Educational Recommender Systems for Metacognitive and Self-Regulated Learning Support: An Algorithmic Systematic Review

This is the authors' abstract. We don't add key points for this paper.

Recommender systems (RSs) increasingly support self-regulated learning (SRL) through personalized resources, feedback, and strategy recommendations. However, traditional content based and collaborative filtering approaches suffer from cold start problems, while machine or deep learning methods raise interpretability concerns and LLM-based recommenders require evaluation of explanation faithfulness and hallucinations. This systematic review examines SRL-oriented recommender algorithms from an algorithmic and pedagogical perspective. Six databases were searched using predefined inclusion and exclusion criteria, yielding 25 eligible studies published between 2020 and 2025. The review traces the evolution from traditional filtering methods to machine or deep learning and emerging LLM-based approaches. Three major challenges were identified: cold start, limited explainability, and LLM hallucinations, alongside gaps in learner system interaction and metacognitive scaffolding. Future SRL-RS research should compare algorithms under controlled data sparsity, evaluate explanation faithfulness and source curation, preserve learner agency, integrate SRL strategies across three phases, and longitudinally assess SRL skill maintenance and transfer.