Tuesday, October 6, 2009

Box Office Tweets

Alex recently launched a very interesting project: trying to forecast a film's success or failure by tracking the number of tweets each received at Twitter prior to release. The idea is similar to David and I were attempting to achieve with Yahoo user reviews. We had reasonable success there, but developing a model based on Yahoo has several drawbacks, the biggest one being that it is highly dependent on how Yahoo designs its interface, and when Yahoo was making a lot of modifications itself, the stability of the model suffered. Twitter, on the other hand, should be much more stable in that regard, and even when its interface does evolve over time, the ease and habit of tweeting should not be affected.

The project is still in the early data gathering stage, but certain traits are already visible after I defined several variables and ran statistic analysis on various combination of them with Alex's data (note: I only applied linear regression for now unless there is significant evidence for quadratic or higher to avoid fitting to noise). The clearest one is that demographics definitely has a strong correlation with the final ratio. If we define the main audience a film attracts into five categories - kids, older females, older males, younger males, and younger females, then the ratio goes up as we move from the first to the last category. Regardless what other variables I pick, the resulting model shows a t-value of over 3.5 for its coefficient, meaning a statistical significance of 99%.

The second trend is that the ratio is inversely proportional to the theater count, or in another word, when a film has lower theater count, it becomes more likely to have a bigger ratio. This is similar to what we saw with Yahoo as well, and we could reason why such finding is plausible. When a film is shown in smaller amount of theaters, it becomes less accessible and will presumably require more passion on the viewers' part to seek it out, and such group would be more intent to tweet about the film.

I will continue to follow Alex's site for updates and his thoughts, and I would encourage others to do so as well. Hopefully I will have more findings to share later.

1 comment:

  1. Great insight! Thank you for your ongoing support and help in slicing the data.

    ReplyDelete