2018 Trending YouTube Video Analysis, What makes a video go trending?
 

I decided to analyze the top trending videos in 2018 for USA the goal of this analysis is to try to identify know common elements have the videos that go trending in youtube and test if we can extrapolate methods from mainstream to the adult world to be able to constantly have better results.

Ok, so let's start with a little bit of descriptive statistics shall we? What are the videos with the most views in this trending group?


These are the top 20%



We can see that the video on this trending group is Childish Gambino “This is America” video with more than 3 billion views and so far at least the next 4 videos have more than a billion views.


The next element we are going to review that was available is the “Dislikes” of a video, so Which are the videos with the biggest number of dislikes in this 2018 trend analysis?



The video with the biggest number of dislikes is the “youtube rewind: the shapes of 2017” with more than 10 millions dislike, after that the video called “So Sorry” and the Childish Gambino video is in third place, is is possible that there is a relationship between the popularity of the video and the number of dislikes it gets?... we will see.


The next element to analyse are the videos with the biggest numbers of likes, here are the results:



The top video here again is the one from Childish Gambino, the second one is “Fake Love” and the third one is Adriana Grande with the video “No tears left to cry”, we can see the Childish Gambino video again, but it is not clear if the “Likes” make the video popular or because the video is popular the viewers click the “Like” button.


Now we will analyse the videos with the most comments, here are the results:



The top 3 videos with the most comments are “Fake Love”, Childish Gambino “This is America” and “So Sorry” , we can see certain pattern here right? We can see that the most popular videos have a lot of interactions nevertheless it is not clear which one is the cause and which one is the effect, so let's keep analysing the data.


We have analysed individual videos so far, now we will show you the relationship between video channels and views:



One can see that once again the channel of Childish Gambino is the one with the most views, Marvel Entertainment is the 4 and Maluma (a spanish singer) it is the 6th once, very good combination of rap music, comics and spanish entertainment, as a side note we can see Taylor Swift in a very distant 15 place, even she is one of the best selling singers.


Once we have analysed the videos, channels, likes, etc.. let's analyse the correlation among variables to be able to identify what variables have a direct relationship and which ones do not have it.


What I am trying to evaluate with this correlation comparisons is to see which variables have a higher correlation and impact in the fact that the videos were trending, the idea is to compare the different combinations of these variables:



The correlation will be a number between 0 and 1 the bigger the number the more correlations exists between the variables.

I will compare first the relationship that exists between Views and Likes for the trending videos.

Likes vs Views.


See the above graphic, for reference I have always marked the Childish Gambino video, the numbers that are shown is the number of likes, the line in the graph is the line that shows the correlation, in this case there is a positive correlation (.80) meaning that one increases the other increases too, lets keep analysing the rest of the variables….

Views vs Dislikes.


These two variables have little correlation between them (.3) meaning not because you have more views the number of dislikes will growth or vice versa.

Views vs Comments.


In this case Views and Comments seem to have little correlation (.47)

Likes vs Dislikes.


I was expecting a lot more correlation between these two, but it is really small (.28)

Comments vs Likes.


This is one of the biggest correlations between two variables in this analysis .70 .


Little correlation between these two variables too, comments, vs dislikes (.49)

Analysing the correlation among variables we get the following table:


This table indicates the likes and views have the biggest correlation and after that the likes and comments, can we conclude those correlations make a video go trending? No, no yet we would need to make an anova analysis BUT even with this information we can get certain conclusions.

CONCLUSIONS AND PRACTICAL APPLICATIONS:

With all the information we have analysed we can see that the videos that were trending in 2018 have several elements in common, the number of dislikes seems to be totally independent to the rest (low correlation), the most important correlation factor are that videos with high number of views have high number of likes, this is important for those that are using bots to make their videos more natural, the second important element is that videos with high number of likes have a high number of comments, you need to take this into account to sell the right signals to the youtube algorithm otherwise your video is clearly being manipulated and youtube will know it.

None of these elements fully explain why a video goes trending BUT it help us to realise how difficult is to get conclusions from tests and analysis, and to separate facts from beliefs based on the data, there are more questions than answers example: what if we analyse the videos based on the category they belong to? Are all the videos the same across different categories? Is there a lot of variation, I will add this information if there is interest, as well as to make an anova analysis to identify from all the variables we have is the one that have the most weight in the trendiness of the video.


