Introduction

Have you ever wondered if it’s possible to make a network of every single Minecraft YouTuber on the platform and figure out how they connect to each other? Ever wanted to spend two months of your life writing, running, and debugging code that has no meaningful purpose whatsoever?

No? You have “better things to do”? You have to “touch grass”? Not sure what that is, sounds gross.

Anyways, check this out.

A render of the graph

This is that graph. 78,000 YouTubers, connected by over 200,000 edges. I’ll admit, my data collection method was a bit flawed, so these may not all be Minecraft channels, and this likely doesn’t include every single channel that’s ever uploaded Minecraft, but I think it’s big enough to get some really interesting data out of.

For example, we can calculate who the mathematically most central person to the Minecraft community is. We can figure out what sub-clusters, or communities, of creators exist within MCYT. We can figure out which small channels are surprisingly important to the space. Stick around to answer these questions and so much more.

If you want to just see the results, the takeaways, and the fun stuff, skip to the Results or the Discussion. But if you’re at all interested in how this data was retrieved in the first place or how the analysis works, then just sit down, get some popcorn, and get ready to read.

Building the Graph

This network is a directed graph, where nodes are creators, and an edge from Creator A to Creator B means that Creator A mentioned Creator B in the title or description of one of their most recent hundred videos. The overall idea is pretty simple, but actually getting the data is an entirely different beast.

You need to crawl across YouTube like a gremlin, scrape videos from creators, detect any scent of mentioning another channel, and then crawl along to each and every one of those other channels to repeat the process.

The process itself looked like this:

1. First, get a seed. No, not a Minecraft seed, no not that seed either you weirdo. A seed of initial channels to start the search from. I got my list of channels from this tier list maker page.
   However, I did also have some criteria that a channel needed to meet in order to be selected as part of the seed:

   1. The majority (>50%) of their recent content (past 20 videos) had to be Minecraft-related

   2. They had to have 100,000 or more subscribers

   3. They had to either have uploaded in the past year, or have more than 10,000,000 collective views

With that said:

1. I made exceptions for two channels for being notable enough (determined by qualitative assessment) despite not having majority Minecraft related content as of recent: TommyInnit and PopularMMOs

2. This does not mean that channels that don’t meet this criteria cannot be selected to be in the final graph. As we’ll see later, a lot of channels that seem to be Non-Minecraft, or have very low subscriber counts, or have been inactive, ended up in the graph.

If you’re interested in seeing the seed list, check out the GitHub repo.

1. Next, use the YouTube Data API to scrape each channel’s last 100 videos to get their data, like video ID, but more importantly: titles and descriptions.

   Now, some readers might be familiar with the YouTube Data API, and might realize that there’s a hard cap of ten thousand tokens that can’t even be appealed unless you’re an actual company. If I’m scraping hundreds of videos from tens of thousands of channels, wouldn’t this take months to complete?

   Well, dear reader, there’s nothing a little intricate code design and an absurd number of API keys can’t fix.

2. Next, find which creators have been mentioned. There are a few ways this could be done. In most cases, the easiest way would be to use natural language processing and a technique called “named entity recognition” or NER.

   The problem is that most NER models are trained to detect things like human or organization names from standard text mediums, like newspapers or blog posts. So it’s a lot less reliable when you’re trying to extract MumboJumbo from the description of a Minecraft video.\

   So this is where I had to resort to the… dark arts. LLMs. gasp.

   It wasn’t my first option, but when the alternative was to spend a month manually training an NLP model to understand that Rekrap is a person and not some flesh-eating disease, I chose to be able to leave my room sometime in my lifetime.

   Using an LLM is pretty simple, though; just feed it the video data and tell it to return all the mentioned channels. The exact prompt I used is in the GitHub repo. I initially used Claude, which got expensive, so I switched to Gemini, which had ridiculous rate limits, and so I ended up landing where most people do when they need a cheap and efficient solution to a tedious problem: the Chinese.

