Title: Algorithmic and meta-algorithmic machine learning natural language processing approaches for stakeholder requirements classification

Authors: Arturo N. Villanueva Jr.; Steven J. Simske

Addresses: Colorado State University, Fort Collins, Colorado, USA ' Colorado State University, Fort Collins, Colorado, USA

Abstract: Requirements engineering begins with discovery at the outset of project acquisition. Documents typically used during this phase include statements of works (SOWs) and requests for proposals (RFPs). One of the first challenges of a systems engineer is to carefully classify requirements into appropriate bins for further processing. This manual process, fundamental to understanding stakeholder needs and architecting and designing the system(s) of interest, is often tedious, particularly for large projects that start out with thousands of requirements embedded in these documents, making the task ripe for automation. For this research, we investigate multiple combinations of algorithms and meta-algorithms to glean insight as to how well they perform on this aspect of one of the more mundane aspects of requirements engineering. We obtain, by running various training corpora representing multiple industries through our pipelines of (meta-)algorithms, some understanding of what works best and what and how they could be improved.

Keywords: natural language processing; NLP; classification; meta-algorithms; machine learning; statements of works; SOWs; requests for proposals; RFPs; document classification.

DOI: 10.1504/IJCSYSE.2022.131033

International Journal of Computational Systems Engineering, 2022 Vol.7 No.1, pp.41 - 56

Received: 19 May 2022
Received in revised form: 04 Apr 2023
Accepted: 04 Apr 2023
Published online: 19 May 2023 *