Automated Labeling and Classification of Business Rules from Software Requirement Specifications

Abstract

Business Rules (BRs) are a critical artifact in the requirements elicitation phase of the software development life cycle. Several taxonomies have been proposed for classification of BRs. In this paper, we utilize Ross's BR classification schema and present an approach to automatically label and classify BRs along this schema. Our approach uses Data Programming (DP) for generating labeled training data needed for training two deep learning-based models to classify the BRs. We obtained an average labeling accuracy of 0.73 for all the BR classes using DP. Upon evaluating the approach on industryspecific dataset, we obtained highest weighted F-score (0.69) with a Bi-LSTM with attention-based model.