Automated extraction of concepts and detection of dependencies are important for analyzing educational texts and supporting applications like curriculum development, learning path suggestions, and intelligent tutoring systems. The aim of these tasks is to automatically extract key domain concepts from educational materials and find dependencies or prerequisites between them. Several approaches are put forward in recent studies, such as weakly supervised methods based on semantic embeddings and clustering, distant supervision techniques that make use of domain-specific glossaries, and supervised transformer-based models. Supervised models frequently exhibit high accuracy, but they rely significantly on large, manually labeled datasets. On the other hand, weakly and distantly supervised approaches drastically lower annotation costs, but they also face issues with seed quality, dictionary coverage, and cross-domain applicability. In order to guide future work in automated educational concept analysis, this study offers a thorough overview of current methods, frequently used datasets, assessment metrics, and recognized limits. It also addresses important issues and unresolved research gaps.