Wednesday, April 08, 2009
The depressing truth about web stardom
Persistence actually reduces the chances of success, according to a new study of YouTube stats
The web is a classic "attention economy"--a system in which people compete for attention. Ask many experts for the secret of success in such a system and they're likely to point to persistence and high quality content.
But a study of the success of videos uploaded to YouTube suggest that quality has little affect on success and persistence seems to actually reduce it.
Fang Wu and Bernardo Huberman at HP Labs in Palo Alto, studied the hit rates of some 10 million videos uploaded by 600,000 users before 30 April 2008. They classified a success as a video that is among the top 1 per cent of those viewed.
Their rather depressing finding is that "the more frequently an individual uploads content the less likely it is that it will reach a success threshold."
That much is clear from their data. What is rather less clear is why this happens. After all, uploaders clearly get better at producing videos as they persist, presumable because their video editing and uploading skills improve. And in fact video producers receive higher ratings for their later videos, even as their success rate declines.
Wu and Huberman suggest a reason: "when a producer submits several videos over time, their novelty and hence their appeal to a wide audience tends to decrease." Perhaps.
A more puzzling question is why the producers persist in the face of declining audience figures. Here Wu and Huberman are a little more convincing. They argue that like gamblers, video producers overestimate the chance of winning when the probabilities are small.
Ref: arxiv.org/abs/0904.0489 :Persistence and Success in the Attention Economy
Massachusetts Institute of Technology
Comments
It is also partially because attention gets divided between many videos rather than concentrating on one. A producer is in a sense competing against himself for a view by producing several videos for the same audience.
It would be more useful to see the statistic of aggregate video views of a prolific producers work compared to the views of a successful viral video from a less prolific collection.
seoetry
04/08/2009
Posts:1
gabrielg01
04/08/2009
Posts:396
One day Wu and Huberman are sitting around in their tragically un-hip little Palo Alto lab, starved for attention outside the narrow circle of similar dweebs, and "Youtube" appears to them in flaming letters across the sky!
"Maybe our little bag of statistical tricks can make us hip after all!"
Grind grind grind grind grind grind grind... and behold! A flaming insight!
Or not.
Let's suppose for one second that Wu and Huberman had applied their simple-minded model to Gawker, a very exclusive internet venue where only about 1% of readers are allowed to post. Here success will be defined as "allowed to post," and after we grind grind grind grind grind a lot of data, we discover that there is a 100% correlation between ever succeeding and success with your very first submission!
Absolutely no one who doesn't succeed with their very first submission ever succeeds!
No correlation with persistence!
No correlation with "quality," according to whatever ad hoc definition of "quality" Wu2 and Huberman2 invent!
Fantastic! Stop the presses! Pay attention to us!
Meanwhile, returning to the inner reality of the internet, someone notices that buried deep within the fine print of Gawker, there's a very discouraging little clause.
"If we don't like your first submission, we will never even look at anything else you may ever submit!"
Ouch!
Youtube works on almost exactly the same principle. The tiniest glimmer of possible popularity depends on making it to the level of "featured videos," and that depends on a relatively miniscule gang of selecters selecting your video. When those guys write you off and attach the "gomer" tag to your moniker, you're done.
There isn't a whole lot of mystery here, and what Wu and Huberman have actually demonstrated is the totally incurable un-hipness of Palo Alto lab rats.
Jacob Freeze
04/08/2009
Posts:4
This is (not far from) the same sort of thoughtless and over-interpreted data-mongering that Wall Street quants employed to "prove" that credit default swaps entail infinitesimal risk, and it's already past time to insist on a little more real-world experience and a little less confidence in an insulated bag of tricks, in the quant curriculum.
The demographic of "quantitative dweebs" has done an incredible amount of damage to all the rest of us, by promoting a phony confidence in their simple-minded models.
Enough is enough!
Jacob Freeze
04/08/2009
Posts:4
The same sort of self-selection occurs in gambling and investing in the stock market. Beginner's luck does exist... if you ignore all the beginners who weren't lucky.
vgogan
04/09/2009
Posts:1