A look into the Non Technical Aspect of the YOUTUBE ALGORITHM. | by Zaid Shaikh | Jan, 2021

0
36

Imagine one fine day you’re surfing through your Home Page feed on YouTube, and somehow amidst of a busy frustrated schedule you’d come accross a newly recommended 🎹 play video, you play it and feel on how great 🎶 can be, but but but there’s a catch, it’s around 5+ years old and more suprisingly you’d open the comments section to meet all your Homies😎, and all the comments you’d see are about

YouTube in 2015 :- no
YouTube in 2016 :- no
YouTube in 2017 :- no
YouTube in 2018 :- no
YouTube in 2019 :- no

YouTube in 2020 :- Well let’s Recommend this video to everyone

And once again our thoughts Square back to the same topic, the mysterious YOUTUBE ALGORITHM.

But before we get to think if there’s really a one way solution to the entire Algorithm, let’s get this one thing clear. The Algorithm and its core Engineering aspect keeps getting changed keeping the Content Creators doubtful about their strategies. Although the Machine Learning/Deep Learning/AI/Analytics keeps getting changed, the FUNDAMENTALS will always remain the same.

Ever thought of what keeps you hooked up for hours after promising your mom for just 5 more minutes on YouTube, it’s probably the Recommendation System that keeps you tied up binge watching Videos.

According to a 2016 published paper by YouTube on its Deep Learning based Recommendation System, it uses 2 Neural Networks, one being a Candidate Generation sorting out a Viewers prefrence and other based on Ranking a Video’s performance.

Candidate Generation based Neural Network Model works on collecting all the past history of a user, followed by creating a subset of videos that a user is likely to watch.

This Neural Network Model essentially builds a replica of what Interests a user seeks, sorting out the best topic of videos.

Eg :- A gamer will be bombarded with Gaming live streams,videos, in his feed, whereas a foodie will counter meals/recipes,hotel & restaurant vlogs on his profile.

Ranking Neural Network Model ranks any given video with various factors such as

  • Impressions :- Number of times the Video surfaced on users discovery(Home page). The video must be played for 1 sec and more than 50% size of the thumbnail should be visible on the screen.
  • Click Through Rate(CTR) :- Ratio of Impressions to the clicks that a video receives. Eg :- For a video with 100k impressions and 10k views for a video, the CTR would be 10%.
  • Total watch time(in minutes).
  • Likes,dislikes(not going to leave you alone as a Youtuber😈) and Comments.

These very factors decide the fate of a Video as well as of the Youtuber whether the Algorithm pushes the video down to people’s recommendation or dump it aside.

This Model also solves the very problem that most viewers find annoying, the Click-bait videos. If a video has high Click through Rate but low watch-time, it’s obvious that the video is Click-bait and the Algorithm pushes it out of the recommendation list.

Apart from these Neural Networks, the platform has also come up with Solutions such as Video Cards, suggested videos at the sidebar, playlists, hashtags etc for better reach of content.

Just when a person would hear about Feedback, they’d possibly think of a User’s feedback solely, but the Feedback Mechanism is more than that.

A powerful tool in the name of YouTube Analytics shows various performance charts such as

  • Keywords
    Keywords are essentially what a user is likely to Type while searching for a Video which equates to reading your customer’s mind on what content he/she wants to consume. This is usually the best SEO strategy to be ahead of the competition.
  • Average watch Time
    How long users watch a creators video(which helps in tailoring the length of videos according to subscribers preference) so that users don’t get left out of short content or either don’t get bored watching long videos.
  • External links to other Social Media platforms which drives additional Traffic.

These analytics along with other options such as Frequently done surveys, Video rating scale for recommending to other users filter out a Video’s performance.

This strategy helps in building up an assigned weight to each video and likely recommend it to viewers on their Home page. Viewers who spend more time on YouTube are recommended videos with more weight but lesser views and likely Users visiting the play icon app once in a while are suggested the best performing videos which keeps any casual user on the platform for a longer duration of time.

Nothing beats Quality of one’s content over Quantity. At the end it’s the depth knowledge, usefulness and how a particular problem a Video solves keeps the viewers hooked on YouTube. This is the reason why channels like Vsauce, MKBHD, AsapSCIENCE would gain large amounts of subscribers and views solely on their Content Quality.

Img src :- YouTube

At the end, every viewer’s time is valued by YouTube and it presents the best possible Content to its user.

Original Source

Advertisement
Dating Website Marketing

This site uses Akismet to reduce spam. Learn how your comment data is processed.