With brands currently willing to purchase “trending topics” on Twitter at a premium, how much more valuable would it be to know what’s trending even before Twitter does?
MIT professor Professor Devavrat Shah and student Stanislav Nikolov have developed an algorithm that can do exactly that, and they claim it has 95 percent accuracy and can make predictions up to five hours in advance. The program doesn’t use pattern recognition because, according to Shah, there’s no set pattern to how things trend on Twitter.
“You [could] try to train for when the jump happens, and how much of a jump happens,” says Shah. “The problem with this is, I don't know that things that trend have a step function. There are a thousand things that could happen."
Instead, their algorithm uses sample sets against real-time activity. Once a topic begins to resemble one of the sample sets, it’s given a “vote” on whether it could trend. More votes points to more of a chance it will trend.
Shah and Nikolov’s ability to see into the future seems sci-fi, but the bottom line for their creation is what it could mean for Twitter’s advertising prospects. With the current price for a trending promo spot falling around $150K (with a bidding element), Twitter could be looking at this new algorithm as a major commercial selling point. There is also talk of using the formula in situations to predict stock prices, ticket sales, and more.
The MIT duo will present the concept at the Interdisciplinary Workshop on Information and Decision in Social Networks at MIT in November 8 and 9.
Learn more at Wired.