Research On Method Of Aspect Mining Based On Identifier Analysis

The integration of Aspect-Oriented Programming (AOP) and refactoring gives a new problem of Aspect-Oriented Refactoring, which consists of two steps, one is aspect mining, the other is aspect refactoring. Aspect mining is not only the precondition of aspect refactoring, but also the most important step of Aspect-Oriented refactoring.The author chooses Identifier Analysis which is a representative method of aspect mining and puts emphases on the improvement of the method. Identifier analysis is confronted with three problems, firstly, it costs too much time to generate concept lattices; secondly, choosing interested concepts manually is excessively complicated; thirdly, it's difficult to be applied to large-scale legacy systems. These problems are caused by using Formal Concept Analysis (FCA) to generate formal concepts which contain a mass of redundant ones. The author noticed that those redundant concepts can be eliminated by filtering elements and attributes in the context as input of FCA algorithm. Accordingly, a series of definitions, propositions and their certifications are showed to ensure that the filter method is based on correct rationale, then a filter algorithm named Pre-Filter which is based upon the foregoing rationale is constructed. Detailedly, Pre-Filter transforms context table into context matrix, and then computes the matrix with mathematical method simply and ingeniously to filter context. Laudably, the applicability of Pre-Filter is not confined to aspect mining; it can be used anywhere as long as there is number limit to concepts' elements and attributes, so it has a common significance.Pre-Filter can be used to reduce inputs of FCA algorithm, so it is placed in front of the step of concept generation in the improved identifier analysis. With Pre-Filter, programmers can not only save much time, but also get less redundant concepts and more legible concept lattices; it makes a progress on the ideal state "what you get is what you need", too. In addition, the workload and complication of choosing crosscutting concepts manually is decreased, and the performance and applicability of identifier analysis is improved.