Revealing Crosscutting Concerns in Textual Requirements Documents: An Exploratory Study with Industry Systems

Abstract

It is well-known that effective requirements analysis plays a crucial role in the quality of software systems. However, the scattered and tangled nature of certain system's concerns can hinder the proper understanding and treatment of import requirements. A key goal of prominent Aspect-Oriented Requirement Engineering (AORE) techniques, such as EA-Miner and Theme/Doc, is to support the automatic identification of crosscutting concerns at textual requirements documents. However, it is still unknown whether and which of these approaches produce accurate results in large text documents and according to the software engineers' expectations. In this context, this paper presents an analysis regarding the accuracy of the aforementioned AORE approaches when processing requirements of two industry software systems. Around 300 pages of requirements descriptions in these systems were the target of our investigation. In general, EA-Miner suffered more than Theme/Doc from the incompleteness and inconsistencies of requirements documents. In addition, other factors can differently influence each approach's accuracy, such as: the participation of requirements engineers, and the level of details provided in the requirements document.