COMET: A ML-Based Tool for Evaluating the Effectiveness of Software Design Communication

Communication is one of the most demanding activities in software development. The effectiveness of communication can be measured by analyzing three interpersonal communication dimensions: Active Discussion, Creative Conflict, and Conversation Management. Previous work relied on manually labeling the communication dimensions to analyze the effectiveness of software design discussions, a process that is time-consuming and not applicable to real-time use. In this study, natural language processing and supervised machine learning are used to create COMET, a tool for automatic classification and assessment of the effectiveness of interpersonal communication during software design meetings. To determine the optimal classification approach, nine different classifiers are examined. The classifier model that performed the best is Random Forest which managed to achieve 93.66% accuracy, 93.76% precision, and 93.63% recall when trained and tested with a stratified 10-fold cross-validation technique.