Mining And Analysis Of Open Source Software Trustworthy Evidence

Trustworthy analysis of open source software (OSS) is the general designationwhich refers to analysis and evaluation of software production and developmentmovement in open source software evolution process, and has great significance forindustrialization and large-scale application of OSS. The OSS evolution processpresents a series of new characters: openness, dynamic, networking and diversification,which accumulated abundant original data on the one hand, and put forward some newrequirements for obtaining and quantifying of OSS trustworthy evidence on the otherhand. The traditional analysis of software trustworthiness are mainly directed atsoftware production, such as source code, document etc, which is obtained mostly bymanual way. The data in software development process is still not fully utilized andresearched for trustworthy analysis. It is difficult to satisfy the requirement ofevolvement process and automatic comprehensive trustworthy analysis. So based onthe OSS data in development process and community which is public and can beobtained freely, research on model and framework, mining and analyzing technologyof evidence for OSS trustworthy evaluation would have important implications on boththeory and practice.Aimed at above question and challenges, the thesis carried out the research ontrustworthy evidence framework and trustworthy evidence for development team inOSS projects with the method of empirical software engineering. Firstly, the internalrelations between open source software data and software attributes are studied, onekind of comprehensive OSS automatic trustworthy evidence framework based onInternet is built. The framework describes software production quality with softwareproduction evidence, and describes participants, structure and collaboration characterswith development team evidence. Then, the relation between development team andsoftware trustworthy is discovered, some kinds of development evidence are mined andanalyzed, and some new kinds of evidence are presented. The feasibility and validity ofthe trustworthy evidence framework are validated. Main contributions of thedissertation are as follows:(1) To the problem of trustworthy evaluation in OSS evolution process, the OSSautomatic trustworthy evidence framework (OSS-ATEF) is built. It includes bothsoftware production evidence and development team evidence, and can satisfy thetrustworthy evaluation in whole software evolution process. The framework solves theproblem of mapping Internet data to software trustworthy evidence. The tools which canselect, compute and obtain trustworthy evidence automatically are presented, and thefeasibility of the framework is confirmed by two famous OSS projects.(2) To the problem of extraction and quantification of contribution and collaboration characters in development team evidence, the development log data ismapped to contribution evidence with a calculation method of developer contribution,and the method to build committer network is proposed. Then, the experiment fordeveloper behavior is built based on the data of Gnome community, the Pareto principlein OSS development process is proved, and the different proportion in different kinds ofdevelopment behavior is discovered. Analysis of committer network reveals thepreparatory distribution law of collaboration is found out, and presents the empiricalbasis of collaboration development characters in OSS evolution process.(3) To the problem of extracting team structure characters in OSS, the experimentfor roles structure in OSS projects with SourceForge community is built. Firstly, thestatistical analysis of distribution and signification of every role types presented, and thestrong relationship between role structure and project’s rank in the open sourcecommunity is certified. Then, the role structure classification model of OSS projects isproposed by the way of relative centrality and absolute centrality, which is the basis onthe way to make use of trustworthy evidence about the structure of OSS software team.(4) According to achieve the high-dimension feature of OSS project’s rolestructure, based on the classification model of role structure of OSS projects, theexperiment to mining the frequent patterns of roles in OSS project with Apriorialgorithm is built. Firstly, all of the target projects are divided three ranges by their rank,and the situation of support of different role configuration is computed in differentrange. The result expressed there is significant difference of support in different range.Then, one kind of classification method is constructed with the projects in high rankrange, and presents the empirical basis for role structure in OSS-ATEF.