Um modelo para a análise de impacto em código fonte usando ontologias e recuperação de informação

Abstract

Changes are inevitable during the product lifecycle. These changes are due different needs, such as the knowledge evolution of business processes, environment or infrastructure changes, etc. Under these circumstances, it is crucial to have the control about the knowledge of what these changes mean to the system. Impact analysis represents the process that creates this knowledge. The impact analysis has a wide meaning for software development, such as the assessment of source code structures or the control of project management constraints. This thesis aims to present a model to analyze the impact in source code using ontologies in order to improve precision and recall of the identified source code structures compared to existing techniques. The use of ontologies integrates a semantic layer under the traditional techniques founded in syntactic analysis. To develop the impact analysis model, it was defined two different sub models: traceability and probability. The traceability model receives as input the source code and domain ontology and generates a traceability ontology populated with links between domain concepts (classes and properties) and source code units (classes, methods and attributes). These links are populated by a lexical and semantic analyzer that categorize, normalize (token generation, expansion and elimination) and compare (stemming) each token. With the traceability ontology and a change request, the probabilistic model classifies each traceability link using Bayesian Belief Networks. To the probability calculus, the node classification used the Google PageRank algorithm and the arrows used frequency analysis TFIDF and conceptual dependency, defined in this thesis. The impact analysis model was implemented as an Eclipse plugin and was empirically evaluated using three controlled experiments.