An Improved Assessing Requirements Quality with ML Methods
Serda Hauser, Stephan Muller-Schmeer, Bogdan Franczyk, Sabine Radomski · 2022 14th International Conference on Computational Intelligence and Communication Networks (CICN) · 2022
This paper describes a method for classification and quality analysis of unspecified information technology (IT) requirements in enterprises using machine learning (ML). During classification, unspecified IT requirements are classified into predefined categories and chapters using a term frequency algorithm and a neural network (NN) based on Tensorflow, which makes a multiclass decision. It bases on the encoding of IT requirements into tensors. The unspecified IT requirements are subject to a quality analysis, and a stencil value is obtained by two algorithms: a pattern template and a Tensorflow model (NN). Both approaches use the Natural Language Toolkit (NLTK) and its Part-of-Speech (PoS) tags. A stencil algorithm uses a PoS-based stencil to compute the stencil value. The Tensorflow model uses other key performance indicators (KPI) and features based on the PoS to make a binary decision about the quality of the requirements. The template thus provides the requirement author with a quality analysis for a possible revision. The described methodology intends to support DevOps in requirements classification and help developers to develop valid artifacts from requirements with little interpretation effort. This paper presents a model driven engineering approach that provides automation and software engineering techniques for managing information in a system (PRIMS).