An attributed network has nodes with attribute vectors. For community detection on an attributed network, we exploit the attributes to disentangle the potentially mixed topological structures. We describe a multiplex representation scheme for overlapping community detection in attributed networks and find consensus communities across layers of different connection structures. We test the method on Twitter, Facebook and Google+ networks and the results are comparable to state of the art. We show that the use of attribute vectors improves detection accuracy. Experiments on synthesized networks show that this method improves the detectability of communities.