Twitter Algorithm Can Predict News PopularityBy: Zach Walton - February 13, 2012
News sites already know how to modify their stories to gain maximum search potential, but what if you could do the same with Twitter?
Researchers at HP Labs were looking into the very matter with their new research, “The Pulse of News Media: Forecasting Popularity.” The three researchers, Bernardo Huberman, Sitaram Asur and Roja Bandari used the API of Feedzilla to collect a sample of 40,000 articles posted to Twitter last August.
The team analyzed the articles and then rated them against four factors – the news outlet that writes and first tweets the article, the information category the article fits into, the relative emotion of the article’s language, and the people and things named in the article.
They find that the most important factor of any retweet potential is the source of the news. The other important factor is the category in which the news covers. Some stories are more prone to retweets than others. Similarly, stories about celebrities or other well-known topics tend to do well.
The one factor that doesn’t have much effect, if any, on its retweet value is a story’s emotional edge. The research found that emotional and objective content both had about the same amount of retweets and shares.
The team took all of this data and put it into an algorithm that they claim can predict the number of tweets an article will receive. The model which works on “low-tweet,” “medium-tweet,” or “high-tweet” classification has an 84 percent accuracy rate.
Huberman speaking to The Atlantic said that this new algorithm could change how news content providers approach Twitter. The algorithm can be applied to a news story and it would tell the writer what to tweak to reach maximum Twitter spread.
This research is apparently part of a bigger effort at HP to find out “how attention is allocated with anything in the web.”
Considering how fast the news of Whitney Houston’s death started spreading only after the AP tweeted about it, I would wager that this research is pretty accurate.
Here’s the study in its entirety: