Validation of Inspection Reviews over Variable Features Set Threshold

Background: Mining software requirement reviews involve natural language processing (NLP) to efficiently validate a true-fault as useful and false-positive as non-useful. Aim: The aim of this paper is to evaluate our proposed mining approach to automate the validation of requirement reviews generated during an inspection of NL requirements document. Method: Our approach utilized two training models; one from requirement reviews and other from online movies. We conducted an empirical study to test our approach using part of speech (POS) against these two trained models and observed trends w.r.t. F-measure and G-mean along with percentage of features used to train two models. Results: The results showed that using training reviews from two different domains report similar trend across evaluation metrics. Our results show that the most stable and promising validation results for F-measure and G-mean are obtained when a model over inspection and movies reviews are trained using feature set threshold value 65% and 45% respectively.