Automated Validation of Requirement Reviews: A Machine Learning Approach

Software development is fault-prone especially during the fuzzy phases (requirements and design). Software inspections are commonly used in industry to detect and fix problems in requirements and design artifacts thereby mitigating the fault propagation to later phases where same faults are harder to find and fix. The output of an inspection process is natural language (NL) reviews that report the location and description of faults in software requirements specification document (SRS). The artifact author must manually read through the reviews and differentiate between true-faults and false-positives before fixing the faults. The time spent in making effective post-inspection decisions (number of true faults and deciding whether to re-inspect) could be spent in doing actual development work. The goal of this research is to automate the validation of inspection reviews, finding common patterns that describe high-quality requirements, identify fault prone requirements pre-inspection, and interrelated requirements to assist fixation of reported faults post-inspection. To accomplish these goals, this research employs various classification approaches, NL processing with semantic analysis and mining solutions from graph theory to requirement reviews and NL requirements. Initial results w.r.t. validation of inspection reviews have shown that our proposed approaches were able to successfully categorize useful and non-useful reviews.