Determining the Level of Detail of Software Requirements

Requirements Engineering is an important step in a software engineering projects. It starts with requirement gathering and documentation of user requirements. If a requirement is documented at the appropriate level of detail, it will not only help improve the quality of the requirements consequently increasing the chances of a successful project, but will also help in the reuse of the same requirements for similar types of projects and problems. This research proposes the use of Parts of Speech (POS) to determine the level of detail of software requirements. These POS based features are used for classification and each requirement is classified as either of the three i.e. Low-level (a statement contains enough information to implement the intended functionality), Intermediate-level (a statement contains multiple tasks and further details will be required in order to implement the particular requirement), High-level (a statement contains many intended functions with incomplete details). These POS frequencies are extracted from natural language text of requirements and are given as input to three classifiers namely Logistic Regression, Multinomial Naïve Bayes (MNB), and Support Vector Machines (SVM). The paper evaluates the performance of each classifier with POSF based features and the traditional Term Frequency Inverse Document Frequency (TFIDF) based features. A 6.8% increase in accuracy for MNB has been observed when POSF based features are used. These results are based on the publicly available requirements that have an inconsistent level of detail. Availability of requirements with consistent level of detail can further improve the results. We used Structural Depth Metric to measure the consistency in level of detail.