With the increasing number of wireless mobile devices, the possibility of mobile communications without network infrastructure is becoming a reality. The decentralized mobile social network (MSN) is a paradigm where nodes can move freely and organize themselves arbitrarily. Therefore, the key challenge is to find routes that can provide satisfactory delivery performance and low average end-to-end delay. Because of the rapid changes of the contact graph's topology in MSN, we require an efficient data forwarding mechanism over the temporal contact graph. This paper proposes a High-Efficiency Routing protocol based on Social metrics in mobile social network (HERS) to decrease the cost of exchanging information. HERS uses the local information including egocentric betweenness, similarity and contact strength to select the next relay node. Then we use the logistic regression to derive the weights of above three metrics. We conduct simulations on real trace datasets, i.e., Cambridge dataset and Infocom05 dataset, and the results demonstrate that HERS outperforms the state-of-the-art SimBet Routing in delivering messages and significantly reduces the average end-to-end delay.