Detecting syntactic ambiguity in requirements specification using Naive Bayes text classification algorithm

Abstract

Requirements engineering is the process of collecting software requirements from stakeholders, defining user expectations for a new product and resolution of requirements problems such as incompleteness, inconsistency and ambiguity of Software Requirements Specification (SRS). Ambiguities in SRS are considered as one of the main problems because one might interpret more than one way and multiple might interpret different interpretations as it might lead to confusion, waste efforts and time consumption. There are many types of ambiguity, which are lexical, syntactic, semantic, and pragmatic. This research focuses on sentence structure and grammar, which is called syntactic ambiguity. Three categories of approaches to detect ambiguities in requirements specification are manual approach, semi-automatic approach using natural language processing techniques and semi-automatic approach using machine learning techniques. Nonetheless, the manual approach requires a lot of efforts, human experts, time consumption and produce low detection rate of defects. On the other hand, some of the natural language processing techniques cannot be used in practical as well as produce misleading output in detecting ambiguity in SRS. Hence, the aim of this research is to apply semi-automatic approach using machine learning Naïve Bayes (NB) text classification technique based on n-gram modeling to detect syntactic ambiguities in Software Requirements Specification (SRS) because NB perform well and accurate in detecting ambiguity. In addition, the finding of this work also proved that NB text classifier achieved (80%) higher accuracy than manual approach (27%) in detecting syntactic ambiguity in SRS.