The State of Metadata in 2011
The big issue with automated metadata creation is that it has always been less than accurate. While most of us love the concept of automated metadata insertion, the reality is that automation accuracy will never be perfect.
Rather than pining for the days when we search more but key less, I expect that 2011 will witness the advent of false-positive-impervious algorithms.
What might they look like, and what benefit do they provide? I can’t speak to the complexities of the algorithms, but the benefit is one of allowing many pieces of information to be stacked on a single frame of video, with the system weighing the potential relevance and accuracy of an intended search against said metadata.
It’s not a far-fetched leap, as researchers for search engines such as Google and Bing spend their days trying to discern the intent of searches, even while mounds of metadata pile up against top sites. Think of it as searching for a needle in 90 haystacks: If you can eliminate all but three of them, the search will probably take less time and be more accurate.
How many fields of data are possible for a single frame of video? I don’t know the answer for all the systems on the market, but Pictron has—for almost a decade—touted the fact that it can handle 32,000 fields of data for every frame of video.
Metadata Discovery Is a Two-Way Street
The idea of video discovery is often pitched as a way for searchers to find pertinent content, but using tags to identify a video is a bit mystifying: It assumes that the person (or machine) adding metadata to a particular video clip knows, months or years in advance of a search, what a searcher will identify as the keywords.
This is backward to a better approach—one that is slowly emerging.
Rather than expecting the searcher to learn a new taxonomy (or, to a greater extreme, a whole new language), the search itself would take the “Do you have any …?” question and use the searcher’s request to determine both the taxonomy for meta-indexing videos as well as the relevance of that taxonomy to the general searching population.
Think of it in terms of the difference between the old-school card games Old Maid and Go Fish. In the game of Old Maid, a player draws a card blindly, often receiving irrelevant or harmful cards; by contrast, in Go Fish, the player (our “searcher”) knows what he or she has in hand and makes a request to another player (our “video repository”) for a specific match to an in-hand card. Even if the request/search turns up empty, there’s less uncertainty.
The next step beyond “Do you have any …?” may very well be like another card game: Uno. In this game, all players—including our “searcher” and our “video repository”—have the ability to react to a highly variable landscape. In Uno, the color, card number, or wild card change constantly, driving a frenetic approach that allows for submission of similarities rather than exact matches.
Will the system be gamed, from time to time, by substitution submissions? Yes, but that’s where the false-positive algorithms come in to play. It’s still a whole lot better than today’s text tag approach, especially if we can begin to use images for atmospheric, environmental, and face-matching searches.
Location to the Nth Power
This one’s fairly obvious and written about so much that it’s almost become a truism: Location matters when it comes to search, even for video search.
Several companies in the space have tied metadata repositories against geolocation filters, thereby disallowing retrieval of content from outside a particular region, in keeping with their media conglomerate customer base.
Ideally, the geo filters will be turned off. But until that day, the next step seems to be a need to allow traveling residents of a particular geography to be able to access content from the road that they have permission to access when they are at home.
To do so requires a coordination of both content-and location-aware searches, coupled with an understanding of whether the request is coming from a permanent resident or someone passing through an area. While it’s hard to address a universal negative in metadata, location awareness needs to come of age quickly for paying customers of TV and other on-demand video subscription services.
Fine-grained metadata is the key to turning an existing library into interactive new products and features.
Companies and Suppliers Mentioned