Weibo and other online social networks have be-come the basic platform for information dissemination and diffusion. Prediction of information forwarding in social networks has attracted a lot of research work, especially how to effectively consider the characteristics of users and information, as well as the interaction between them. In this paper, we construct a weighted heterogeneous network based on users and information, and propose an improved heterogeneous network graph representation algorithm Mpath-wMetapath2vec to generate low dimensional representation for forwarding prediction. The experimental results on Weibo dataset show that our model outperforms the algorithms without considering node features and other graph vector representation algorithms.