In order to help learners identify suitable learning resources and sequences sets for personalized learning from a large number of online learning resources, a label propagation algorithm on directed edge weights (LPADEW) was proposed for clustering micro-learning units and sequence that belong to the same learning period for specific learners. Two improvements to the label propagation algorithm were made: the label update order was determined according to the utilization rate of the unit node, the randomness of node update order was reduced. The accumulated value of the directed edge weights of the pre-neighbor and the post-neighbor of the current element node was used to update the label, and the label weight was used in the label updating strategy, which can not only reduce the randomness of label updating, but also avoids the formation of giant cluster. Experimental result shows that LPADEW algorithm performs well on real datasets of micro-learning and synthetic datasets.