Information Retrieval - Based Solution for Software Requirements Classification and Mapping

Abstract

in software engineering, the process of requirements elicitation and specification is considered as a base for all other development process. This means that any fault or mistake in the requirements definition will negatively affect the whole process of software development and consequently affect the cost, time, and effort of the developers and users. Traditionally, the process of requirement elicitation and categorization was done manually and based on the experience of the developers. However, a lot of problem came up because of the absence of automatic approaches. This paper presents a novel approach to improve the process of software requirements classification and mapping. An Information Retrieval (IR) method, namely Latent Drichelt Allocation (LDA) will be used for classification process. A corpus of software requirements also will be built to be used as input space for LDA algorithm. Typically, each requirement will have a corresponding document in the corpus. We conducted two distinct experiments. The first one is to extract the topics of software requirements, and the second one is for mapping and linking any new requirement to the most existing relevant requirements. The results showed that the proposed approach overwhelmed the state-of-art approaches.