Grupo Alarcos · Conference paper · 2016
Using text mining to locate concepts in execution traces
Mostefai Abdelkader, Ricardo Pérez del Castillo, Malki Mimoun
Concept location is a key activity during software modernization since it allows maintainers to exactly determine what pieces of source code support a specific concept. When concepts supported by an information system are getting outdated or misaligned, concept location becomes a time-consuming and error-prone task. Moreover, information systems embed significant business knowledge over time that is neither present nor documented anywhere else. For this reason, this paper proposes an approach to locate concepts in execution traces, which contain the actual and current information derived from the execution of the existing code. This approach, based on text segmentation tools, first retrieves and depicts relevant information from the execution traces, and then, it provides a mapping between such information and particular pieces of source code. The main idea of the approach is to take the file of the trace as text where the content of each function (i.e., method) appearing in the trace is a sentence composing the text and then applying the C99 algorithm to segment the text to a thematically coherent segments. The C99 is the state of the art algorithm proposed in text mining community to segment natural text. Each produced segment is composed by a set of functions or methods and bears a concept. Also, under this idea all techniques, proposed by text mining community, for text segmentation will be accessible to the software evolution community. We evaluated our approach on different execution scenarios of JHotDraw and ArgoUML.