3. Each time you find a channel, queue it to be searched later. Also, make sure to store which “hop” it was found on. Channels that are immediately linked to from a seed channel are at Hop 1; channels that are linked from Hop 1 channels are at Hop 2, and so on.
   For the sake of not going too far out from the Minecraft community, not spending a year on compute time, and not draining my bank account for DeepSeek tokens, I decided to cut off the search at Hop 2 channels. That might seem short, but even then we got over seventy thousand channels and started to leak out of the Minecraft community.

4. Also, each time you find a channel, create a node for it on the graph, and make an edge from the source channel to the discovered channel. These edges also store the video data for the video that the channel was mentioned in for super sick analysis later.

5. And then… just wait. Let it run 24/7 for a week or so.

6. Finally: profit.

Analyzing the Graph

Turning something into a graph makes it instantly more cool, since graphs make certain properties way more apparent. Take your friend group, for example. Make a graph showing who’s friends with who. Go out as many hops as you like. You’ll find out who your unknown mutuals are, who’s the most connected person, who’s the most central person, stuff like that.

Like, look how cool mine is!

Anyways, I should define a couple of graph properties so that they actually make sense when I talk about them later:

Property Definitions

Degree: The degree of a node is the total number of connections it has. The in-degree of Creator A is how many people link to Creator A, and the out-degree of Creator A is how many people Creator A links to.

Degree Ratio: The degree ratio is defined as the ratio of in-degree to out-degree. A large degree ratio implies that a node is linked to quite a lot, but does not link out to many other channels.

Centrality: A measure of how important a node is in a graph. There are actually a few different types of centrality, like degree centrality, closeness centrality, or eigenvector centrality. But in this analysis, I mostly focused on betweenness centrality. Which is a measure of how many shortest paths the node sits on. In non-nerd speak: it shows how “bridge-like” a node is. If the node is removed, how many creators would be more separated than they are now.

Hidden Influence: This is actually a custom measure I made myself. It’s a composite score of degree and the inverse of subscriber count. Basically, channels with a lot of connections but not a lot of subscribers score highly, and channels with relatively few connections compared to their degree score poorly.

Community: A subset of the graph where nodes within a community are more connected to nodes in the same community than those outside of it.

Insularity: How closed off a community is. High insularity means that nodes within a community are literally only connected to other nodes within the community. Low insularity, or porosity, means that a community frequently connects with nodes in other communities.

Bridging: If a creator “bridges” a community, it means that they interact with that community. It’s similar to insularity, but for individual creators. A high number of bridges means the creator is well connected outside of their own community, and vice versa.

Cosine Similarity: A measure of how similar two channels are based on who they mention.

Imagine that the total outgoing edges from a channel can be summarized in an arrow (or vector). The direction that this arrow points is dependent on which specific channels are mentioned, and how many times the channel was mentioned.

Cosine similarity is just taking the arrows for two different channels and finding the cosine of the angle between them.

Channels that point in the same direction have a cosine similarity of 1, and channels that point in completely different directions have a similarity of 0, since, if you remember from high school trig, the cosine of zero is one and the cosine of ninety is zero.

Stay in school kids, it makes you better at Minecraft.

For more information on how exactly I made some of these measurements, and a more detailed explanation of them, check out the GitHub.

Jeez. That was a lot. Can we get to the fun stuff now?

Results

Basic Graph Stats

The final graph had 78,275 nodes and 207,846 edges. The approximated diameter (longest simple path) is 13. The average path length is 6.1, mirroring the popular “six degrees of separation” phenomenon in many social networks

Degree Distribution

The degree distribution follows a power law, which is to be expected for a social network. The top creators are exponentially more connected than the rest of the network.

Identified Communities

35 unique communities were identified, and are shown in the figures. For a full list, check out the web app. Two different Music communities were identified, along with several distinct cultural communities, such as German, Brazilian, and Indian.

