Although self-regulated learning (SRL) plays an important role in supporting online learning performance, the lack of student self-regulation skills poses a persistent problem to many educators. Recommender systems have the potential to promote SRL by delivering personalized feedback and tailoring learning strategies to meet individual learners' needs. However, fully-supervised learning recommenders require extensive data, while unsupervised learning lacks expert or teacher guidance. To address these problems, we propose a theory blended semi-supervised learning method to implement a SRL recommender system. Specifically, we developed a fully-supervised machine learning recommender and a semi-supervised machine learning recommender to provide relevant suggestions for enhancing student SRL skills. In the first phase of our study, we investigated whether a semi-supervised algorithm could predict students' SRL more effectively than a fully supervised algorithm. In the second phase, we developed two recommender systems, one based on the semi-supervised algorithm and the other on the fully supervised one, to provide relevant recommendations for enhancing student SRL skills. We conducted an experiment involving two distinct groups of online students who received SRL suggestions from the respective recommenders. Our results showed that the semi-supervised learning recommender significantly enhanced students' SRL compared to the fully-supervised one, with a large effect size (F(1, 81) = 12.879, partial [eta][superscript 2] = 0.14). This research offers insights for enhancing SRL recommender systems and serves as a practical guide for future studies.