In this paper, we consider the problem of efficient social media access on mobile devices, and propose an Offline Online Social Media (O2SM) Middleware to: (i) rank the social media streams based the probability that a given user views a given content item, and (ii) invest the limited resources (network, energy, and storage) on prefetching only those social media streams that are most likely to be watched when mobile devices have good Internet connectivity. The ranking scheme leverages social network information to drive a logistic regression based technique that is subsequently exploited to design an utility based content prefetching mechanism. We implemented O2SM and a corresponding app, oFacebook, on Android platforms. We evaluated O2SM via trace data gathered from a user study with real world users executing oFacebook. Our experimental results indicate that O2SM exhibits superior viewing performance and energy efficiency for mobile social media apps; its lightweight nature makes it easily deployable on mobile platforms.