With the advance of social media networks, people are sharing contents in an unprecedented scale. This makes social networks such as microblogs an ideal place for spreading rumors. Although different types of information are available in a post on social media, traditional approaches in rumor detection leverage only the text of the post, which limits their accuracy in detection. In this paper, we propose a provenance-aware approach based on recurrent neural network to combine the provenance information and the text of the post itself to improve the accuracy of rumor detection. Experimental results on a real-world dataset show that our technique is able to outperform state-of-the-art approaches in rumor detection.