Recent work suggested that, in the age of data overload produced by sensors, social media, and IoT devices, a key new type of network transport protocols will be one that offers representative summaries of requested data, retrieved at a consumer-controlled degree of granularity. Given the over-abundance of data, consumers will seldom need all data on a topic, but rather will increasingly favor an appropriate sampling for summarization purposes. The paper explores such sampling as a novel service enabled by information-centric networking paradigms that name data objects, not hosts. By naming data objects, it becomes possible to selectively retrieve them, but the properties of the resulting sampling depend on the naming scheme. This paper describes an automated object naming service, called Espresso, that facilitates content sampling over information- centric networks. We show how Espresso, combined with a trivial retrieval policy, translates the sampling problem into a naming problem, and customizes the naming to different applications' sampling needs. Experimental results show that the computational overhead of automated naming is affordable. The service is first evaluated in simulation, demonstrating a higher sampled-data utility to the consumer, while balancing retrieved data importance and diversity. Social network applications are then introduced, where naming is produced by Espresso. Results demonstrate the advantages of Espresso, compared to baselines, in terms of retrieving meaningful media data summaries.