The recommendation system became increasingly crucial to YouTube's frenetic push for growth. Then the model would predict which videos you'd be most likely to actually watch, and presto: recommendations, more personalized than ever. The model would take your actions (whether you'd finished a video, say, or hit Like) and blend that with other information it had gleaned (your search history, geographic region, gender, and age, for example a user's “watch history” became increasingly significant too). By 2015, they would also introduce neural-net models to craft recommendations. Instead, they focused on “watch time,” or how long viewers stayed with a video it seemed to them a far better metric of genuine interest. Goodrow and his team decided to stop ranking videos based on clicks. Even if a viewer immediately bailed, the click would goose the view count higher, boosting the video's recommendations. Goodrow noticed another problem caused by YouTube's focus on views, which was that it encouraged creators to use misleading tactics-like racy thumbnails-to dupe people into clicking. In 2011, Google tapped Cristos Goodrow, who was then director of engineering, to oversee YouTube's search engine and recommendation system.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |