Multi-Granularity Characteristics Analysis In Large Scale Software Networks

The evolution and update of software play an important role in the softwareengineering, which has provided a guarantee for quality control, iterative developmentand reconstruction of software. Previous works mainly analyze the evolution of softwarenetwork at class level, but the object-oriented software has appeared multiple levels,multiple granularities, and multiple integration mode of organization. If the research islimited to a single level of granularity that can not completely reflect the structuralcharacteristics of the software, so we should research the software evolution from amulti-granularity perspective.In view of such problem, this paper abstracts the object-oriented software intopackage granularity, class granularity and method granularity of three different levels ofthe software network, and statistics the evolution of characteristic value in thethree-granularity software network, provides quantitative basis for the softwaredevelopment and maintenance.This paper firstly analyzes the evolution of multi-granularity software based on thedegree, the results show that the average degrees, greatest degrees and the averagedegrees of nodes with high degree values in the three levels software network generallyhave upward trend evolution; the class granularity and method granularity of softwarenetwork have a similar evolution rules, which reflects the fractal feature. The nodes withlow in-and-out degrees in three kinds of software networks account for the vast majority,which is more prominent in the method and class level. The degree distributions ofmulti-granularity software networks all have power-law characteristics, and power-lawexponents of method and class level of software networks show slowly declining orstable evolution trend, while it is upward trend but remain minimal in the packagesoftware network.Secondly, this paper does evolution analysis on center index—betweenness andcloseness in multi-granularity software networks. This paper discusses the evolution trend of the average values and ranges of stable values，distribution characteristics andevolution, and referrers to the correlations, and lastly classifies and analyzes theevolution trends.Finally, this paper studies the evolution trend of the standard structure entropy inmulti-granularity software network. The results show that standards structure entropy inthe three kinds of granularity software networks with the versions of software evolutionpresent a decreasing trend, which indicates that the software structure towards the orderlydevelopment, and then statistics the ranges of stable values about standard structureentropy. It has great practical significance in guiding the software upgrade, controllingthe software complexity and researching how software design ideas impact of softwarearchitecture.