Study With Me

StudyTube

Introduction


Across the globe, students dedicate countless hours to their education, grappling with the demands of schoolwork during the day and homework by night. But unless you're some kind of genius, to keep up with society's pressure to do well in school can be a difficult task and one many people give up on.  Studying is rarely fun but it's something most have to deal with every day.  But studying isn't only academic- whether it's professionals seeking to upskill, someone discovering a new language or hobby, or a kid just looking to get better at  a video game, learning is everywhere. 


Whether it’s to free up more time to spend with loved ones or catch up to your classmates, making studying more efficient would be a lifesaver for many.  That's why there are countless books, articles, and even university courses promising students the best and most effective ways to improve their studying strategies. 


In terms of accessibility and popularity though, it is hard to beat the second most visited website in the world.

YouTube may mostly be known as a source of entertainment- but it also has a vast ocean of diverse educational content. From makeup tutorials and home repair to software tutorials and nature documentaries, it offers something for everyone.  It's even developed its own genre for the study of studying- commonly referred to as "studytube". Whether it's "study with me" sessions, day-in-the-life vlogs, or tips for acing your next exam, studytubers have created a space that's both informative and incredibly supportive. 


However, it seems like just about every day there is a new video that claims to have the BEST way of learning that's going to save you so much time and headache, and if you don't click on this video you'll forever be behind those that do know the Top Ten Secret Tips of Ivy League students, and your life will be ruined... Ok, maybe most people don't go that far. Regardless, the proliferation of these types of videos can overwhelm viewers. With new videos continuously being released, it becomes challenging to discern if they are genuinely useful or if the content is hidden beneath layers of clickbait. 

How do you navigate this vast landscape of educational content? 

 

The best choice of action is probably to just stick to a few trusted videos and apply their tips rather than getting stuck in an endless search for the perfect study hack. But if you're the curious type who wants to know the context, or if you really feel like you might be missing something... Then hopefully this project will help!

Project

Goal:

The goal of this project is to create a library that organizes content from 800 videos across 100 channels, offering a comprehensive overview of StudyTube. This will not only provide access to the wealth of content available but also allow for an understanding of YouTube's educational landscape and the unique insights it offers.


Inspirations for this project include my noticing that for most study advice videos, the top comment is usually an outline of the video's key points. 

Whether it's to save viewers time or to act as a review, it means most viewers appreciate having the content of the video summarized.


Data Collection


Search Process

I searched terms such as ‘study tips’ and ‘how to learn fast’. I then followed the algorithm and clicked on videos in the sidebar.

For each study video I went on, I also went to their channel to see if they had more videos or a playlist of their study-related videos (which most channels did have).


Selection Criteria

Because the focus of this project is on general studying advice, videos are focused on general studying strategies and school topics, so videos that are based on a certain major (Top 10 Engineering Major Tips) or a specific test (How to pass your GREs) are not included.


Collecting Data

Using the search process above, I got 812 videos. I then used the YouTube API to gather detailed information from each one, such as date published, length, channel name, etc.

After that, it was time to let OpenAI’s Whisper do the heavy work (https://github.com/Dicklesworthstone/bulk_transcribe_youtube_videos_from_playlist)  to automatically transcribe each video in better quality than YouTube’s automatic transcriptions. 

Finally, with the assistance of GPT 3.5, I extracted summaries/key points from each video and categorized each point, checking manually for errors. 

I also added a few columns of my own for further analysis- categorizing each channel according to a channel type and noting each YouTuber's gender.


The Data Itself


A transcript is a form of data that was never meant to be. 


YouTube scripts differ greatly from podcasts, articles, and books in that they are are written or improvised for the video medium. If a speaker is present, you miss out on body language and facial expressions. Even without a person on screen, not only is the visual presentation component completely gone, but tonal cues are impossible to get. 

These study advice video transcripts are particularly different from courses too- While they are presentations, similar in structure to a basic persuasive essay, they are usually much more casual in format.  Noting these qualities can be important if working directly in NLP, which I originally tried to do (Before realizing manually annotating thousands of sentences was no fun).


Thankfully, these textual qualities don't affect the content too much, especially the way we are extracting the key points (A mix of manual labor and LLMs)


Rather it's things like YouTube being a live datasource. Not only are new videos published every day that could have been good candidates for this dataset, but videos are often privated/deleted. This happened in the middle of my work as well, so it was a good reminder to account for that.

