Research On Software Static Structural Networks: Topological Characteristics And Metrics

The traditional method of software metrics can not be applied to large-scale software effectively, so how to measure the software becomes a challenge in software field. In recent years, researchers find that the software structural networks show the same characteristics with the complex networks. Furthermore, the software system (topological) structure has an effect on the quality characteristics, which will boost the development of software metrics. In response to these problems, based on the interdisciplinary research, the thesis with the aim of quantifying software metrics tries to propose a set of systematic method to measure software network.Firstly, the Software Networks Metrics Platform is designed and realized. The platform can analyze the source code of the software, generate the software network, calculate the topology eigenvalues, and measure software network with the methods of measuring the eigenvalues which are based on Deviation of Eigenvalues of Software Network.Secondly, based on the results of the open-source software samples analyzed by Software Networks Metrics Platform, the macro-topology eigenvalues with statistical methods are analysed, The results indicate that the average shortest path length, the average degree of nodes and graph coreness increase by the number of nodes in software network, while the standard structure entropy decreases. However, there is no obvious relation between the average clustering coefficient and the number of nodes.Finally, based on characteristics of Macro-Topology eigenvalues, the definitions of deviation of eigenvalues of software network and the relative quality characteristics of software on scale are given. The average shortest path length deviation (D(AP)), the average degree deviation (D(AD)) as well as the standard structure entropy deviation (D(E)) are evaluated and it is that shows how the eigenvalues affect the software efficiency, the complexity and the order of code structure (Maintainability). The experiment results show that the method is feasible and effective. Through the analysis, we get that when D(AP)>0.7, D(AD)>0.5 or D(E)>0.05, the corresponding characteristics of software quality can be optimized significantly.The thesis not only offers a professional tool to research on software networks, but also provides a theoretical basis and methods for metrics of software networks.