this method constructs a weight matrix of standard software requirements specification and combines them with sentence-level text vectors.We use the demand document data set to conduct experiments on the text classification task of extracting key sentences.Experimental results show that compared with the sentence vector-based text classification method,After incorporating the weight matrix,the accuracy of key sentence classification increased by 5.82%.After fusing the weight matrix,the feature fusion can reduce sentences' impact in the document-independent directory on the classification results.Second,aiming at the problem that the current named entity recognition methods based on sequence annotation is not accurate in extracting graph element information,this paper takes data flow diagram rules as prior knowledge and proposes a diagram element named entity recognition method combining prior knowledge.This method encodes the previous knowledge into a feature template,connects the context information of the model to mine the characteristics of the diagram elements,obtains the output sequence representation with the most considerable conditional probability,and check the rationality of the conventional prior knowledge of the output sequence representation.We use a data set of text sentences with diagram elements to conduct experiments to extract diagram elements from data flow diagrams.Experimental results show that compared with the named entity recognition method based on sequence labeling,the diagram elements' extraction accuracy is improved by 1.9%.Besides,the graph elements extracted after prior knowledge verification are closer to human cognition.Based on the above research,this paper designed a tool to automatically extract the element information of the data flow diagram in the text,and use real enterprise software requirements documents to test it.The test results show that the tool can effectively extract the graph element information contained in the text,and automatically generate a reference toplevel data flow graph,which saves the time of manual analysis and is useful for generating deeper data flow graphs.