Communities ranked by insularity

Community sizes

The insularity for each community ranged from approximately 0.68 to 1, with the Analog Horror/Undertale community and the Music 2 community being completely insular.

Hidden Influence

The top 20 creators with the greatest hidden influence, along with the 20 creators with the least hidden influence, are shown below.

Top 20 and Bottom 20 Hidden Influence Channels

Although many of the creators at the top of the leaderboard have substantial subscriber counts (avg = 6,975,113), they are well connected enough within the Minecraft community to merit large hidden influence scores.

Conversely, the creators at the bottom of the leaderboard have extremely large subscriber counts (avg = 155,888,888), but are largely irrelevant in the Minecraft community. MrBeast is the second-lowest scorer in terms of hidden influence, despite being well known among the Minecraft community.

Degree Ratio

The top 20 creators with the greatest degree ratio, along with the 20 creators with the lowest degree ratio, are shown below.

Top 20 and Bottom 20 Degree Ratio Channels

Many of the channels with low degree ratios are documentary channels (e.g., XayXay, Abyssoft), second channels for existing creators (e.g., SandwichLord VODS, jschlatt VODS), fan channels (e.g., izethetic), or data collection channels (e.g., MaxData, Dangraphs).

Bridges

The top 20 creators with the most bridges are shown below.

The majority of the high scorers are not predominantly “Minecraft” channels; rather, they seem to either be channels that are generally popular and prevalent in the YouTube space (e.g., MrBeast, Ryan Trahan), or documentary/data collection channels (e.g., Abyssoft, XayXay).

Creators ranked by how many communities they bridge

Centrality

The top 20 most central creators, as measured by betweenness centrality, are shown below. The top scorer had a centrality of 0.006, meaning they lie on 0.6% of all shortest paths between creators.

Unlike the Bridges leaderboard, many of these creators are well known and present within the Minecraft space specifically.

Creators ranked by centrality

Subcount Statistics

A table of average and median subcounts for each community, along with their skew (how much large or small outliers skew the mean), can be found below.

The Indonesian community has the largest median subscriber count, at 626,000, whereas Modern MCYT has the smallest median subscriber count, at 338.

Subscriber count statistics by community

Discussion

Language Communities

Six unique language/national communities (Indian, Indonesian, Brazilian, Spanish, French, and German) emerged from the community identification process. This is interesting yet not too surprising in hindsight. Even though the community identification process had no explicit linguistic filter, it makes sense that creators more frequently mention others who speak the same language.

Out of curiosity, I took a look at the communities that popular bilingual Minecraft YouTuber and international Minecraft SMP founder Quackity is connected to. Quackity is connected to 12 communities outside his own (Core MCYT), which include 4/6 of the discovered language communities. It seems that he isn’t directly connected with any creators from the Indonesian or German communities.

Interestingly, there actually are no Minecraft channels that link to both the Indonesian and German communities, and there is only one channel in the entire graph that links to every single language community. The social platform data analysis channel: MaxData

The Segregation of Minecraft

Minecraft itself was broken into four primary subcommunities, which also happened to be the four largest communities in the network: Hypixel, Mature Minecraft/Hermitcraft, Core MCYT, Modern MCYT (listed in descending order of size).

Hypixel

The Hypixel community consists of your classic Hypixel/Minigame channels, as well as a lot of the competitive/speedrunning Minecraft channels. For Hypixel, think NotNico, Lego Maestro, Purpled. For competitive, think SpeedSilver, Feinberg, Fruitberries, Couriway. Despite how well some of these channels may fit in the other Minecraft communities, the connections within the Hypixel community are so insular that they are separated from the larger MCYT landscape.

Mature Minecraft

Mature Minecraft/Hermitcraft mostly consists of older, more “laid-back” members of the MCYT community. This has a strong overlap with the Hermitcraft community, given their relative age compared to other Minecraft creators. Many redstone and technical creators also fell into this community, such as Rays Works, ianxofour, ilmango, Purplers, and more. I was personally surprised that redstone didn’t come out as its own community, but given the close ties between technical Minecraft and the Hermitcraft community, it’s not too unusual.

