From Descriptive to Predictive: Forecasting Emerging Research Areas in Software Traceability Using NLP from Systematic Studies

Systematic literature reviews (SLRs) and systematic mapping studies (SMSs) are common studies in any discipline to describe and classify past works, and to inform a research field of potential new areas of investigation. This last task is typically achieved by observing gaps in past works, and hinting at the possibility of future research in those gaps. Using an NLP-driven methodology, this paper proposes a meta-analysis to extend current systematic methodologies of literature reviews and mapping studies. Our work leverages a Word2Vec model, pre-trained in the software engineering domain, and is combined with a time series analysis. Our aim is to forecast future trajectories of research outlined in systematic studies, rather than just describing them. Using the same dataset from our own previous mapping study, we were able to go beyond descriptively analysing the data that we gathered, or to barely 'guess' future directions. In this paper, we show how recent advancements in the field of our SMS, and the use of time series, enabled us to forecast future trends in the same field. Our proposed methodology sets a precedent for exploring the potential of language models coupled with time series in the context of systematically reviewing the literature.