Hey there!

As we all know, Pewds turned off the comments section for his YouTube videos a while ago. In my opinion, this was a good choice and not only Pewds, but also other YouTubers have problems with bad, negative or just stupid comments. Therefore, I think the problem should rather be solved, where it actually comes up - in the technical aspects of the YouTube comments section.

I have written the following text and tweeted it to some YouTubers, which mentioned the low quality of the YouTube comments in recent time. I hope that some parts of this solution or idea are useful and might even bring back a nice and good YouTube comment section in the future.

Maybe some of you even have additional ideas, which could actually solve the problem. Let us know!



MOANING ABOUT BAD #youtube #comments? - NOT ANYMORE!
Idea for finding a long term solution #YoutubeCommentSolutions

1) The problem

...is actually a serious one. Negative and bad comments on YouTube videos are not only annoying for the authors of the videos, but also for other community members. Unfortunately, these insults, self-ads or other crappy stuff is nowadays common in such comment sections. There is no point in manually removing such comments because there are simply too many bad comments and just after you might have deleted some of them, even more are there afterwards. This is the reason why I think that only intelligent computer systems can help to get rid of such comments and are the only opportunity for a long term solution.

2) Background knowledge

Just recently, Google published some research results regarding advanced methods for statistical machine translation systems. For example, these SMT systems are used in Google Translate. Apparently, Google Translate does not actually "know" how to translate texts from one language to another one, since this is impossible for computers today. Instead, their systems "learned" how translations can be done by analyzing a whole bunch of already existing translations of humans. Basically, the idea is to use probabilities, which measure, what sentence is probably the best translation for a given sentence. Often the results of Google Translate are not perfect, but translation is very complicated in comparision to the idea, which is presented under 3).

This whole topic might seem unrelated to YouTube comments, but actually it is indeed related to it. This proves that Google already has the knowledge and technology to use such statistical systems and could use it for YouTube comments aswell. In this case, the statistics could be used to determine, whether a comment is such a negative and bad comment or not. The results will never be perfect, but using these mechanisms, YouTube could get rid of a whole bunch of these comments automatically. If I remember correctly, PewDiePie once said in a video that there is a kind of simple blacklist word-filter. This method is very simple and the results are not satisfying because most bad comments will not get filtered out and even some good comments might get sorted out. The ratio between these two could be way better with a statistical system, which does the whole magic.

3) The idea

There could be two sets of such statistical filters: A GENERAL set, which includes filters for insults, ads and so on. The second set should be a PRIVATE filter set, which only applies to personal negative comments. Every Youtube user can activate or deactivate specific filters from these sets for his own comment section. For example, one could activate the general filter for insults, but not the filter for self-ads.

The private filters should be used to get rid of comments, which are more related to the actual YouTube user. For example, people complaining about PewDiePie not playing horror games, CinnamonToastKen's girlfriend's clothes in front of the camera or simply Zoella poking Jim Chapman in the bum. These are the different cases for private filters because they are obviously not useful for all YouTube users.

Especially in the beginning, such a statistical comment filter system needs a lot of data to work correctly and build up statistics for the detection of bad comments. This is, where humans need to help the statistical filter systems to "learn" the correct detection of bad comments. For every comment, there could be a small dropdown menu, in which you can choose, whether a comment should be reported for a specific filter or not.

The general filters could use data of all YouTube users. Once a certain amount of users (relative to the amount of viewers) reported a comment as bad, the comment could be actually removed and the filters could be provided with the comments as data to "learn" from. The reports of YouTube partners might be counted as "more important" than the reports of usual YouTube users.

The private filters obviously can only take data from the specific YouTube user himself. In order to avoid possibly wrong conclusions of the private filters, a private filter should only get active after a specific amount of data was provided. For example, PewDiePie's private filter for horror games should only get active, once he has reported 15 or 20 comments for this filter, so the results of the filter are quite reliable.

4) Your opinion

What do you think of this idea? Maybe some of the youtubers or even the whole community could suggest the development of such a system to Google to get rid of bad and negative YouTube comments.

Hey Okuu and thanks for sharing your opinion!

On the one hand, you are right. People can still communicate here and share their thoughts and opinions with each other - just the same as in the comments section.

But on the other hand, I think that this forum here is rather a "workaround" than a real solution for the problem. Of course it actually works fine for the whole BroArmy. Unfortunately, not only Pewds, but rather the whole YouTube community is negatively affected by these bad and stupid comments.

Just think of the comment's section of other YouTuber's, such as Marzia. Although there is a section in this forum for her, she still has the comments active for her YouTube channel and there she might still receive these comments.

Because this thing is a problem for YouTube overall, I think that this forum here works fine for the BroArmy, but does not actually solve the problem overall. What do you think about that?

Hey Nixx!

Yeah, participation in channels and in their comments is indeed optional. But don't you think, that simply avoiding the comments section for all channels and videos would somehow lead to a worse YouTube overall experience?

In addition to that, I guess that also Pewds would benefit from re-activating the YouTube comments section, IF there were no negative comments, because there would be a higher overall communication within the BroArmy-community. I just compared the numbers of comments of old videos with the numbers of posts in this forum for the videos nowadays.

Old YouTube average amount of comments: ~ 10.000 comments
Video section average amount of comments: ~200 comments

Now let us assume that 75% percent of the YouTube comments were negative comments at that time, which might be a pretty high assumption. Then there would still be 2500 good comments in comparision to 200 good comments, which is about 12 times more.

I think the reason for this effect is the laziness of people: It is way simpler to just throw out a comment under a YouTube video than registering in the forums here and do the same thing here. I don't want to say that the forums here are bad or anything like that, but for most people it is an "obstacle" to do a registration somewhere and most people are then simply to lazy to register, which leads to a lower amount of active users.

