Recently, one of our clients (a large video service) came to us with the task to increase the clickability and viewability of videos on their service.
It was obvious that this was a job for our data department.
In a short time, they developed a system of video recommendations. This system is an algorithm based on impressions and CTR. At the heart of this algorithm is the mathematical apparatus — the lower boundary of Wilson’s confidence interval (Wilson) for the Bernoulli parameter, it serves to calculate the rating for each video. According to this rating, each user is offered a selection of clips according to his interests.
As a result, our system has increased the clickability by 19% and the viewability by 24%.
By the way, this rating allows you to work with various content (in this case it was video).
Also, in the future, we plan to use Machine learning in this scheme, which will allow us to improve these numbers.