Passing Thought

November 8, 2015

Youtube recommendations are generally great, but tend to weigh my most recent (read: last two weeks) interests disproportionately higher than anything I have seen in the past. Binge-watch anything in particular and the algorithm rapidly assimilates its recommendations around that interest to “feed the fire”. 

Because I saw a couple of Schumacher F1 videos, and checked out some SUV reviews - I now have a feed littered with other recommendations of this ilk even though my interest graph over a wider temporal bound is far more diverse.

Youtube’s data likely indicates that it is monetarily more significant to promote the binge habit (symmetric recommendations) over diversity (asymmetric recommendations). I posit that there is a growing corpus of Youtubers who change their viewing behavior and view fewer videos of a given kind in order to help this naive algorithm keep diversity up. 

Clearly, interests can be fleeting. Perhaps the algorithm can make topical approximations of “stickiness” to counter this. If a person consistently engages with a topic over time, it should “fade” more slowly off the recommendations grid. Obviously, the behavior of other folks with interests similar to mine would also be a signal input.

tl;dr

the Youtube recommendation algorithm’s agility is impressive, but I wish it wouldn’t forget so easily that humans are capable of fleeting obsession and enduring curiosity.