Requirements Complexity Definition and Classification using Natural Language Processing

Abstract

This paper addresses the risks and impacts from use of subjective assessment of requirements complexity in planning, estimation, and test phases of software development. The risks are to delivery schedule and product quality and impacts are significant functionalities planned for later cycles, compressed times for testing and late detection of high severity defects. A solution to objectively determine complexity and ranking of requirements that mitigates these risks and impacts is discussed. Applying natural language processing, key words are identified in the given set of requirements, and their weights measured to determine the complexity class distribution and ranking. This classification and the ranked requirements data is used in estimation, planning, and defining development and test execution sequence. Objectively determined complexity classes improves accuracy in test estimation and planning. The complexity-based test sequencing mitigates risk to product quality and execution schedule by early discovery of high severity defects providing the project team a sufficient time to recover. This also lowers the project risk to subsequent phases as ranked complex requirements are tested early and simpler requirements are tested at the end of the cycle. Results and benefits obtained in applying this solution to several projects are presented contrasting with projects that did not use this.