Last time I wrote about “(You)s“, or replies, as unofficial post rating system of 4chan.
Besides being the best way for detection of popular posts, 4chan’s reply system, provides us with data about shapes of conversations inside individual threads. By treating a thread as a graph, with posts as nodes and replies as edges that connect those nodes, we can get insight about a number of things about the thread and people that post in it.
For example we can easily detect centers of conversation and evaluate how strong they are. Or strings of separate conversations and how many and how long they are. Or how many disconnected posts are floating around in the thread etc.
We can then start classifying threads based on their network structure and see how it mirrors the type of content that can be found in those threads.
In this post I’ll look at 4chan threads as networks of replies, make some pretty images of graphs, and draw some conclusions based on those visual representations of threads.
I chose to draw edges as undirected, but it is purely an esthetic choice in this case. Instead of drawing lines as arrows, I chose to make nodes as big as number of replies they got. So the number of (You)s determines the size of a node. This way we can better see impact of individual posts, and I get to enjoy muh minimalism in graph design.
I decided to paint OP (original poster) node red, just to easily locate it and see its effect on the graph structure.
Now I’ve explained the rules by which a graph is created, we can get to graph images! Oh, and clean your monitor since some nodes are as small as 1 pixel and you don’t want dirty screen polluting the data.
The role of OP
Let’s have a look at graphs of a few threads that illustrate various roles OP can have in a thread.
This is a /b/ (Random) thread in which posters post results they got after doing the popular 8values quiz about political orientation. In threads like this, we can expect to find three main types of posts: quiz results, comments on other people’s quiz results, and both in one post.
When OP asks a question or explicitly sets the thread topic like in this case, every post of quiz result type is on-topic since it answers OP’s question. With this in mind, a lot of posts replying to OP are just being on-topic, like most disconnected nodes. We could consider OP’s post inflated in this sense, since most of its replies are directed towards the whole thread and are unnecessarily replying to OP.
Conversations in this thread get started with posts of quiz result type, and posts of comment and “both” type form strings of replies which can be rather long in some cases. The longer the string and the less it forks, the bigger are chances of there being only two people arguing (about their political views, in this case).
So it all started with OP who set the topic, and then posters found their political enemies and started arguing.
This is “Who do you prefer, Illya or Kuro?” from /a/ (Anime & Manga). Basically OP asks users to make a choice between two (underage, of course) anime characters.
Here OP acts the same way as in previous thread, but you can see how replies get connected more often. My guess is it’s because here people are all on the same page – they are all fans of the show and they are just playing a game of “would you rather”. They have no reason to be hostile and end their conversations.
This is a common type of thread on /b/ – rolling thread. OP sets the rules and everybody else rolls – makes a blank post or posts anything just to interpret the ID number of their post according to OPs rules. Users can roll to see some outcome or play the magic 8-ball game with set rules.
In this case too, replying to OP is the same as just posting in the thread. Strings of replies, however, are not conversations like in previous examples. Here those are usually “rerolls” and in rare cases people commenting on roll results.
And now for something completely different. 4chan’s Quest board /qst/ is used for playing different kinds of games using the imageboard infrastructure. In this case, the game is called Pokemon Island Adventure Quests and OP keeps the thread going by updating the game world state. Then other users vote for the next action by replying to OP and so on.
You can definitely see in the graph where voting is happening and which are OP’s posts. OP’s initial post in which rules of the game are explained is insignificant in comparison to OP’s posts where voting takes place.
Compare the Pokemon game with this one, Evil Nest Quest 3, also from /qst/. You don’t have to look at the content to see that this quest uses individual narrative roleplaying rather than collective voting mechanism.
General threads are threads that appear often in a board and serve as a sort of container for a topic, focusing the discussion on the single topic that interests a large number of board users.
Here’s a Trump general thread from /pol/ (Politically Incorrect). It is pretty decentralized and nodes are mostly connected in conversations. Focal points of all general threads are news, questions and bait – provocation from someone that isn’t interested in the topic.
These are two generals from /a/ – Little Witch Academia and Dragon Ball Super threads, and one from /lit/ (Literature) – Science Fiction and Fantasy General. As I suggested before, entertainment fandom seems to result in longer strings of replies than political fandom due to the nature of the subject.
And above are some generals in which asking questions is more common. The Beginner General from /ic/ (Artwork / Critique) and Stupid Questions Thread and Daily Programming Thread from /g/ (Technology) are more about helping each other deal with concrete problems than voicing opinions on the subject.
This manifests as a lot of closed shapes in graph structure. Many users ask specific questions, which creates free nodes. Then multiple users reply to these with their answers. Finally, the user that asked the question thanks everybody that answered him or reformulates the question. Longer strings represent troubleshooting process.
