Aspect Mining Method Research Based On Vector Space Models

The traditional programming techniques can not modular system concerns,so theseconcerns are typically implemented in multiple methods throughout the system and these arenamed to be crosscutting concerns.The use of an aspect-oriented programming paradigm,theimplementation and use of these crosscutting concerns can be implemented into aspects, thus,soluting the problem that legacy system is difficult to understand,maintain and evolve, due tothe useness of traditional programming techniques.Changing legacy system intoAspect-oriented system has become a hot research topic. In order to achieve the legacy systemto Aspect-oriented system, the process of aspect mining is carried out to necessarily identifythe crosscutting concerns in the legacy system. Once identified, the crosscutting concerns canthen be refactored into aspects. This study investigated automated aspect mining method inwhich we focus on the study that identify crosscutting concerns using vector space models torepresent method in the source code of the legacy system.Firistly, this paper analyzes the mining process that the existing Aspect Mining methodbased on Vector Space Models, finding that the existing vector space model attributes arebased on the class level in the source code. We continue to further analysis of the source code.According to the called method and method name decomposition of token word,we proposethree improved vector space models, they are FIV-class, FIV-method andFIV-method-tokens,then we give the construction algorithm of each model and descript themining process of using heuristic-based k-means clustering Algorithm for aspect mining.According to the JHotDraw source code,based on the same clustering Algorithm,we use theexisting vector space models and improved vector space models to mine Aspect and give theexperiment procedures and results. The experimental results show that, compared with theexisting vector space models we find that three vector space models that we proposedsignificantly improved on the results of DISP.In order to enhance the efficiency of aspect mining method based on vector space modelfurther, in this paper, the existing two based on heuristic clustering algorithm arecompared,we prove that HAC algorithm is more suitable for aspect mining method based onvector space model than HACO algorithm, then we give the mining process of aspect mining based on HAC clustering algorithm and using three improved vector space model. Accordingto the JHotDraw source code, heuristic-based k-means clustering Algorithm and HACclustering Algorithm,we use the existing vector space model and the improved vector spacemodel to mine Aspect and give the experiment procedures and results. The experimentalresults show that, compared with the the existing vector space models we find that based onthe HAC algorithm three vector space models we proposed both significantly improved on theresults of DISP and DIV, the precision and recall rate in identification of crosscutting concernimprove significantly too.We also evidence that agglomerative hierarchical clustering is moresuitable than partitional clustering for aspect mining based on vector space models.