Since Google has the technology and the knowledge to realize the idea mentioned above with the help of intelligent systems, it is quite a shame that they do not use these possibilities to greatly improve the quality of YouTube comments.

In order to clarify the idea a bit, I want to provide an example:

Small example:

Learning phase:
Comment 1: "Come to my channel and like and subscribe!" -> Reported as self-ad by YouTube community
Comment 2: "Like my videos and subscribe to my channel!" -> Reported as self-ad by Pewds
Comment 3: "Subscribe to my channel!" -> Reported as self-ad by YouTube community

Processing phase:
A combination of the words "channel", "my", "like", "subscribe" is probably self-ad and should be filtered out.

Automatically filtering phase:
Comment 4: "Like and subscribe my channel!" -> Automatically filtered out
Comment 5: "My channel is great! Like and subscribe!!" -> Automatically filtered out


This example might seem a bit artificial, but I think it clarifies the idea the best. Of course the system would not work pretty well for just 5 comments, but you should rather think of thousands of comments to learn from and of even more thousands of comments, which could then be filtered out automatically.

Hey Nixx,

it's nice to see that somebody else is also thinking about that topic, thanks for your answer.

Regarding your example of obstructions:
You are right that these make it harder to detect this comment as a negative comment, but I do not think that it is impossible to get even these sorted out, when enough effort is spent, e.g. removing dots and special characters or checking for similarity to "regular" words and using these instead then.

I do not think, that these systems are actually used in forums or comments sections nowadays. The approaches are quite similar, that is true. But the small detailed twist lies in that aspect, that today's systems often use "static" information rather than "statistical" information.

The small but important word in the given example above is, that the "combination" of these words is probably a bad comment.

For the long-comment example you provided above, "static" systems would just see that a certain amount of "bad" words is contained in it and therefore it would get blocked. If I understood Pewds correctly in one video, a similar blacklist system is already in use for YouTube comments. This is actually the reason why this would indeed lead to wrong results.

In contrast to that, the "statistical" systems do rather complicated probability calculations. If the probability that the comment is a bad one is above a certain threshold, the comment gets blocked. For our both examples combined that would mean: The "statistical" system only knows that the combination of the words "my", "channel", "like", "subscribe" in a very close area is probably a bad comment. That characteristic does not apply to the example you provided because there is no actual combination of these words, but instead they are spread all over the comment, which would reduce the probability in the calculations significantly.

In this point it is actually a question of details and because of the shortness of the sentences I provided in the example above, that aspect probably did not come out that clearly. I hope that this explanation now clarified that a bit.

This is actually the same step that researchers did for machine translation systems. In the beginning just simple dictionaries were used, each word was translated step by step. Obviously this "static" approach leads to crappy results - as it does for bad-comment-detection nowadays. With the small change of details and rather analyzing the context of each sentence by the usage of the probabilities, the quality of translations actually increased a lot because the words are more translated in their correct context.

Of course there will never be a "perfect" solution. There will always be some comments, which are removed incorrectly, and comments, which are not removed, although they should be removed. But I think that there is still much room for way better results.

In the other points you might be right:
When people are too lazy to register in this forums, they might also be too lazy to participate in reporting bad comments. I do not know, if YouTube partners themselves could spend enough efforts to collect enough data. In contrast to usual users, they might actually have a higher motivation level to do this and they could get a higher priority, when they report such a comment.
Of course a lower amount of comments is easier to manage. But I don't know if they actually need to be managed that intensively. I can imagine that Pewds is interested in the comments of the BroArmy and it is difficult to read all comments when there are several thousands of them. But on the other hand the communication inside the community (just as we do right now ) would be better and more fruitful, if there were more comments again.

What frequency of winky emoticons are you talking about, Nixx? Hehe .

Yeah, that is in fact that point, in which all good thoughts and ideas will find their end - unfortunately.

In a first step, I just wanted to check out, what you other Bros think about that topic, since we all directly noticed the results of that issue. I also hope to raise some awareness, that Google actually itself suggests technologies in research, which are very well suited as a solution for this YouTube-comment issue, but they refuse to use it in order to solve that problem.

As you just said, I am also a bit disappointed that nothing changed after Pewds turned off comments. Not only Pewds, but all YouTubers have exactly the same problem. Even some media reported on Pewds turning off the comments as a dramatic event, while YouTube and Google simply relax and chill in their nice seats and enjoy their time.

EDIT:
Okay, thanks for the information on that, mKeRix. Then the previous paragraph changes it's meaning to irrelevant and not really helping changes. And that some comments slip through, although this should not happen, just shows the high-quality of the software.

I do not actually think that contacting YouTube would lead to any improvements right now, just because of the reasons you mentioned, Nixx. It is too difficult / expensive / never-change-a-running-system-ish for them to change it. But there is still the hope that - maybe in a far far future - people will increase pressure on Google and YouTube to actually force some change.

Hey Piggeh! Yeah, that is exactly the point. As a worldwide company, they should definitely have the ressources not only for doing related research in this area, but also in actually making use of it.

Unfortunately, they will probably not change anything, unless they experience a lot of public pressure and see the necessity to change it or can make profit out of it, as Nixx pointed out earlier.

I think that there is actually a way how they can earn money with it. In this context "luckily" not only YouTube, but nearly any other web page has the same issues with those bad comments. Just think of reviews for products on Amazon, insulting tweets on Twitter and so on. They might be able to get millions out of it, once other huge internet companies would use their system. As a small motivation boost, they might also consider the opportunity of spending some cents for those people, who actually help to collect the data on YouTube.

Actually, this seems to be a plan for turning (verbal) dirt into gold. What do you think of it?