The lack of effective support for lifelong learning remains a persistent challenge within professional associations, often leading to fragmented pathways, unstructured course selection, and inefficient use of resources. In such contexts, traditional collaborative filtering methods are unsuitable due to data sparsity and course attendance policies: most users attend only a few courses and courses are available only for limited periods. In this paper, we examine the extent to which metadata-driven recommendation approaches and diverse recommendation architectures, particularly those leveraging semantic item identifiers, can serve as a viable solution for course recommendation in these professional scenarios. Experiments on historical course data collected from an engineers’ association training school show that, under a partitioning strategy that closely reflects real-world conditions, the proposed approaches generate highly relevant recommendations, demonstrating their potential to effectively support course recommendation in such environments.