Web searches predict success for films, games and songs
- 28 September 2010
- From the section Technology
Web search terms are a good predictor of the success of films, songs and video games even weeks ahead of release, researchers say.
The findings echo a study in April showing that the number of mentions of a film on Twitter could predict its opening box-office take.
However, in some cases predictions based on search do not significantly improve on those made with other data.
The research appears in Proceedings of the National Academy of Sciences.
It was carried out by a team at the research arm of search firm Yahoo.
They compared the total number of searches containing a given term, such as the title of an upcoming film, determining how much that number correlated with the film's success on release.
They did the same with songs on the US' Billboard 100 music chart and video games.
In all three cases, the correlations were strong - a much-searched term was a good predictor of the eventual success, even as much as six weeks ahead of time.
However, as Duncan Watts, a co-author on the study, told BBC News, "predictions are good or bad only relative to something else".
"There are other ways to predict the success of movies based on other kinds of data - how many screens it's opening on or what the budget is.
"So for each of our cases we construct a baseline model - which is what you would've (used to predict) if you didn't have search data, and then we throw the search query volume into the mix and see if our predictions get better."
For the most part, the search data did not much improve on those made based on different, more traditional data.
Study co-author Jake Hofman explained that the search data seemed to make the most difference in the case where the data available ahead of release was limited, such as video games.
"There are some places where you really don't have good outside information and search is reliable, like the non-sequel video games," he explained to BBC News.
"Then there are situations where search data (leads to) a pretty marginal improvement."
Bernardo Huberman, the HP researcher who showed substantially similar results using the volume of tweets about films, called the new effort is "good work", in no small part because "they have a lot of data - they're Yahoo".
He pointed out that prior work by Google researchers indicated a similar correlation exists between house prices and search terms.
"The picture that's starting to emerge is one in which millions of people are essentially browsing for data, and us being able to capture that browsing allows you to predict," he told BBC News.
Yahoo's Dr Watts remarked on possible future implications of the approach, saying that "you could imagine expanding this study to other kinds of consumer behaviour, thinking about hotel vacancy rates or how many people will be flying into Las Vegas this weekend - all of that is probably correlated with some kind of search activity".
But for Dr Huberman, that possibility raises an issue of some concern.
"In principle, these companies are sitting on their ability to predict consumer behaviour that could be used for their own benefit, or sold - the interesting question is whether this data should be restricted."
However, the crystal ball that people's technological habits might represent is not, he said, always the most important thing.
"In life, it's not so interesting sometimes to predict the future, but to know what to do once you know what the future will look like."