Facebook launched the Like button 18 months ago and it has had a huge impact on how people browse and share information and form associations with other entities. Within days websites had integrated Facebook social plugins which made it super easy to feed stuff back to Facebook and share with your friends in a frictionless way. Though Facebook started collecting information about every webpage you went to as long as there was any social plugin on that page, you still had to take an additional step to decide if something was worth sharing with your friends or else they would never see it. Let’s take an example:
I visit the NYT webpage and read a couple of stories, say A and B. I then decide that story A is worth sharing and hit the “Recommend” button and it gets posted to my feed. My friend arrives on NYT and sees the headline for story B. He doesn’t know that I checked it out as well but he is interested in it and even clicks on it but he never shares it either. Then a third friend is now on NYT trying to decide what she should read. Given the old scenario, only story A would be recommended to her. The information about story B and two friends interacting with it has been lost.
Maybe it’s lost of for a good reason – it probably wasn’t worth sharing. One could argue it keeps the signal to noise ratio high. But the best way to deal with information overload is generating more information, not less. With enough training data, and meta information like time spent and other derived engagement metrics it won’t be too hard to use that lost information to come up with even better suggestions.