In a recent whitepaper, Ooyala broke down how consumers connect with video and how you can use recommendation engines to help drive that discovery. I thought I would break into the paper a bit just to look a those ways since it's interesting.
Having great content is one thing, having consumers find it is entirely a different beast altogether. Even just deciding on how you're going to recommend content requires some heavy mental lifting. You could simply throw all that awesome content you made into a library and hope for the best, or you could choose one of these ways, which will probably work far better.
The Algorithm Method
For those that aren't quite sure what the algorithm method is, it's when an algorithm is used to recommend content to consumers. Most recommendation systems are going to be using an algorithm of some sort, unless you've got a team of humans curating content and recommending it to consumers.
Our curiosity about a topic is often not satisfied after seeing the first piece of content about it. This means recommendation systems can rely in part on the relationships between the content in videos to provide appealing suggestions about which other videos you might also want to see.
What kinds of relationships are there between content?
So we can attack it all from a strictly data or metadata approach. Genre, subject matter, director, product type, etc. This also gets into things like language, serialization, live and on-demand content. That's one way to do it and is one of the most popular ways which we see in many places. You watched an action movie, here are more action movies. You watched a Kurosawa film, here are more Kurosawa films...
This version of the recommendation system is based on the behavior of groups of people. You probably see it every week when you go to Amazon and the page says "other people who purchased this also purchased X, Y and Z." A group of people did something that you are now in the midst of doing, perhaps you're interested in following in their footsteps?
Things that go viral do so for a reason, because people found the content interesting and wanted to share it with others. The important thing is that when a topic is hot, like the Red Bull Stratos jump this weekend, it's because it's interesting and timely.
...popularity and trending are different in important ways: popularity means that a large number of people have watched over the long term; trending identifies unusual short-lived spikes in watching patterns...
This will, of course, require a real-time analytics package or tracking that you can then pull data from to start recommending whatever is hot at that very moment. When recommending based on trending it's far less predictable as you can't always predict what will be popular on any given day.
Which System is Best for you?
There's also the potential to use multiple types of recommendation engines in various places. Find which have the best results and use them the most. After all, it's all about getting the content in front of the consumers, right?
Presenting recommendations to viewers in the most appealing way is critical. For some applications, distributing the content throughout the website is best: in these cases the recommendation system will need to be designed to be queried by the publisher’s web
application to return recommendations for integration into the overall page composition.
Where a more “lean-back” TV-style experience is desired, it may be better to show a few key recommendations directly in the player so viewers can enjoy immersive full-screen playback with only occasional mouse or keyboard interactions to course-correct.
Connecting Consumers with Content does a great job of explaining what recommendation systems are, and more importantly, how they can help an online video publisher not only have the content discovered and viewed but also generate more revenue. So get moving!