I would also like to point out that it’s very funny that Kevin MacLeod got put here. If you don’t know, Kevin MacLeod makes a bunch of copyright free music that YouTubers love to use. You’ve absolutely heard one of his tracks, even if you don’t know it (see Sneaky Snitch or Monkeys Spinning Monkeys).

Core MCYT

Next, we have the Core MCYT. These are channels that I determined to be from the 2013 to 2022 era of Minecraft YouTube. That’s a long time period, especially in internet years, so it captures both the OG creators like DanTDM, CaptainSparklez, and StampyLongHead, and the more recent “DreamSMP” creators, like Dream, Technoblade, and TommyInnit. I think the key connecting factor between these two sets of creators, and the reason why the community wasn’t split into two, is the 2019 phenomenon we call “Minecraft Monday”. That would also explain why PewDiePie and James Charles are here.

Modern MCYT

Lastly: Modern MCYT. This is your Lifesteal creators, your Unstable SMP players, your Parrot, ClownPierce, Rekrap, and, very surprisingly: MumboJumbo? Nearing the end of 2021, Minecraft YouTube started to make a pivot that, I believe, was kickstarted by the start of the Lifesteal SMP in July of 2021.

Also, unlike how the old MCYTs and more contemporary MCYTs came together in projects like Minecraft Monday and SMPEarth, the “Modern MCYT” community never had that sort of bridging event, resulting in their separation from the rest of the network.

Moreover, unlike previous eras of Minecraft YouTube, when the game was gradually building a footing for itself on the platform and establishing itself to have a substantial audience, Modern MCYT emerged when Minecraft had already proven itself to be extremely lucrative for content creation. So, consequently, this stage was/is characterized by an emphasis on algorithmic optimization through fast-paced editing, advanced storytelling, and attention-grabbing concepts. Core MCYT was Minecraft players making YouTube videos, Modern MCYT is YouTubers making Minecraft videos.

And I believe this is why MumboJumbo fell into this category rather than the Hermitcraft or Core MCYT communities.

Keywords and the Algorithm

Upon further inspection of the graph, practically all of the edges connected to Mumbo in the Modern MCYT community are incoming, meaning he doesn’t directly link to anyone else in the Modern MCYT community. I thought this was strange, so I found the video data for some of the videos that link from Modern MCYT creators to Mumbo, and I found this beautiful little tidbit at the bottom of the description in practically every Lifesteal or Modern SMP video:

“Not Technoblade / TommyInnit / or any other Dream SMP member. This is a new and original Minecraft idea like HermitCraft or LastLife with Grian and MumboJumbo. Survival is harder than Hardcore Minecraft because PVP is on.”

If you don’t understand why someone would put a bunch of slightly related stuff in their description, don’t worry; I was confused at first too. But after some research, I learned that this is what we in the biz call: keyword spamming. Just throw a bunch of popular names into the description and hope the algorithm associates you with them.

That specific line of text was present in a lot of the videos that were linked to Mumbo, so I think it’s safe to assume that that’s a large part of the reason why he was associated with Modern MCYT. At the end of the day, the graph measures mentions, not relations, and mentionship is not at all equal to relationship.

Unexpected Communities

So at this point I think it’s worth mentioning the elephant in the room. Well, actually the 25 elephants. The 25/35 communities that seem to have nothing to do with Minecraft whatsoever.

And honestly, I think that’s one of the most interesting things that this graph revealed. Minecraft creators are pervasive enough within the larger content creation landscape for an initial seed list of only MCYTs to breach containment into Pokémon, Education, Shooter Games, and more.

