Adaptive learning is getting more and more attention in the modern age because it offers an alternative to the traditional "one-size fits-all" pedagogical approach. This paper presents an extended work that improves the efficiency of an ant colony optimization (ACO) model for adaptive learning, where a fuzzy knowledge extraction model was established to extract recommendation knowledge from learning experiences of past learners. Specifically, this research improves the model by devising a more efficient algorithm that requires smaller numbers of training cycles and learners to find high-quality adaptive learning paths. The key approach is to cut down non-contributing search subspaces effectively through the negative feedback given by past learners. Simulation results showed that this new ingredient added to the ACO-based model makes it more suitable for practical content recommenders.