Recovery and Analysis of Architecture Descriptions using Centrality Measures

TL;DR

This paper introduces a novel method for recovering architecture descriptions by applying social network centrality measures to program elements, facilitating understanding of system structure.

Contribution

It proposes a two-phase approach using centrality measures and layer assignment techniques, including rule-based and machine learning methods, for architecture recovery.

Findings

Both layer assignment techniques achieve high accuracy.

The approach effectively identifies system architecture layers.

Centrality measures correlate with architectural roles.

Abstract

The necessity of an explicit architecture description has been continuously emphasized to communicate the system functionality and for system maintenance activities. This paper presents an approach to extract architecture descriptions using the {\em centrality measures} from the theory of Social Network Analysis. The architecture recovery approach presented in this paper works in two phases. The first phase aims to calculate centrality measures for each program element in the system. The second phase assumes that the system has been designed around the layered architecture style and assigns layers to each program element. Two techniques to assign program elements are presented. The first technique of layer assignment uses a set of pre-defined rules, while the second technique learns the rules of assignment from a pre-labelled data set. The paper presents the evaluation of both approaches.