And also, since the graph consists of virtually one component (there are one or two microscopic disconnected components of 1-2 nodes), each of these channels does somehow connect to a larger Minecraft community. Even the communities with 100% insularity (Analog Horror/Undertale and Music 2), though they might not have any outgoing edges to a Minecraft community, they have incoming edges from a Minecraft or Minecraft-adjacent community.

Adding one last thing before I move on: the Yogscast someone got themselves their own community. Despite how foundational they were to early MCYT, they didn’t fit into any of the other Minecraft communities and were influential enough to warrant one all for themselves. Though, it’s not all too surprising given how large and diverse the Yogscast team is now.

MrBeast

Is it possible to do a YouTube analysis project without mentioning the big man himself? The media mogul? The Lunchly leader? The Feastables founder? You know who I’m talking about: Mr. Jimothy Beast himself.

Mr. Beast scored #1 in sub count (no shit, really?), #3 in terms of centrality, #1 in bridge count, and, interestingly, #-2 in hidden influence. No, not #2: # - 2. He came second to last in terms of hidden influence, scoring 18.03, only better than T-Series’s score of checks notes… 0.

Centrality and Bridge Count

His high scores in centrality and bridge count make sense. Mr. Beast interacts with a lot of different types of creators on the platform, so naturally, he’s going to be well connected outside of his own community. This also inherently increases his centrality since betweenness centrality is a measure of how many shortest paths have MrBeast on them. Since MrBeast connects many different communities, the shortest path between two creators from different communities is likely to be through MrBeast.

Out of curiosity, I looked at exactly which communities the Beast doesn’t bridge, since he only bridges 26 out of the 35 total communities. These turned out to be: Guns/Military, Animators, Yogscast, Hardcore Gaming, Mixed, Indonesian, SCP/Lore/Anime, Analog Horror/Undertale, and Music 2. The last two make complete sense, since those two communities have 100% insularity; it’s unlikely that MrBeast would be directly connected to either of them.

However, I was surprised by his apparent lack of connection to the Animation and Hardcore Gaming communities. My intuition was that he’d be linked to the Animators through JaidenAnimations, but it turns out that Jaiden actually ended up in the Pokémon community, which is Beast’s connection there.

The Hardcore Gaming community uses hardcore with two definitions: difficult games, and taking games very seriously (gaming culture, gaming industry, etc.). So it seems that MrBeast isn’t very relevant or connected within the broader video gaming culture.

Hidden Influence

His pitifully low hidden influence is also reasonable given his massive subscriber count. The other channels at the bottom of the hidden influence leaderboard are [literally]{.underline} the other most-subscribed channels on the platform. Though, I would have expected him to place just a tad bit higher given his strong connections with the content creation community when compared to massive brands like T-Series and Cocomelon. Subscriber count can’t be the sole reason, since even relatively large channels like Dream (30m+ subscribers) scored very highly.

So, MrBeast has virtually no hidden influence, and comes in at #3 in terms of centrality. But who comes in #1? Who is the most important and central person in the entire Minecraft community? Who is the glue that holds 17 different communities together? Who beat Minecraft with a steering wheel and farmed potatoes day and night for eight months?

Technoblade

Of course, it’s Techno. The man who participated in nearly every major Minecraft SMP and event for over a decade. The man feared across the entire community for his unparalleled PvP, and beloved for his unparalleled content and personality.

I knew Techno would place well, but I didn’t expect him to be #1, showing just how important he was even four years after his passing. Because, besides his centrality, I also think that he was a part of the 2022 pivotal shift in Minecraft YouTube, splitting Modern MCYT from Core MCYT.

Technoblade was genuinely one of the best creators on the platform. It’s hard to overstate how unique his combination of humor, skill, drive, and just general character was, especially in the Minecraft community. And although he passed almost five years ago, there is absolutely no doubt in my mind that he has and will continue to shape the Minecraft community for years to come. He taught us the power of ADHD, the art of war, and made it clear that “no matter how ridiculous the odds may seem, within us resides the power to overcome these challenges and achieve something beautiful.”

Technoblade never dies.