From an interview with Scott Kirsner, of the newly opened up Variety.
Variety: What are the challenges of measuring the popularity of videos across multiple sites? Is a view on MySpace really the same thing as a view on Revver or YouTube?
VVC: No, different sites count a video as being viewed at different points in the viewing cycle, and of course some sites still don't publish view data at all. It's precisely because of the difficulty of verifying and normalising view data that we don't use view data as the basis for our video charts. We collect our own data, measuring the number of people embedding and linking to videos on their own blogs and personal websites. This buzz metric lets us publish a list of the most popular videos that's unaffected by definitional discrepancies, view-fraud or marketing expenditure.
Variety: Do you feel that most sites' data about the number of times a video is viewed is pretty reliable?
VVC: No, absolutely not. With many video sites, you can simply open up the page with your video on it and set your browser to automatically refresh the page. This is a really simple way to inflate viewing figures, requiring almost no specialist knowledge.
We estimate that at least 200,000 videos a day are currently being uploaded to video sharing sites. Getting your video noticed amongst this ocean of content is hard. So the prize for getting your video onto the homepages or most viewed lists of MySpace or YouTube is huge, and the incentive to game these sites really strong.
To combat this, some video sites are beginning to use third parties to verify view figures, particularly in cases where a marketing agency has paid to have their video campaign placed and promoted on the site.
Variety: What about compiling an accurate number of views for a video that may be posted on multiple sites, by multiple people?
VVC: That's hard. We have tools to help us detect the same video on multiple sites (or on the same site, posted by different people) and we aggregate figures where this happens. Some of the video sites are working on audio and visual 'fingerprinting' but these techniques are being developed chiefly to detect copyright-infringing clips. They haven't yet been proven to work at Internet-scale and don't address some of the subtler issues of content identity that we're concerned with. From our point of view, a longer clip or a differently edited clip may be materially different content but fingerprinting techniques, which are based on sampling, are not designed to care about these issues, because they're not particularly salient to the enforcement of intellectual property rights.
Variety: It seems like we don't really know much about how much video content is being viewed on sites like NBC.com or BBC.co.uk or ABC.com, do we? Those sites don't seem to present the number of views they deliver, for a show like "Lost," for example, as public information....
VVC: This is something we'd like to work on, but it will require greater co-operation from the sites themselves. We'd be delighted to talk to any large media outlet that delivers a significant number of video streams.
Variety: Are there other problems or barriers that still exist, in terms of measuring which video content people are watching?
VVC: Sure. There are other well-known deceptive practices designed to increase viewing figures. Many videos are misleadingly titled. And there's an increasing tendency to manipulate the thumbnail image produced for the video by briefly inserting an unrelated still image halfway into the video. Images of bare skin tend to work pretty well. And deceptive practices aside, legitimate, paid-for marketing activities are increasingly being used to drive traffic to commercially produced video content that's hosted on YouTube or MySpace. So it's incredibly difficult to get a real measure of what, in a grass roots, organic sense is truly popular just from viewing figures alone.
Recent Comments