Using Text Mining Techniques to Extract Rationale from Existing Documentation

Abstract

Software development and maintenance require making many decisions over the lifetime of the software. The decision problems, alternative solu- tions, and the arguments for and against these solutions comprise the sys- tem’s rationale. This information is potentially valuable as a record of the developer and maintainers’ intent. Unfortunately, this information is not explicitly captured in a structured form that can be easily analyzed. Still, while rationale is not explicitly captured, that does not mean that rationale is not captured at all—decisions are documented in many ways throughout the development process. This paper tackles the issue of extracting ra- tionale from text by describing a mechanism for using two existing tools, GATE (General Architecture for Text Engineering) and WEKA (Waikato Environment for Knowledge Analysis) to build classification models for text mining of rationale. We used this mechanism to evaluate different combinations of text features and machine learning algorithms to extract rationale from Chrome bug reports. Our results are comparable in accuracy to to those obtained by human annotators.