Exploratory Analysis


Once all 800 videos are in, we can begin asking the questions. 

How long would it take to watch all the videos? How many people have these videos actually reached?

For total length we have 5 and a half days of content, or 132 hours of lecture- if it was a college course, it'd be three semesters long. (15 weeks * 2.5 hours per week for a single class).  As for total views, that comes to about 400 million. 

Total subscribers across all channels wouldn't be a very accurate indicator because there are some huge channels who only make one or two videos on study advice (Such as TED, Talks at Google, etc.). If you exclude the mega-channels who aren't really part of the video genre though, (who have about 100 million subs together) we get a total subscriber count of about 50 million.  I go into more detail on channel types and comparison later, at the Tableau dashboard portion of this project.

For our oldest video in the dataset, we have Scott Young's "Learn Faster", wheras the newest video is jspark's "This SECRET makes STUDYING FUN and EFFORTLESS". It's a surprisingly fitting contrast to the evolution of this genre how social media has matured as a whole (to focus more on clickbait-esque titles).

Scott Young is famous for his MIT challenge, where he took it on himself to learn the entirety of MIT's CS degree in one year.

jspark probably hadn't even entered first grade when the first video here was created, but he is now a highschool student who is catching up to Scott in subscribers. These channels are both still active, with Scott making an hour long conversation video with one of the current giants in studytube, Dr. Justin Sung.

Since we have the titles for each video, we can also check things like which videos claim to be the best! 

I used a bunch of keywords related to the more outlandish claims, and checked for the top ten videos by views.

So in a sense there you have it, the Top Ten best Top Best Study Advice videos!


We can also look at the channels that make the most use of clickbait words in their videos. Here we can see that seven out of the 100 videos have ALL their videos be... designed to garner attention.

We can also search within the full transcripts.

For example, by counting each video that contains "sponsor" (so "sponsored by", etc. will be included as well) we can quickly find out which channels have the most sponsored videos.

Tips

Now for the most important part of the project- extracting the actual content from the transcripts. 

This project was initially meant to be an NLP project, but after trying and failing to get anything to work better than ChatGPT, I decided for simplicity's sake I would just go ahead and use it to summarize the key points, and just check for errors myself. 

Taking each transcript and boiling them down to their main points, I then categorized each "point" according to a list of study advice types such as Self Care, Active Recall, Feynman technique, etc. Unfortunately an objective list of study techniques probably could never exist, but this is based on my own research/generally accepted terms.

Here are some word clouds so you get an idea of their content. (From left to right: Anki, Self-Care, Social)


For full list of categories and their explanations, click to expand.

Organize: Organizing your study space/materials.




Mindset: Cultivating a positive and forgiving attitude towards oneself.




Environment: Optimizing physical surroundings for better focus.




Schedule: Planning your day with specific time allocations for tasks.




Chunk: Breaking down information into manageable parts.




Self-Care: Prioritizing personal health and well-being during study sessions.




Resources/Tools: Utilizing tools and resources to enhance productivity.




Encode: Linking new information to what you already know.




Notetaking: Effective strategies for recording and reviewing information.




Active Recall: Testing yourself to enhance memory retention.




Breaks: Taking short breaks to improve overall study efficiency.




Anki: Using digital flashcards for efficient study revision.




Social: Leveraging group study for a more enjoyable learning process.




Prepare: Starting early with preparation and review for cumulative learning.




Prioritize: Focusing on one subject at a time to enhance learning.




Engage: Participating actively in class for better information retention.




Mindmap: Using visual diagrams to organize information and ideas.




Ask for Help: Seeking clarification and understanding when needed.




Deliberate Practice: Concentrating on weak areas for improvement.




Interleave: Mixing different subjects or types of problems to enhance learning.




Teach Others: Enhancing understanding by explaining concepts to others.




Memorize: Committing essential facts to memory for quick recall.




Music: Listening to music to reduce stress and enhance focus.



Prime/Skim: Previewing material to focus on key points during deeper study.




Read: Regular reading practice to improve comprehension skills.




Rote Repetition: Practicing repeatedly for skill mastery.




Cue: Using sensory cues to enhance memory retention.




Mnemonic: Employing memory aids to recall information.




Exam: Techniques for maximizing performance on exams.




Fact: Understanding the process of learning through experience.




Memory Palace: Using spatial memory to organize and recall information.




Consistency: Building steady habits to support long-term learning goals.