Adult

I may be pushing the envelope too far by extrapolating these conclusions to the adult tubes since the algorithm is different but it definitely may be worth to test the results if views, likes and comments can be made with a bot if this can make the videos be more trendy and get more visits.


If there is interest to know exactly what element is the most important to made a video go viral let me know and I will do it and share it with the community.

Good post OP but there is one other variable that we don't have access to that plays a pretty big part as well.

With "views" there is also a hidden stats for "view time". View time weighs very heavy on how much the "view" counts toward making it trending. The reason for this, is to stop someone from trying to make a vid trend by simply spamming a video link to get views.

The way they filter that is by using the view time stat to know if people are actually watching the video or just clicking to it and closing it because it wasn't good. Meaning "view time" the hidden stat is really more important than just views alone.

Likes have already been confirmed by many youtubers as a key element, but thew view time is the trick that would be hard to get around in order to make a "tend" video. It simply requires good content..


edit.. There is also talk about length of the video mattering as well as some people claim they should be over 10 mins, but IMO it's obvious with music videos being less than that that the 10 min idea is not accurate.

Bookmarked. Thank you for this analysis.

Interesting post.

Definitely shows correlation, but hard to tell which data point causes the other to move or if it is mostly exponential.

I would say dislikes to likes ratio is an effect instead of a cause.

Media picks up the video and shares it for some reason. Look at the new Gillette ad. People go see the "toxic masculinity" ad everyday just so they can comment about how much it sucks. Emotion feeds traffic.

I wanted really bad to have a white woman in a MAGA hat making out with a black woman and put that on youtube. It would have got a ton of views. But I did not want either of the women to get attacked on street due to the video.

^that would be awesome. not the attacked part tho, that ain't going to happen bro stop being paranoid.

Do it...

i like poo

Still busy studying the graphs

https://media.tenor.com/images/91782...4b4f/tenor.gif

That was insightful, patadeperro :thumbsup.

Good call about the view time, but as you said that information was not available so I have to deal with what I could get my hands on, but yes you are right there are more variables involved for sure.

Np, happy to get any feedback you may have about it, cheers.

What? what do you mean? the correlations between variables was clearly established, and there is no exponential anything, this was a linear correlation.

Good call, in reality I was just showing correlation between variables, not a cause and effect relationship, but yes you are right some of the actions may be due to other factors.


Don't worry, an entire lifetime won't be enough for you to even read the post, you just keep trolling.

Thanks CaptainHowdy, cheers.

this is prolly the most informative post we've had since 2016 when this forum turned into policowars

also did you consider the thumb image? I don't know how you would measure or get empirical data on that but having an image that gets clicked more comparing to the same video content with a different image, thus leading into the views, likes, comments etc

Thanks for your comments, two things regarding the images used, that information was not available to be analysed and second even if it were available here we start getting into a very difficult area that is the analyse of images due the volume of information would have been extremely difficult to know what elements from that image helped to make the videos trend or not identifying specific elements of an image from a computer perspective is extremely difficult but I do understand the relevance of that analysis.

I understand you used a linear model and you conclusively proved correlation. Wasn't that your point, that they correlate?

I'm saying that we don't know if engagement linearly increases views or views linearly increase engagement or if each increases the other for an exponential effect and, if so, which influences the other more. Does that make sense?

Now I understand your point but I think you are confusing the things I was not trying to prove there was a correlation it was more to know if correlation existed IF it existed define how much correlated each variable are, now the graphs are showing that some variables have bigger correlation than other BUT does not show a cause effect relationship, as they said correlation does not mean causation, meaning yes there is a correlation between variables BUT this does not mean there is causation specifically there is a high correlation between likes and views BUT we can't say that higher number of likes cause higher number of views not that higher number of views cause higher number of likes, we have only established there is a positive correlation between these two variables.... and here comes the real issue with statistics it is not easy to REALLY establish cause and effect relationships.

Bumpin' . . .

Firstly, I am sorry that you have a sense of humor of a slug.

Secondly, I have now actually "chomped" thru your statistics, and ahm...you have NOT concluded anything about trending, but just correlated high # visitors and high # of likes and dislikes and comments. It makes a total sense, even without reading that drivel your call statistics. Everybody knows if something has a billion views, that it will have caomparable high number of likes, dislikes and/ or comments. Just look at Trump's twitter. That is logic, not statistics. Your statistics ( even though you have evaluated billions of data bits) is akin to this statistical marvel:

" We have determined that 50 % of viewers liked the clip and 50 % viewers did not" ...."Hmmm and what about the third viewer ?"

As to the graphs themselves:

You have neglected the fact that even if the power of sample is adequate, it is actually quite massive, the probability of correctly rejecting the null hypothesis when it is false (inverse of the type II error rate). Experiments with low power have a higher probability of incorrectly accepting the null hypothesis—that is, committing a type II error and concluding that there is no effect when there actually is (I.e. there is real covariation between the cause and effect).

There are many sources of potential error in massive data analysis, many of which are due to the interest in “long tails” that often accompany the collection of massive data. Events in the “long tail” may be vanishingly rare even in a massive data set. For example, in consumer-facing information technology, there may be little data available for many individuals even in very large data sets.

Your graphs show a great regression towards the mean. From this explanation it is also clear that the more extreme sample you select for your conclusions there is a higher likelihood of a regression toward the mean in any subsequent measure or any post test.

I could go on and debunk all your stats but I am bored reading that the Earth is a sphere.


Also before you call somebody a troll look at your own " buy from me useless software" track record.

That is the goal, to separate beliefs from facts and sometimes statistics help us to confirm our beliefs, such is the case

what third viewer?

You are wrong, there was no hypothesis set, I was just analysing the variables involved in order to create an hypothesis we would need to make an anova anaysis or a t-test analysis that I invite you to make or @adultking that says he is an expert in AI and big data, you are just repeating words without understanding really what you are saying.... there was not hypothesis created YET, that would be the next step on an analysis like this


Please debunk them, that is the goal of this post, happy to share the data collected and cleaned, I am always happy to learn

No I wont your comments are always this kind of nonsense and I am here proving I just not promote my software..... unless you that even when you tried to debunk anything and repeat words you dont understand .... you failed, but happy to learn your elements to debunk the information I put here, that is the goal of this post, cheers.

Most everyone plays in the same way with a few fundamentals....
How many are drawn to it and how long they watch it.
Some that have the additional analytics might also include how much this was 'shared' through other means and that alone will spike a chart.
email, tweet, message etc...
Likes and dislikes are the 'engagement factor'. They were motivated enough to pick either.
How much weight is place on each is on the beholder because not everyone has access the same analytics.

And just to proof how big of a troll you are.... and how you are unable to grasp ANYTHING I posted, I read what you wrote and even the concepts are true they did not apply for the type of analysis I made and then I found this:

https://i.imgur.com/yiNMnRe.png

You copied and pasted an entire graph from a book explaining big data mistakes

Source

Meaning you DID NOT understand anything I posted, you went to google searched for "big data analysis errors", you copied and pasted an entire parragraph without understanding the context in which these errors occurr........ you came to the thread to try to show "the mistakes I made" when the part of the book you copied and pasted does not even apply for the analysis I created....:1orglaugh:1orglaugh:1orglaugh:1orglaug h

I am all for learning and correcting mistakes..... but... but.... what you did is just ridiculous.... can you be more pathetic than this?

I'd like to see more posts like this here.

Hopefully ! :2 cents: :thumbsup

A tweet containing a vid from facial abuse had 2.3 million views in 2 days

My intial gif reply was actually to give your thread a bump. Of course you did not get it. Then, I have said above that you have NO sense of humour, which I have proven again.

Of course I have pasted that shit from random google searches and did not even try to read that crap. Yet and at least for a day, you took it as reasonable assumtions. Therefore, it shows that neither of us know shit about stats or do not give rats ass if they are phoney...!

I do not need stats comparing apples to oranges, even though it is logical that if somebody has a million views he will have more positive or negative comments then a guy with 10,000 views ...wow what a discovery :1orglaugh:thumbsup

You should chill instead of getting righteous. I am done here.:2 cents: