With the proliferation of social Web applications, users can now collaboratively author, share and access hypermedia learning resources, contributing to richer learning experiences outside formal education. These resources may or may not be educational. However, they can be harnessed for educational purposes by adapting and personalizing them to different learner needs. We propose to collaboratively annotate learning resources with a lightweight resource annotation metadata schema complemented by a folksonomy-derived semantic model. The annotation metadata schema follows a novel policy to associate numerical ratings to learners' subjective expression of opinions on learning resources. On the other hand, the semantic model serves to support a collaborative resource recommendation algorithm based on the k-nearest neighbour approach. Proof-of-concept is demonstrated via a prototype Web-based recommender. Preliminary results indicate learners are confident with the recommended learning resources in terms of accuracy, usefulness, novelty and information adequacy. Formalization of folksonomy in annotating learning resources as well as in opinion mining from the crowd to rate resources may not only support personalization but also engage learners more effectively in technology-enhanced learning.