Tableau Dashboard

As mentioned previously, within the studytube genre I further divided the channels into types for a better analysis of the genre as a whole. Here's a quick guide.

Channel Guide:

Aesthetic:

A genre that focuses on pretty videos and notes. They have a softer style.

Expert:

A youtuber that brands themselves as an expert on the subject. Usually an adult or working professional.

Show:

Channels like TED ED, Talks at Google, Crash Course, etc.
High production value, usually not based on a single personality or topic.

Student:

A youtuber that brands themselves as a fellow student.
Usually an academically successful one, who shares tips that worked for them.

Vlogger:

Similar to a student youtuber, but more holistic(?)

They have at most 50% videos pertaining directly to studying, with more vlogs about their travels or school life. 



Using these types makes it easier to look at channel trends. 

For example, previously we saw that the earliest video was an "Expert" type video by Scott Young, and the latest was a "Student" type by jspark.

A single video from first and last isn't really enough to get a grasp on how things are trending as a whole, but by looking at all the channels that have been active by year, we get this:

Student-type videos have become increasingly popular, particularly from the late 2010s through 2023, where there was a noticeable jump to 50%. This trend continues into 2024, with Student types being  the majority as well. 

However, the data I collected is still limited. Even though I tried my best to look around, it may have been that Youtube's algorithm promotes more recent videos.  It could also be that student channels tend to not stick around, so because the videos I collected were mostly recent, the older student channels disappeared. Student channels do seem to be easier to start but also die, and have less of a need to build a "brand" compared to Expert types. 

Insights


Top Videos and Top Channels:

Here we have the top ten videos by views, likes, and comments, to give you a quick glance at the best videos according to Youtube's algorithm.
The only noteworthy outlier here is the top video by comments, which happened because there was a "call to action" in the video. A reminder of how comment count is a metric that can be dependent on something that has nothing to do with the content of the video.
Top Channels is mostly to give you an overview of the most successful channels. When set to ALL, Show types dominate because they are usually made by professional groups/have a huge variety of content. Student type videos have some of the highest like to view ratios, possibly because of the cameraderie the audience may feel towards them.


Types Overview:

While there are more Student channels than any other type, Expert channels have created the most videos.
This makes sense as Student/Expert types would have the most content focused on studying, and while it's easier to make a small channel with the qualifications of student, experts may treat being a youtuber more seriously and make more videos overall. Show type channels obviously demolish the rest in average subs and views.


Quality and Engagement:

Student type videos seem to have the most likes, and Expert type videos are up there when we look at comments. From a cursory look at the videos, Expert videos might have more people say "thanks for the tips", possibly because they are more authoritative in nature? Regardless, that brings us to our next dashboard...


Gender:

Here we can see the clear difference in terms of gender representation with the types of channels, despite the number of channels for each gender being pretty much equal. Aesthetic and Vlog types are almost completely female, while Expert and Student are majority male. Of course, with my sample size not being very big, it's hard to come to any strong conclusions- but it's still a very noticeable effect.


Tips:

Expert type channels make up most of the videos, and Show type the least, so there's nothing too surprising here. The proportion for each type seems similar, which means most likely the type of channel doesnt have much of an effect on what categories of tips will be talked about.


Corpus Overview:

The majority of channels Youtube recommended are more recent, but there are still many classic channels from the past that have survived relevancy. The data is also a ~bit~ unbalanced, considering the median videos per channel is 2, but some very prolific channels such as Study with Jess have... 140.

Web App

To bring this analysis to life, I cobbled together a web app that serves as a dynamic catalog for these videos. 


 Users can search, filter, and sort through the library, diving deep into specific categories or exploring miscellaneous study advice for fresh perspectives. 

If you're interested in any of the categories, of those categories is a weakness, you can quickly see how the most popular videos talk about it, see if recent videos have added anything new, or just look in general to see which channel has a style that matches.




Clicking on a video leads to a detailed page featuring an embedded player, key points, tags, and a complete transcript, ensuring a comprehensive view of the content.



Conclusion


This web app offers a practical tool for learners to find videos that match their learning needs, complete with summaries, tags, and transcripts. 

Whether you're trying to find a new study hack, learn more about methods you're already familiar with, or gain context to the evolution of these ideas as a whole- this library is meant to simplify your search. Hopefully it's in the spirit of the power of community and technology in making education more open and accessible to everyone.