Let’s take a look at a couple of threads focused on content instead of OP or topic.
Content can be original (OC) or reposted from a source internal or external to the imageboard.
Here are two threads from a content-oriented board /gif/. Both are full of GIF and WEBM files, but one has content of pornographic nature, while the other aims to be funny (hence the name “YLYL” – You Laugh You Lose).
In the porno thread common reply can be asking for sauce (source material), while in the funny one common reply is declaring that you lost (because you laughed. This type of reply acts as a smiley face in a way, since for some reason 4chan users use emoticons only ironically).
This difference makes the funny one manifest in shorter strings of replies, since “I lost” comments don’t require a response of any kind, while asking for sauce does. They have stronger focal nodes too, which seems to indicate that sense of humor of /gif/ users is shared more than their taste in pornography.
Above are two threads from /wg/ (Wallpapers / General). One is about Vaporwave wallpapers and the other is about cartoon wallpapers.
Notice the parachute-like formation in the cartoon wallpapers thread and how it’s exactly opposite of the exploding shapes in the one about Vaporwave wallpapers.
In original content based threads there is a common practice of users rating content from many other posts. This manifests as a small node linking to a lot of other nodes.
Here’s the structure of a /wg/ Desktop Rate Thread. In it many posts are replying to many other posts and they end up forming the mess at the center of the graph.
Here’s a closer view of the graph, along with the text of a typical post.
And here are two threads from /t/ (Torrents), both starting off with big dumps of content. One is about GNU/Linux and IT learning resources and the other is about science fiction content. OPs in both threads provide the biggest data sources, so they stay the focus of the thread, at least for a while.
This is a trap (cross-dresser) thread from /b/. The content is mixed – some of it is original and some of it is reposted. A small number of users is using this thread as “doing requests” thread (there are also threads in which users ask for ideas for drawings, where users dare each other etc).
Original content in this kind of thread is mostly done as a reply to a request and this results in long strings of requests and replies with possible branching, depending on the nature of the content (how fast a request can be entertained).
And to wrap up this section about content providers, take a look at this YLYL thread from /b/. That huge black star of a node must be some extremely hilarious meme to outshine all other posts with its supreme dankness! Let’s take a look at the post that made such impact on the thread:
And let that be intro to the next section…
Here I’ll show you graphs of some bait threads, where OP is definitely fishing for replies. You can track how effective the bait is by looking at the size of the node.
As expected, baiting works best in boards that score high on neuroticism spectrum, so most drastic examples will probably be found on /b/ and /pol/.
Above is a thread from /pol/, in which OP asks “Why are people so fat in all the pictures I see of Americas far right?” Of course, the bait works and it gets replies of all kinds (serious, triggered, more bait etc). Feeding the troll results in bait clustering.
This one is also from /pol/ and it’s title is “How to debate a Jew?”
Here’s the quote of the OP of the thread:
I was trying to debate a liberal Jew today.
He made me look like a complete idiot because he always went and asked me “what is race” and “what is marriage” and being annoying in that way. Just like in fucking Mein Kampf. Like just as he describes it. But I couldn’t simply expose him for being a jew because “oy vey ad hominem” and the fact that it was in Berkeley.
Teach me senpai how to debate the jew without pointing him out directly.
Guaranteed replies – some call OP out for being stupid, some give him serious advice, some try to out OP as a troll, some try to troll OP etc… A bait cluster is formed.
BTW, notice how the noose shapes are beginning to form. Spooky…
The last example from /pol/ shows that not all bait posts must be just for lulz.
This thread’s title is “UNITED KINGDOM PM ANNOUNCES MASSIVE “ISLAMOPHOBIA” CRACKDOWN” (all caps!) and it shows a picture of an article from The Daily Beast.
News that are regarded as scandalous have similar effect to bait, but thread structure is “cleaner” then in pure bait threads.
Last but definitely not least is a typical bait thread from /b/. It’s a “Your mother will die in her sleep if you don’t reply to this post” thread.
On some 4chan boards this technique is used for destroying threads by filling them with garbage replies, but it is just senseless bait in this case.
Just look at it.
This post was all about what we can infer from simply looking at visual representations of these graphs.
I hope these images and my observations at least made you think about the relationship between content and structure of 4chan threads.
I’m quite sure threads can be classified more precisely and fed into some machine learning software that would learn to deduct types of thread content and types of posters and interactions only from thread structure.
This could find many uses in anonymous social networks of future dystopia. Combined with psychometrics, it can be used for targeted advertising. It can be used for planning and analyzing strategies of marketing campaigns. It can be used to train bots.
Oh and if you don’t reply to this post, your mother will die in her sleep tonight…