A Machine Learning Toolkit for Selecting Studies and Topics in Systematic Literature Reviews

Abstract

Scholars conduct systematic literature reviews to summarize knowledge and identify gaps in understanding. Machine learning can assist researchers in carrying out these studies. This paper introduces a machine learning toolkit that employs Network Analysis and Natural Language Processing methods to extract textual features and categorize academic papers. The toolkit comprises two algorithms that enable researchers to: (a) select relevant studies for a given theme; and (b) identify the main topics within that theme. We demonstrate the effectiveness of our toolkit by analyzing three streams of literature: cobranding, coopetition, and the psychological resilience of entrepreneurs. By comparing the results obtained through our toolkit with previously published literature reviews, we highlight its advantages in enhancing transparency, coherence, and comprehensiveness in literature reviews. We also provide quantitative evidence about the toolkit's efficacy in addressing the challenges inherent in conducting a literature review, as compared with state-of-the-art Natural Language Processing methods. Finally, we discuss the critical role of researchers in implementing and overseeing a literature review aided by our toolkit.