Strategy of youtube’s recommendation system



The internet has much to do with associative memory by linking related pages. The formation of certain links can have important implication for the traffic. For a website like youtube, such connections are formed between different videos on the site via their video recommendation system.

This system will select a list of videos that it thinks you might want to watch based on your previous selections, but it is not necessarily in the best interest of youtube to recommend the most related videos. In fact, when youtube tried to optimize the click-through rate of their recommendations, they found that on average, viewers actually were not satisfied with the videos that they clicked on. Trying to maximize clicks, the algorithm would find videos that on the surface look good – maybe with an interesting picture, or a controversial title – but are not always substantial or genuine. Users would hop quickly from one video to the next once they realize that the recommendations are misleading, all the while increasing the amount of pages visited. Youtube has switched their strategy to maximize viewing time instead, but saw a substantial decrease in ad revenue because they could no longer rely on gimmicky videos to gain large amounts of views.

The recommendation system accounts for a huge number of views for youtube. Although the plurality of views are generated directly from a search on youtube, video recommendations rank a close second. According to a recent study, recommendations are actually even more important than they would seem at first glance because most of the direct searches pertain only to very popular videos, while the vast majority of the videos on youtube derive a huge amount of their viewership from so-called “related videos” system.

An important implication of this is that youtube can exert an incredibly amount of control over what its viewers watch, as evinced by the way that they could inflate their view numbers by recommending many misleading videos.

Even with their strategy to maximize the amount of time viewers watch videos rather than maximize views, it may not be in youtube’s best interest to recommend videos that are most relevant to what you have just watched.
Youtube wants to increase view diversity with their recommendation system. By giving you a recommendation that you probably would not have stumbled upon, youtube is increasing the time you spend watching their videos and increasing their ad revenue.
Source: unknown