How does Google force videos on me? Click below to click YouTube

When browsing YouTube, everyone thought, he recommends videos for you. Get involved with them in a working algorithm developed by Google’s programming elite. Let’s take a look at the problem of manipulation with its former development, the level of trend will be explained by youtuber Tary and finally we will reveal how it is recommended to influence it.

Mon, that’s how it happened. Very boring and don’t realize it, click on YouTub with the right panel of the proposal. Often in the comments they lead vtyIm on the weird part of YouTube again (in hell: I’m on the weird page of YouTub again). How do we get to it?

Rolls-Royce between programs

It will introduce a smart algorithm there. The development within the company is called the most sophisticated referral system of the present. And why are they so proud of him? The system recommends the use of many sites and services, but YouTube has to deal with the following factors:

  • Volume: You’ll find billions of active videos on YouTube. So the algorithm has to work with a huge amount of data. You can cheat on astronomical slahere.
  • Accuracy: Upload 400 hours of video every minute. Even you have to process the software and evaluate their relevance for each individual hunter.
  • a: Designs are easy in the electronics store. When someone browses laptops, recommend them under the product of another laptop or accessories. But user breeding on YouTube is extremely difficult to predict. Sometimes you want to play energetic music, other times you can laugh at youtuber, in the evening you can start a tutorial on a tie and watch soundtracks at night. around the algorithm you can find some formulas in your breeding and then recommend them on the basis of them.

The algorithm is currently being developed by Google Brain – Alphabet’s best brains dealing with artificial intelligence (AI). Of course, you have published the algorithm on YouTube, but it has published its operation (PDF, in English). If you don’t want to go through complex English, let’s try to simplify it.

Explained simplyFilter dog two hundred

In order for the AI ​​to experience a large number of videos and select a dozen measures per measure, use only two hundred. The first is the candidate generation. This is due to the history and context (demographics, location, words in the search) of the selection from hundreds of millions to hundreds.

This fraction goes through the second hundred, called ranking, which added real-time testing to the cadmium of the video – and only the highest ones will be sent to your main site or the right panel of the proposal.


Until 2012, cars rewarded for the number of views – that is, clicking on the video. This created clickbait videos for which the user did not succeed. The alphabet therefore created a new middle ground rule in 2012. The most important thing is the long time of watching the video and the user who is watching it regularly, the rest is supported.

Therefore, it is likely that the videos you saw will appear in it. If you spend a long time with them and click on them according to your mood here and there again, YouTube has no reason not to recommend it again until you get tired. A typical example is long mixes of music or hours of noting a bang, which people bind as a podcast and listen to the cell.

Explanation of the compositionNeurons study subconsciously with him

And let’s get deeper and deeper. The predecessor of the current software was the so-called matrix factorization. The complex part of the phrase knows the method recommended for the basis of previous activities, in which it looks for hidden formulas. You then refine the preferences of the user and better filter the results.

According to Google’s development, this method is a remnant of sloit. They therefore developed a neural s, which proceeds from factorization and simplifies it. The algorithm is based primarily on our history, along with changes in hidden features, which, according to developments, go into more depth than female feedback (lb / don’t like me, questionnaires). We can imagine that the method does not draw from the consciousness of our subconscious.

Visually plays a role


Schma zen vbru

For the software, the history of tracking is as important as the viewing. Anything you type in the search engine is indexed and reflected in subsequent proposals. No matter if you clicked on the found videos. Demographic data are added to the data summary – age, gender and status, available in Google. Vvoji realize that the user prefers new videos, and therefore the date of publication will be among the key factors.

Only a hundred videos will go through the first hundred. Thanks to the zenmu sample, a factor much more can be involved in the second hundred. Some of them are binrn, so they have only two mon results (eg if the user is stung). Most have millions of different variants (eg the last search term) and there are hundreds of categories. The software has to process it quickly, due to its development that the neural light and all the factors fail all the time and it works with what it currently does.

