One of the biggest values of a real-time data product is its ability to highlight unusual activity on your site - activity that directly translates into actionable information. However, in order to identify what's "unusual," you must have a good sense as to what is "usual."
Over the course of our research, we’ve found that the number of active visits on an article, as a function of the time since it was posted, generally follows a few typical patterns. Armed with those patterns, we designed a newsbeat "Spike Alert" system to identify and alert our users when an article's traffic suddenly and unexpectedly deviates too far above its expected levels.
For every article, we evaluate the incoming traffic on a source by source (i.e. Social, Search, Links) basis to identify whether there is a spike in traffic from that source. (The number of active visits in the Spike Alert email is specifically from that source, not for the article as a whole.)
We determine that a spike is occurring by taking into account a number of factors, including:
- Predicted increase in traffic. This is to avoid sending alerts for expected, and therefore, uninteresting, increases. For example, an article posted at 2 a.m. that starts to pick up traffic at 9 a.m. is not that interesting.
- Rate of increase in traffic
- Absolute increase in traffic
Newsbeat Spike Alerts are configurable at a site, section, or author level and can be received by email, SMS, or iPhone push notification.
As an initial means of throttling the number of alerts we send, we created a default thresholding system based on the site's 30-day maximum number of active visits. Sites with large amounts of traffic will have a higher threshold than sites with smaller audiences. Going forward, we're exploring a number of different options to allow for more customized Spike Alerts.
It's incredibly important that we hear your feedback, so we can continuously improve our product. Please email us at email@example.com with questions or feedback on Spike Alerts, or any other newsbeat feature!