An Integration System for Software Product Archiving, Analysis, and Retrieving System

点击翻译按钮获取中文摘要

In this research, we have developed a component search system, named SPARS-J, which treats the source files of Java classes as components. In the system, we use a novel ranking method to narrow retrieved software components from reusable libraries. We define a component rank model based on a graph representation scheme of the component library. In this model, a collection of software components is represented as a weighted directed graph, i.e., the nodes of the graph correspond to components and the edges linking the nodes correspond to cross component usage. Similar components are clustered into one node so that the effect of duplicated components is removed. The nodes in the graph are ranked by their weights, which are defined as the elements of the eigenvector of an adjacent matrix for the directed graph. The resulting rank, named component rank, is used to prioritize the query result so that highly ranked components are quickly seen by the user. The idea behind component rank originates from computing impact factors (called influence weights) of published papers. This approach has been extended to ranking Web documents on the Internet. SPARS-J has been applied to various collections of Java programs, such as JDK, programs downloaded from the Internet, and business applications from two companies. The results show that a class frequently invoked by other classes (such as those that implement fundamental and standard data structures) generally has a high rank, and that nonstandard and special classes typically have a low ranking. Two companies use SPARS-J for automatic management of their software assets, and SPARS-J shows very promising results.