What are Facebook and Twitter Predicted Interactions?

NewsWhip tracks the world’s media - capturing new articles within minutes of publication. It then tracks the growth of social interactions with the story and weights the increments by the rate of change (acceleration / deceleration). Using predictive analytics, these measurements are extrapolated to estimate the number of interactions a story will have earned by the time it is double its current age.

This enables precision crisis management and an immediate understanding of the stories that will have the greatest impact on the public.

How it works

We measure the social interactions on each article at regular intervals to create a time-series of historic values. This allows us to get a speedometer rating of how quickly the article is accelerating or decelerating.

We apply a predictive analytics technique known as double-exponential smoothing to this time-series to generate a prediction for each article. The idea is to give more importance to recent values in the time-series. Thus, as observations get older (in time), their importance becomes exponentially smaller.

We predict the score each article will have when it is double its current age. This means that the longer we observe a story, the further into the future we can predict with accuracy. The maximum distance we’ll predict into the future is 24 hours, as most articles taper off after that time and so we measure less frequently.

We form a new prediction every time we collect a new datapoint for a story. This ensures that we make use of the best available information at the time and that if an article’s trajectory unexpectedly changes course, we are aware and can keep our clients informed.

In Spike, our predictions are shown as ranges of possible values together with a line showing a central estimate or most likely value for the future outcomes. There is an 80% probability that the article will not exceed this range.

Past forecast errors are used as the basis for determining the likely range of possible outcomes. The greater the historical errors, the greater the extent of uncertainty surrounding any point estimate.