The streaming movie giant Netflix have a knack of being able to suggest what movies to watch, stats suggest that an amazing three quarters of all customers select their movie entertainment this way.
Ever wondered just how Netflix manage this? Well it’s all down to the way their impressive algorithm system works, carefully analyzing massive amounts of data to try to ensure that the recommendations that it gives match as close as possible the individual viewers requirements.
Just what goes into the Netflix recommendation algorithm?
Well lots and lots of data, but you would expect to be told that! More specifically Netflix looks at the information collected from millions of its customers to decide just what to offer:
- Approximately 30 million plays each day
- What movies are being watched
- How much of the movie is watched
- Whether the customer continues to watch sequels or similar movies
- Geo-location data
- Device information
- Social media data from the likes of Facebook and Twitter
However, even when Netflix has all of this information it can still be really difficult to figure out exactly what its users will want to watch; one of the key factors is in knowing what movies normally follow other movies.
Netflix’s current algorithm, BellKor’s Pragmatic Chaos was built and tested over a 3 year period by a number of testers from AT&T, Consutig and Commendo Research, as part of the 1 Million dollar competition that was set up by Netflix. The algorithm proved to be ten per cent better at collecting and understanding the data; and continues to collect information from viewers to this day.
The easy part is collecting the data
When a company has such a large data base to go at, the collection element of the data is a relatively simple task; the much harder part is trying to understand just what it all means. It would appear that connections play a very large part in how the algorithm works; an example of this would be that if a number of people watched a particular movie and then a large proportion of those people go on to watch another movie then that itself would form a strong connection. It is these connections that form a large part on how Netflix offer its movie recommendations.
Netflix admit that despite how effective their current algorithm is, there is room for improvement; they hope at some stage to show customers content that they will view to completion and then recommend the next thing that they will view go on to view to completion.
From the noises come out of Netflix towers, this doesn’t seem too far off from happening.