Like for example: I watch a random youtube video & on that video it mentions something (an idea, item, person etc.) that isn’t related to the topic of the video, it caught my interest & before I complete the full sentence or word on youtube search bar the specific thing that I was interested it is suddenly on the top row of search recommendations, that’s more strange is that the video I viewed wasn’t even popular nor mainstream but instead it’s an obscure one & this is not even the first time it happened.
My hypothesis for this is that youtube has an AI that listened to the audio of every video uploaded & collects the words said on the videos which they use when someone watch a video that the AI recognize then recommends it on search bar, or maybe not so if I’m wrong about my theory can someone explain how youtube does this?
In: Technology
Youtube auto-generates closed captions for videos, so it would make the most sense that those words are used in some form of backend search or tag function associated with that video.
There’s also the possibility that other people have also searched for a similar item after watching the same video, so again there’s more associated searches already associated with the currently playing video and Google is remembering that when offering suggestions for searches to you.
YouTube can predict what you might search next because it uses advanced machine learning algorithms that analyze your viewing habits and interactions with videos. This technology learns from patterns in what you watch and how you engage with content, as well as how other users behave. YouTube combines this with natural language processing, which helps understand and predict search queries by analyzing the text related to videos, such as titles and descriptions.
Additionally, YouTube’s algorithms use collaborative filtering, which pools data from all users to identify trends and make predictions. This means if other people who watched the same videos as you often search for something specific afterward, YouTube might suggest that search to you, too.
In essence, YouTube’s ability to anticipate your search queries is based on analyzing large amounts of data to identify patterns, rather than listening to the content of the videos themselves. It’s about understanding behavior on a broad scale and using that information to make educated guesses about what you might be interested in next.
Youtube has AI that can check the audio for stuff discussed in it. It looks at the comments of people who responded to that video. It looks at common searches *other* people made immediately after watching that video. It keeps track of your habits, searches, and interests. And it has a very basic AI for filing in generic search stuff (same as Google search does).
All of that feeds together into predicting what stuff you, personally, might be interested in seeing.
It’s a proprietary algorithm so you can’t know EXACTLY how it works, but it includes a combination of factors like metadata, previous behavior, and other user behavior.
Google is going to pull info from videos via uploader data, comment data, and transcription data. Let’s say this pile of metadata for a single video says: “Ryan Gosling, Dancing, Cute, Hip Hop, Funny.”
You Google “funny hip hop dance”, and boom there’s Ryan Gosling.
If you watched a clip of Ryan Gosling, and a clip of Jimmy Fallon, Youtube is going to recommend a clip of Jimmy Fallon interviewing Ryan Gosling. If you watch that, then Youtube is going to recommend Jimmy Fallon being interviewed, or Ryan Gosling being interview on a different talk show.
It works in the same way that Amazon says “other customers ALSO bought THIS!”
If someone else watched everything you watched and then randomly watched a monster truck rally, Youtube is going to recommend the monster truck rally video because it made someone else with similar tastes click before, so they take a shot with you.
If you don’t click, then Yotube downgrades the keyword “monster trucks” for your profile’s interests. If you keep clicking on monster trucks, Youtube makes that a bigger focus and starts to ignore clips of talk shows.
I wind up watching a lot of weird 10 second videos that have 1 million or more views, so Youtube will keep recommending those almost irrespective of the content.
The whole idea is to keep you clicking and watching ads so they are going to make educated guesses and follow patterns that have have gotten you or other users to click.
Edit: as for your questions about predictive text, it works mostly the same way. But for the most part it’s filling in what other users have searched for. Google processes around 2 trillion searches per day, so you can’t start Googling anything that hasn’t already been Googled a billion times prior today.
Latest Answers