Software System Networked Modeling And Key Node Identification Methodological Study

With the rapidly increasing demand for software functions,the structure scale of the software system is growing constantly and the interaction methods between different software are becoming more and more complicated,resulting in greater complexity of software systems which pose a negative effect on runtime reliability,management and maintenance of software systems.A feasible solution is to describe the structure of software systems reasonably and recognize the key parts of software systems accurately by introducing complex networks to the researches of software network.Hence,we can effectively explore the inherent structural features of software systems from a global and local perspective,which is of great significance for boosting reliability,maintaining software systems,prevention of vulnerabilities and error location.Therefore,by carrying out researches on various models of network,we proposed corresponding modeling methods and analyzed the research results efficiently.The main contents of this paper are as follows:Firstly,we concentrate on the networked modeling of software systems.On the one hand,investigating the integral feature of software systems by selecting the appropriate number of subsystems as components and establishing a macro network model from a larger granularity level;On the other hand,taking classes as the basic granularity units and improving the weighting algorithm of network of class scale by considering multiple influences between classes,which helps us to acquire the micro network model and make it easy to analyze the code details of software systems.Then,in terms of key nodes identification,different key node identification methods are designed with different emphasis.On the one hand,from the perspective of optimizing the ranking index of evaluation nodes,the robustness measure function is reconstructed by structural weighting,and the firefly algorithm of discretization is implemented by introducing the concept of commutator,by adding the initial solution structure based on good point set.Partition search,variable global attractiveness and adaptive random items enable the optimal search to converge accurately,thus achieving accurate identification of key nodes of the network.On the other hand,from the perspective of synthesizing different metrics,ListNet is introduced into node sorting to fuse different metrics,and multiple ListNet-based learners are integrated based on AdaBoost algorithm to further improve the generalization performance of ListNet and the overall key node identification has higher accuracy.Finally,in order to verify the effectiveness of the proposed method,the macro and micro network models are established respectively by analyzing the characteristics of the actual software network system 16000 and three typical piece of software.Based on the four network models,simulations are carried out for the proposed key nodes identification methods.From the simulation results,it is apparent that the established network models of different hierarchical level are able to effectively depict the structure features of software systems both globally and locally,which matches the actual situation of software and provides an accurate network model for further research.Compared with other similar type of methods,the two key nodes identification methods with different emphasis can achieve higher accuracy and people can choose corresponding one according to actual work to deal with a variety of application scenarios.