Hacker Newsnew | past | comments | ask | show | jobs | submitlogin
Ask HN: How to Learn
6 points by _fvh3 on Sept 6, 2022 | hide | past | favorite | 8 comments
Im looking to learn a previously ill/undocumented topic: the YouTube algorithm, yes I know it sounds childish etc, however im making high quality content of software development, AI dev, etc whilst making it digestible for YouTube, I know about the YouTube "meta" and the patterns which make a successful video but im not sure how to learn the algorithm.

From my current "research" I have gathered these things about successful content creation: https://pixeldrain.com/u/Ebazr6dT

I've made a whole page of notes and theory from professional social media analysts and top YT creators from these links: https://pixeldrain.com/u/cKUQB63g I am really just spotting patterns and learning this basic factors of the algorithm which may just be correlation, not causation, I know that the top creators know "the algorithm" and have actually learned it however it isn't publicly available on the internet as they have previously stated this in the past.

So im asking, how does someone learn about an undocumented topic without just "spotting patterns", im also asking for resources that show how the algorithm works or potential connections to people who know about it or are interested in it

Id figured this is a good place to ask as many people here have worked on complex algorithms and content creation



I assume you've seen this: https://www.youtube.com/watch?v=3A8kawxMOcQ (starting at 7:10)?

From the video vs your write up, it sounds like you it might be helpful to find a few other people to discuss your theories with them.

As for learn about an undocumented topic without just "spotting patterns" - that's a much broader / longer discussion. I've been a part of and adjacent to a few groups in that have done this professionally and the answer is mundane - think really hard about incentives and actually do the spotting of the patterns in a way where you aren't lying to yourself. If you do it in a group you can find ways to test things faster than just doing it by yourself.

As to where to find people people interested in this, maybe check out youtube channels just under the YouTube Partner Program eligibility. Which means they'll be hungry to try to get over the hurdle. As you work with people you'll slowly see who is serious and you can try to focus on working with them more closely.


I would never have thought about gathering a group of people and experimenting, I have just been studying what the top creators say and what is successful without actually testing anything or putting anything into practice.

I am in a few communities where I think people will share an interest in YouTube and will potentially join a discord server where everything can be co-ordinated, I think I will try something like that. Thanks for your reply!


I work for a media company that works with YouTube creators. From my experience you should add following items to your list: - new uploaded videos drive traffic to old videos of the same channel - schedule consistency and calendar: 2-3 videos at least each week. It's very important - links from other channels: it helps when other channels promote your videos. - live streams help boost views (see https://gyre.pro/ - I'm not affiliated with this, but I know creators use it and get good results)

Generally services like vidiq or tubebuddy help you. Many creators use them.

The algorithm that you talk about changes quite often. Most of creators make lots of experiments to find their algorithms.

Make sure to stick to YouTube community guidelines. Seriously, YouTube can ban you forever if something in their view goes wrong.


You are going to make so-called reverse engineering which is an essence of learning (if not talking about dealing with some bits of machine language only). But I have no idea how are you going to accomplish the task having neither code/program nor data such as users and videos.


I dont think that human can learn this algorithms.

You should use machine learning. I think that those are ranking algorithm. Basically create a training set of known examples that you have, and try to predict the same behavior.

I do not think that this can be done via rule based approach.


For mobile 1st link: https://pixeldrain.com/u/Ebazr6dT (Paragraph 2); 2nd link: https://pixeldrain.com/u/cKUQB63g (Paragraph 3)


Thumbnails with sex appeal help with views.


This guy knows what he is talking about




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: