Metric Study, Based On The Impact Of The Software Network Software

Software metrics is one of the guarantees to make software design and development scientific. The software complexity metrics is an important aspect of software metrics, which is closely related with software reliability, maintainability. With the increasing scale of software systems and the increasing complexity of issues addressed by it, which is a new raised challenge for software development process management and quality assurance of software products. Therefore, how to understand, measure, predict, control and improve software complexity becomes one of extremely important problems in the field of software engineering.Studies have shown that large-scale software system as an artificial intelligence system, also shares global network characteristics such as small-world and scale-free. So far, researchers have proposed some metrics to measure software complexity, utilized the theories of complex network, methods, which overcome some shortcomings of traditional complexity measurement methods, such as just focusing on the micro structures statistics and lack of structural integrality. It has become an interdisciplinary filed of complex network and traditional software engineering.In this paper, we first summarized and reviewed related work in the new field, and then abstracted software system into three hierarchical influence network models that are separate but mapping to each other. Eight java open-source software systems were selected as study cases. The results show that they have small-world and scale-free characteristics obviously. According to some common special phenomenon of the corpus network models, combined with the principles of software engineering practice and based on the comparison of traditional software complexity metrics, we proposed a series of metrics to complexity of software system, which include the flowing three aspects,1).modularity measurement based on package influence network,2).change-proneness measurement based on feature influence network,3).cohesion measurement based on feature influence network. These metrics are all verified validity by theories and empirical validation. Finally, a case tool, namely Loach, is briefly introduced, which is developed for extracting and analysis software influence network models.