Semantic Topological Relation Based Use Case Diagram Retrieval

There is a certain degree of degree thesis plagiarism phenomenon in China,of which software engineering thesis is one of the most serious type.At present,3 Chinese journal databases can be used for degree thesis duplicate checking,they are CNKI,Wan Fang Data and VTTMS.Based on the feedback information from degree thesis duplicate checking systems,we cannot only obtain the similarity between our papers and others' but also prevent bad academic phenomena such as copy and plagiarism to improve thesis quality.However,the 3 databases can only recognize words and some tables except images.Based on our observations,there is a lower text repetition rate but a higher UCDs similarity degree in those plagiarized degree theses.Indeed,the UCDs reflect the personalized needs of software systems,there should not be similar ones.Thus,it is of great significance to identify the semantic similarity of software engineering Use Case Diagrams(UCDs)in degree thesis duplicate checking.In software engineering papers,many Use Case Diagrams(UCDs)have close or same semantics despite their different appearances.The semantic of UCD lies in the topological relations between its component elements.Due to the features of UCD,the topological relations of its component elements have special semantics.On one hand,it does not only have limit topological relations types but also share different semantics probably despite of the same topological relation,as is the fact that we cannot tell which is from association,include,extend and generalization based on the classic topological models;on the other hand,if the direction relation of a pair of component elements in a UCD gets changed,instead,we think that their semantic of topological relation keeps the same.The key of UCD retrieval lies in the comparison of topological relation semantics,which means that many spatial relation model-based methods can be inappropriate for UCD retrieval.Content-based Image Retrieval(CBIR)is a new type of method for image retrieval based on image content and its context,which mainly contains two key work: feature extraction and similarity comparison.It has three 3 layers of processing in terms of technical level: 1)low-level features-based or their mixture-based image retrieval,such as shape,color,texture and spatial relation;2)modeling from image internal target object for image retrieval;3)reasoning from semantics lie in image internal target object and context for image retrieval.A variety kind of research have arouse in the field of image retrieval from early text-based image retrieval method to content-based image retrieval nowadays.Many image retrieval approaches can be applied to retrieve UCDs,including document retrieval,circuit diagram symbol retrieval,engineering drawings retrieval and hand drawings retrieval,etc.According to our research,these methods can recognize UCDs that are similar in appearance,they can hardly work well when the UCDs have different appearance despite their semantics are the same.Based on the above work,a new schema for UCD retrieval based on semantic topological relation has been proposed.The main idea of this paper is listed as follow:1)Give a brief introduction to the key concepts and important characteristics,introduce the component elements of UCD,especially relation,the one has semantics,and we also introduce the research topic of this paper by using a concrete case.2)Give a brief introduction to content-based image retrieval technologies,analysis the development situation and existing problems of severalrelated topics to our research,including document retrieval,circuit diagram retrieval,engineering drawings retrieval and hand-drawing retrieval etc.3)Give a detail introduction to our proposed method,at first,graph primitives extraction,including actor,use case and relations;and then,construct a semantic topological relation graph based on the semantic topological relations to describe the structural information,in this step,we translate a UCD into a weighted diagraph and use different weights to distinguish different topological relations that share semantics;at last,define similarity degree to complete retrieval,which is in fact a graph matching work,we use an improved graph matching algorithm to reduce the complexity.4)Establish a dataset and do experiments to validate our method in terms of efficiency of UCD retrieval,including qualitative and quantitative experiments and compare to other image retrieval methods.