In addition, not only the attributes of the user but also the video itself are involved in the game. Is there a lot more in Brazil? Did the people from this video click on it? Similarly, the software finds out about channels. Do you watch his videos regularly? Don’t go to the end for them? For have you grown a kanl to remove?

From the obtained data, the algorithm then generates a nominal score for the video, which is based on the assumption of two things – whether you click on it and how long you spend on it. tve vs njak video that is constantly displayed in me? It’s dark, high. You can dream it by ignoring it until it disappears, or by clicking on it and running away for a second. The song is not software.

TrendyPositive discrimination in practice

On the main page you will also meet the so-called trends. This is a panel of videos that are currently popular. But the popularity is not high. Some trends are predictable, like new dogs from a famous artist, others are quite surprising. That is why we dream of a combination of popularity and news, development.

The order trend (# 1, # 2…) is determined mainly by the rate of growth of the crop, the place of origin and other factors. Therefore, the video with the most watches may not be necessary in the first place. The fact that sweat does not play the main role is also confirmed by the well-known esk youtuber Tary.

A year ago, the trend got dark, which had an abnormal number of likes and comments. And thanks to that, I gained 400,000 customers a year. It worked like I said: It can hold 50,000 likes. The user managed it, Youtube recognized that it was trendy and threw it there. And that was true of the comment, spn youtuber describes.

Since December last year, the vbru system has changed. There are also new videos with trends and likes. It’s thanks to the category they fall into. In particular, YouTube has defined a certain number of flowers from individual categories, to Tara. For example, the vlogue of riding a motorcycle may be more of a musical novelty, but thanks to the car-moto category, it will enter the trend.

ManipulationDoes uml influence bh society?

Recently, the impact of fake news on the election results. The question is whether ours does not affect anything from our own ad. One of the people who saw under the guise – the development of the design algorithm – spoke for The Guardian about the YouTube threat to the small population.

YouTube is TV like reality, but it is distorted that we take more time on it, to French programmer Guillaume Chaslot. Google released him in 2013. Official reason: performance issues. Chaslot’s version: he tried to make a change so that the algorithm was vc fr.

After the crash, he began to pay deep attention to the machine unit he was involved in. And during the US presidential election, he noticed an anomaly. It was special. Regardless of whether you started looking for Trump or Clinton, the video throws were pushed into the Protrumpov area, to Chaslot.

The author of the link in The Guardian, Paul Lewis, decided to test his theory. He went through a thousand of the most recent videos associated with the election and reached the eye: 551: 92 (360 neutrals). YouTube was more likely to recommend videos that helped Trump and the Clintons. The algorithm was certainly not preset, but thanks to the user, it started to behave in this way. Simply put: people liked Trump videos more, so YouTube started recommending them.

Chaslot launched his own website, where he explains how YouTube says he promotes conspiracy theories or favors some election candidates.

From the user’s point of viewHow can I influence the recommended?

So how can you get involved in the offered videos? As a user, you have the opportunity. One of them is Brick: When you start watching Brick videos, you will get more personalized designs, explains Albta Houzarov from Google for Technet.

If you are not satisfied with the recommendations, you can remove videos, channels and sections that are not of interest to you and improve the recommendations. You can do this with the help of Nezajm m.


If you delete two or more videos from one channel, this channel will not be recommended for you, Houzarov continues. It is only on potai at the moment.

It is possible that you have removed the channels that you did not care about, and now you have. Correct this on the page, where you click on the open menu and then on Feedback Uninteresting. It is recommended to restore it by clicking on Delete feedback. And it is recommended that you also influence the search history.

Let’s say you’re looking for some darkness, like changing a wheel. But it’s your video with wheels in your recommendation. Just go into the history of watching and searching and delete these specific videos and killers, adds Houzarov with a warning that deleting can not be reversed.

Pt will take a look at other YouTub’s weapons software – the Content ID program, which watches over the author’s first and blocks videos that copy content according to it.