In opportunistic networks, the nodes usually exploit a contact opportunity to perform hop-by-hop routing, since an end-to-end path between the source node and destination node may not exist. Most social-based routing protocols use social information extracted from real-world encounter networks to select an appropriate message relay. A protocol based on encounter history, however, takes time to build up a knowledge database from which to take routing decisions. An opportunistic routing protocol which extracts social information from multiple social networks, can be an alternative approach to avoid suboptimal paths due to partial information on encounters. While contact information changes constantly and it takes time to identify strong social ties, online social network ties remain rather stable and can be used to augment available partial contact information. In this paper, we propose a novel opportunistic routing approach, called ML-SOR (Multi-layer Social Network based Routing), which extracts social network information from multiple social contexts. To select an e↵ective forwarding node, ML-SOR measures the forwarding capability of a node when compared to an encountered node in terms of node centrality, tie strength and link prediction. These metrics are computed by ML-SOR on di↵erent social network layers. Trace driven simulations show that ML-SOR, when compared to other schemes, is able to deliver messages with high probability while keeping overhead ratio very small.