How to identify influencers in online media and social networks? A vast number of methods have been proposed to this aim, ranging from measuring the size of information cascades (e.g. counting message forwardings) to weighting user profiles depending on the considered topic (e.g. with Personalized PageRank). This paper proposes a new approach to influencer identification, rooted on two remarks. Firstly, some prior knowledgebased information can be available in real-world settings: the fact that some persons truly are influencers with respect to a given topic. Secondly, any potential influencer can operate through several channels: newspaper articles, blogs and/or social networks. To accommodate multi-channel influences in a compound way, a feature construction mechanism is proposed to describe potential influencers, by automatically extracting features from interaction graphs (e.g. twitter social network) and contents (tweets). Finally, a supervised machine learning approach is proposed to tackle influencer identification, automatically finding the characteristic influence patterns based on the available features and the (handful of) known influencers. Empirical validation on a dataset of over 10 million authors and 110 million tweets tackles influencer identification with respect to five topics (human ressources, e-marketing, fashion, high-tech, wine). The lessons learned are twofold: firstly, a small set of features is sufficient to identify influencers; secondly, characteristic influence patterns exhibit significant differences depending on the topic of interest