Online learning provides learners with appropriate learning environments and a large amount of various educational resources, while eliminating the constraints of time and space. How to optimize the learning experience for learners has become a top priority. We focus on helping learners choose suitable resources, maintaining their enthusiasm for learning, and preventing them from dropping out. In this paper, we propose a complete solution to address the above challenges. First, we design a semantic online learning environment based on resource-sharing, where learners can perform a variety of learning-related activities (e.g., share, access, and vote), and learners’ processes are described by knowledge graph. Then, we derive learners’ features from their activity records collected by web logs, and form learners’ feature models. Finally, we combine knowledge based embedding with collaborative learning to provide personalized recommendation services for learners.