Construction and analysis of vector space models for use in aspect mining

Abstract

A software system is designed so that its concerns are as independent as possible. Concerns upon which other concerns depend are called crosscutting concerns, examples of which are logging, authentication, and session management. Crosscutting concerns in a software system have the potential to increase the number of defects over time as the system is evolved. Aspect-oriented programming provides an additional layer of abstraction to the object-oriented programming paradigm for the purpose of separating concerns. The search for crosscutting concerns is referred to as aspect mining.Previous aspect mining algorithms used aggregated metric values as components in the vector space model. In this paper a new method for constructing vector space models is proposed that attempts to retain the detail present in the relationships between the elements of a software application. This is done through the use of pattern matrices derived from the non-aggregated metrics. The non-aggregated vector space models are then used in a clustering-based aspect mining algorithm and their performance is evaluated. The results show that this new approach to constructing vector space models is a viable one but needs further investigation. Issues with current measures for evaluating clustering-based aspect mining algorithms are highlighted and directions for further research are given.