Automatic detection of constraints in software documentation
Date
2021
Authors
Ghosh, Joy, author
Moreno Cubillos, Laura, advisor
Ghosh, Sudipto, committee member
Vijayasarathy, Leo, committee member
Journal Title
Journal ISSN
Volume Title
Abstract
Software documentation is an important resource when maintaining and evolving software, as it supports developers in program understanding. To keep it up to date, developers need to verify that all the constraints affected by a change in source code are consistently described in the documentation. The process of detecting all the constraints in the documentation and cross-checking the constraints in the source code is time-consuming. An approach capable of automatically identifying software constraints in documentation could facilitate the process of detecting constraints, which are necessary to cross-check documentation and source code. In this thesis, we explore different machine learning algorithms to build binary classification models that assign sentences extracted from software documentation to one of two categories: constraints and non-constraints. The models are trained on a data set that consists of 368 manually tagged sentences from four open-source software systems. We evaluate the performance of the different models (Decision tree, Naive Bayes, Support Vector Machine, Fine-tuned BERT) based on precision, recall and F1-score. Our best model (i.e., a decision tree featuring bigrams) was able to achieve 74.0% precision, 83.8% recall and an F1-score of 0.79. This suggests that our results are promising and that it is possible to build machine learning based models for the automatic detection of constraints in the software documentation.