Designing an approach for highlighting requirements from elicitation interviews

Abstract

Software projects start with correctly understanding what the problem is, why this is a problem and who should be involved to solve this problem. This is done by requirements elicitation, a phase of Requirements Engineering that highly relies on the communicative skills of the requirements engineer. While there are several techniques for requirements elicitation, most involve a conversation, between the requirements engineer and the client. Nowadays, most of these conversations are recorded, since online communication has spiked since the COVID-19 pandemic in 2019. This allows for the exploration of the transcripts of these conversations using Natural Language Processing (NLP). By using Natural Language Processing for Requirements Engineering (NLP4RE), we can extract the questions asked in the interview, after which we can classify whether these questions are relevant or not. As a final task, we can categorize these questions. By providing an overview of these relevant questions, with their category, to a requirements engineer and allowing them to read the answer to those questions, this approach provides a direct transfer of knowledge by showing parts of the conversation. This can uncover a missed requirement or clear out any misunderstandings. In this context, we propose our approach to highlighting requirements-relevant information in requirements elicitation interviews, by using NLP in combination with Artificial Intelligence. The performance of this tool is evaluated using tagged conversations from a Requirements Engineering course, given at Utrecht University.