Micro Level Source Code Summarization of Optimal Set of Object Oriented Classes

Maintenance of software is effectively possible only after proper understanding of its modulesand components. Optimal summary of the different components can help a lot in understandingtheir internal logic. Hence source code summaries must include the description of thefunctionalities and the intent behind the existence of various source code entities. In this paperwe propose a new methodology for source code summarization performed at class level. Theproposed methodology identifies an optimal set of classes which are more beneficial forsummarization. Our generated summary includes all important information such as its properties,architectural details, its dependencies on other classes, internal implementation details andinheritance etc related information. This entire summary is generated in natural language usingan automated process. An empirical study of the proposed approach has been carried out overseven open source software and qualitative evaluation also has been carried out for certainimportant classes with help of experienced developers. In order to demonstrate the usefulness ofour proposed approach, our generated summary is compared with the competitive commonlyused summarization method. The comparison is done using pyramid approach which usesweights as a means to compare two or more summaries. The results clearly suggest that ourapproach generates summaries are quantitatively as well as qualitatively more useful fromcomprehension as well as maintenance view point.