Using computer algorithm, scientists can now predict the virality of your Facebook photos. The prediction is based on the shares of your Facebook photos.
Researchers from Stanford informed that they will take clues for prediction of Facebook photos virality to lie in ‘cascades’. ‘Cascades’ describe photos and videos, which are shared multiple times.
The research team includes Jure Leskovec – an assistant professor of computer science, Justin Cheng – a Stanford doctoral student, Lada Adamic and P Alex Dow from the Facebook team and Jon Kleinberg – a computer scientist from Cornell University. The team started the research by analyzing 150,000 Facebook photos, which have been shared at least five times.
Jure informs that the team wasn’t clear on the use of the information cascades for prediction since it rarely happens.
As per the data shared by Facebook and University scientists, only 1 in 20 photos shared on the social networking sites gets shared hardly once. In addition to this, 1 in 4,000 photos gets shared more than 500 times.
In one of the papers that will be presented at the International World Wide Web Conference, the researchers will be describing how they made accurate predictions, 8 out of 10 times when photos cascades would double in shares. For example, if a photo is shared 10 times, will it get 20? Moreover, if it got 500 shared, will it be shared 1,000 times, and so on?
The first analysis report states that in a cascade, there was a 50:50 chance of the shared to be doubled. The scientists have considered the variables that may aid them to predict doubling events accurately. This included the rate and speed of the photos shared, and the sharing structure of sharing (for example – photos reposted in different networks created stronger cascades). The best predictor of cascade growth was the speed of photo sharing. The next factor was how the photos were shared i.e. across different networks and groups
After considering various factors into their analysis, the researchers were able to predict the doubling events accurately almost 80 times which soon reached 88 percent.