Identifying Media in the Modern Era

Steven Posick, ESPN associate director, Enterprise Software Development,started by defining terms. What is identity? Identity is a characteristic of an Object that uniquely distinguishes it from all other objects within a Domain, which is defined as "a set of common requirements, terminology and functionality." A source of authority is an authoritative source or source of record for data or a domain and may be an organizer,  service or an algorithm and implements the requirements, terminology and so on. What is media? "For the terms of this presentation it is data that describes or depicts the events that transpired within an interval or point of time," said Posick, noting it was the simplest definition he could come up with given the constraints of the presentation. The domains of media identity are a hierarchy of identity domains in which media may be identified and classified. A media element is the smallest unit of media, representing the smallest interval or point of time. A media segment is a sets of media elements that represents a continuous interval of time.

Use cases include searching for media associated to real or fictitious events, finding the best media file/instance to act upon; finding a suitable alternative for an instance of media and searching for media containing or contained within other media (the latter being "something we commonly have to do at ESPN," said Posick, who described cutting highlights out of a larger piece of media).

The domains of media identity include events (real world events, movies), media (episodes), material (unique set of pixels), clone (unique set of bits), instance (physical/tangible instance, a file or stream) and segment/element (atomic elements, parts or sub-sections). Identity domains give the ability to create an ontology showing the relationship of these identities in a simple fashion. He also showed these relationships as a hierarchical "branching tree" graphic as well as groupings by media type and other parameters. "When searching for media of real or fictitious events, it's assumed the user knows what event they're searching for and therefore can find the media, then the material and then the instances," said Posick, who noted that this chain of events can also take place in reverse, i.e., starting with the known instance or file.

He showed a use case of finding a suitable alternative for an instance of media that may be in a different format. "You can use media and material identity to find other instances/files," said Posick. "Searching for media contained with other media, you have to know the media you're searching for. Then you can use composition, or component parts, to determine source of that media and then find instances/files for media containing the source media."

In order to maintain the relationship, media identifiers have to be created and assigned through ingest or edit. For material identity, it's just like media identity, said Posick. Similarly, you have to create a new identity whenever it's not pixel-for-pixel identical. Media composition is a list of "pointers to media" each entry called a composition segment and each composition segment representing a media segment and a contiguous block of source media.

Searching for media associated to an event, media is associated to one or more events, material is associated to media and media instances are associated to material (and comprised of media segments/elements). Searching for media containing other media, each media asset, material asset and media instance has a composition. Where that leaves us is with a simple iterative process for searching for media. "You search for the media source, and then query that for media segments, which can be ordered by any relevancy you want," Posick said. "You extract the identities for those segments and then search again with the same process, iterating until you find the results you want."