A Multinomial Naïve Bayes Classifier for identifying Actors and Use Cases from Software Requirement Specification documents

Abstract

A software Requirements Specification (SRS) document is an NL (Natural Language) written textual specification that documents the functional and non-functional requirements of the system and various expectations of clients in a software development project. To understand the different requirements of the system, developers make use of this SRS document. In this paper, we apply Naive Bayes classifiers - Multinomial and Gaussian over different SRS documents and classify the software requirement entities (Actors and Use Cases) using Machine Learning based methods. SRS documents of 28 different systems are considered for our purpose and we define labels for the entities Actor and Use Case. Multinomial Naive Bayes is a popular classifier because of its computational efficiency and relatively good predictive performance. Out of the classifiers tried out, the Multinomial Naive Bayes recognizes Actors and Use Cases with an accuracy of 91%. Actors and Use Cases can be extracted with high accuracy from the SRS documents using Multinomial Naive Bayes, which then can be used for plotting the Use Case diagram of the system. Automated UML (Unified Modeling Language) model generation approaches have a very prominent role in an agile development environment where requirements change frequently. In this work, we attempt to automate the Requirement Engineering (RE) phase that can improve and accelerate the entire Software Development Life Cycle (SDLC).