Systematic Review of the published Explainable Educational Recommendation Systems

Abstract

The goal of this paper is to systematically review the available literature on the topic of explainable recommendation systems in education, especially when recommendation systems are integrated as a part of learning management systems (LMS). The focus years for analyzing available literature are the years between 2010 and 2022, period when online learning is expanding and online learning platforms are continuously being developed, which makes these years relevant for scoping this review. The topic of interest in this research are recommendation algorithms whose results can be explained and interpreted. The first part of the methodology used in the paper utilizes an NLP-powered toolkit that automates a big part of the review process by automatically analyzing articles indexed in the IEEE Xplore, PubMed, Springer, Elsevier and MDPI digital libraries. The toolkit relies on the PRISMA methodology for standardizing systematic reviews. First, a quantitative analysis of all available literature is performed, followed by a qualitative analysis of the few selected articles which indeed focus on the explainability when implementing recommendation systems in educational context. The relevant articles are analyzed in detail and compared on multiple indicators like the field of work, tools and techniques used, and how explainability is achieved. The results show that although the amount of available research is growing and new learning management systems are being developed at a fast pace in the last few years, the explainability of the machine learning techniques used in the recommendation systems is not a popular topic among the researchers and developers with research interest in educational context. The amount of the available literature for explainable recommendation systems in educational environment is scarce, but is expected to grow following the global trend of explainable artificial intelligence $(\mathrm{x}\mathrm{A}\mathrm{I})$ as key technique for practical implementation of advanced AI models.