Predicting Social Trends Two Months In Advance

There are plenty of marketers out there whose sole occupation is tracking social media to detect new trends about to surface, and they’re not alone: from law enforcement and public health workers to economists and stock market speculators, there’s a wide range of disciplines that stand to benefit from predictive social media analysis. Towards that end a new study from researchers at Yale, the University of California-San Diego, the Universidad Autónoma of Madrid, and NICTA of Australia, formulated a technique that they claim can forecast social media trends up to two months in advance.

The study, described in a paper titled “Using Friends as Sensors to Detect Global-Scale Contagious Outbreaks” and published in the online journal PLoS One, focused on the role of “sensors,” meaning central individuals with more social connections who act as sentinels for wider networks, registering an emergent trend before it becomes prevalent across the network and the Internet at large. One advantage of the system is it allows researchers to predict trends based on a relatively small sample size using “local monitoring,” rather than trying to wrangle data from the entire social media universe.

The researchers looked at extended social networks, meaning randomly selected individuals plus their circle of friends and their friends’ friends, ultimately analyzing 40 million individual Twitter users and a total of 15 billion followers (when friends’ social circles are included). By randomly selecting individual users and a subset of their immediate and extended social circles, then analyzing the subject matter of tweets, the researchers identified certain “sensor-friends,” occupying central positions in social networks,” who became aware of emerging trends well before the less-connected individual users, who served as a control group.

For example the study, which focused on Twitter usage in 2009, predicted the rise of the hashtag #Obamacare two months before it became prevalent on Twitter -- an especially noteworthy accomplishment when you consider the term is a neologism that had just been coined at that time. However this was an extreme example: more typical lead times ranged from 20 hours to a week ahead of time.

According to the study authors, you only need data from around 50,000 Twitter users to predict what content will “go viral” across the Internet using “local monitoring” -- although they caution that they can’t forecast which channels (meaning specific social groups on Twitter) will end up propagating the viral content. 

“Social contagion” has been a popular subject of study in recent years. Previously PLoS One published a study suggesting that emotional moods, as well as ideas, can propagate socially across online networks. The study, titled “Detecting Emotional Contagion in Massive Social Networks,” analyzed around 1 billion Facebook posts by 1 million users using a system called Linguistic Inquiry and Word Count, then overlaid this sentiment with weather data showing when it was raining in a particular location. After showing the correlation between bad weather and negative emotional states in rainy cities, they then examined how Facebook posts by users experiencing bad weather affected Facebook posts by friends in places with good weather.

Also last year, a study by researchers at the U.S. Military Academy at West Point’s Network Science Center, titled “A Scalable Heuristic for Viral Marketing Under the Tipping Model,” examined 36 social networks (including nine academic collaboration networks, three e-mail networks, and 24 networks extracted from social media sites) to deduce an approach for identifying “seed groups” -- the subsets of larger groups who are most likely to propagate a viral marketing message.

Recommend (1)