Various prior studies have leveraged cloud computing and big data techniques to promote adaptive micro open learning. However, this novel way of open education resource (OER) delivery and access suffers from the cold start problem of learner information. In this paper, we introduce a service oriented solution to assist OER providers and instructors to deal with the sparsity of data in OER recommendation using an ontological approach. Learners' features are predicted by spreading activation and demographic similarity based inference. An evolutionary algorithm is provided to realize the OER recommendation in terms of heuristic rules.