Twitter is very useful for figuring out "what is happening right now." For example, Twitter has been used to figure out where an earthquake
is happening. But instead of what's happening now, can it also be useful for telling what will happen in the future? Two researchers at HP Labs, Sitaram Asur and Bernardo A. Huberman, postulate that Twitter can actually be used to predict future box office results for movies
. By building a computer model around the tweets for 24 movies, they claimed were able to predict the opening weekend box office results with 97.3% accuracy. An accuracy of around 97% seems unnaturally high, so I dug a little further into the actual report
and found that both of these numbers were a result of combining each approach with the number of theaters associated with each release. Without taking into account the number of theaters in the release, tweets alone were able to predict at 93%. According to Asur, the correlation between theater counts alone and box office results is a paltry 39%. Considering that theater counts are probably a good indication of how well a studio thinks a film is going to perform, it seems that Hollywood should take notice of these forecasts.
The supposed gold-standard of box office predictions is Hollywood Stock Exchange, a prediction market. The researchers found that HSX had a 96.5% accuracy (when also combined with number of theaters). It's impressive that HSX gets a fair amount of accuracy despite it using "fake" currency. HSX is currently awaiting
a ruling from the U.S. Commodity Futures Trading Commission that will allow it to trade using real money. When that happens, the accuracy of its prediction market should increase. At that point it may be interesting to see how much more closely the two prediction systems compare, since the apparent edge that the Twitter algorithm has right now seems too small to be significant.
In any case, with Twitter, since it's also a communications medium, tweets about a movie not only indicate interest, but also can ultimately influence
a movie's sales at the box office. However, this effect seems distinctly different than the traditional notion of "observer effect," since the act of measuring the interest in a movie can actually occur without directly participating in the Twitter conversation. That said, if it becomes common practice to predict box office success by analyzing Twitter, the incentive to exploit this by spamming Twitter with tweets about an upcoming movie increases greatly. After all, the many predictions that foretold that Avatar
would break box office records certainly played a self-fulfilling part in its breaking of those box office records. Furthermore, Tweets were taken for the 7 days before the weekend -- which means there would not be enough time to make any adjustment to any marketing (in the event of a dire prediction), and certainly not to production budgets, which have already been spent.
So, while, in their report, Asur and Huberman conclude:
While in this study we focused on the problem of predicting box office revenues of movies for the sake of having a clear metric of comparison with other methods, this method can be extended to a large panoply of topics, ranging from the future rating of products to agenda setting and election outcomes. At a deeper level, this work shows how social media expresses a collective wisdom which, when properly tapped, can yield an extremely powerful and accurate indicator of future outcomes.
The simple fact is that there is still quite a chasm to cross before Twitter can be used practically for any type of prediction or forecasting activity.