A common framework for aspect mining based on crosscutting concern sorts

Abstract

The increasing number of aspect mining techniques proposed in literature calls for a methodological way of comparing and combining them in order to assess, and improve on, their quality. This paper addresses this situation by proposing a common framework based on crosscutting concern sorts which allows for consistent assessment, comparison and combination of aspect mining techniques. The framework identifies a set of requirements that ensure homogeneity in formulating the mining goals, presenting the results and assessing their quality. We demonstrate feasibility of the approach by retrofitting an existing aspect mining technique to the framework, and by using it to design and implement two new mining techniques. We apply the three techniques to a known aspect mining benchmark and show how they can be consistently assessed and combined to increase the quality of the results. The techniques and combinations are implemented in FINT, our publicly available free aspect mining tool.