Open Access

ARTICLE

Using Artificial Intelligence Techniques in the Requirement Engineering Stage of Traditional SDLC Process

by Afam Okonkwo^*, Pius Onobhayedo, Charles Igah

Computer and Information Sciences, Pan Atlantic University, Ibeju-Lekki, Lagos, 73688, Nigeria

* Corresponding Author: Afam Okonkwo. Email:

Journal on Artificial Intelligence 2024, 6, 379-401. https://doi.org/10.32604/jai.2024.058649

Received 17 September 2024; Accepted 04 December 2024; Issue published 31 December 2024

Abstract

Artificial Intelligence, in general, and particularly Natural language Processing (NLP) has made unprecedented progress recently in many areas of life, automating and enabling a lot of activities such as speech recognition, language translations, search engines, and text-generations, among others. Software engineering and Software Development Life Cycle (SDLC) is also not left out. Indeed, one of the most critical starting points of SDLC is the requirement engineering stage which, traditionally, has been dominated by business analysts. Unfortunately, these analysts have always done the job not just in a monotonous way, but also in an error-prone, tedious, and inefficient manner, thus leading to poorly crafted works with lots of requirement creep and sometimes technical debts. This work, which is the first iteration in a series, looks at how this crucial initial stage could not just be automated but also improved using the latest techniques in Artificial Intelligence and NLP. Using the popular and available PROMISE dataset, the emphasis, for this first part, is on improving requirement engineering, particularly the classification of Functional and Non-functional Requirements. Transformer-powered BERT (Bidirectional Encoder Representations from Transformers) Large Language Model (LLM) was adopted with validation performances of 0.93, 0.88, and 0.88. The experimental results showed that Base-BERT LLM, its distilled counterpart, Distil-BERT, and its domain-specific version, Code-BERT, can be reliable in these tasks. We believe that our findings could encourage the adoption of LLM, such as BERT, in Requirement Engineering (RE)-related tasks like the FR/NFR classification. This kind of insight can help RE researchers as well as industry practitioners in their future work.

---

Keywords

---