Scoping reviews are a type of research synthesis that aim to map the literature on a particular topic or research area. Though originally intended to provide a quick overview of a field of research, scoping review teams have been overwhelmed in recent years by a deluge of available research literature. This work presents the interdisciplinary development of a semi-automated scoping review methodology aimed at increasing the objectivity and speed of discovery in scoping reviews as well as the scalability of the scoping review process to datasets with tens of thousands of publications. To this end we leverage modern representation learning algorithms based on transformer models and established clustering methods to discover evidence maps, key themes within the data, knowledge gaps within the literature, and assess the feasibility of follow-on systematic reviews within a certain topic. To demonstrate the wide applicability of this methodology, we apply the here proposed semi-automated method to two separate datasets, a Virtual Human dataset with more than 30,000 peer-reviewed academic articles and a smaller Self-Avatar dataset with less than 500 peer-reviewed articles. To enable collaboration, we provide full access to analyzed datasets, keyword and author word clouds, as well as interactive evidence maps.