Open Educational Resources (OER) are increasingly welcomed by people as the demand for education. However, how to make full use of them and make learners study efficiently could be some tackle problems to be solved. In this paper, we start data processing from Yale Open Courses and propose an efficient deep learning based model capturing context similarity by using word embeddings and LSTM to predict the most relevant sessions from any text paragraph from courses so that help to recommend courses or generate learning roads for OER users. The results of the experiment indicate that the model shows high performance on OER courses dataset in the experiment and also provides ideas for sessions' prediction and recommendation.