Open in
Meme App
Open
Safari
Continue
Chrome
Continue
![Isis Meme: pepe
sad-frog
savepepe
smug-frog-b anti-meme
smug-frog-a
apustaja
0.6
0.4:
02-
Figure 6: Inter-cluster distance between allelusters with frog memes. Clusters are labeled with the origin (4 for 4chan, D for The Donald, and
G for Gab) and the meme name. To ease re adabi lity, we do not display all labels, abbreviate meme names, and only show an excerpt of all
relationships
in /pol, we find the #TrumpAnime controversy event [55]
Although, due o space constraints, this analysis is limited
where a political individual (Rick Wilson) offended the alt
right community, Donald Trump supporters, and anime fans
(an oddly intersecting set of interests of /pol/ users). Simi
larly, on The Donald and Gab, we find the #Cnnblack mail [31]
event, refering to the (alleged) blackmail of the Reddit user
that created the infamous video of Donald Tnump wresding
the CNN
to a single "family" of memes, our distance metric can actu
ally provide useful insights regarding the phylogenetic rela
tionships of any clusters. In fact, more extensive an alysis of
these relationships (through our pipeline) can facilitate the un
derstanding of the diffusion of ideas and infomation across the
Web, and provide a rigorous technique for large-scale analysis
of Intemet culture.
4.1.2 Memes' Branching Nature
4.1.3 Meme Visualization
We also use the custom distance metric (see Equation 1)
to visualize the clusters with annotations. We build a graph
G =(V.E), where V are the medoids of annotated clus-
Next, we study how memes evolve by looking at variants
across different clus ers. Intuitively, clusters that look alike
and/or are part of the same meme are grouped together under
the same branch of an evolutionary tree. We use the custom
distance metric introduced in Section 2.3, aiming to infer the
phylogenetic relationship between variants of memes. Since
there are 12.6K annotated clusters, we only report on a subset
ters and E the connections between medoids with distance un-
der a threshold K. Figure 7 shows a snapshot of the graph for
K0.45, chosen based on the frogs analysis dis cussed above.
In particular, we select this thresho kd as the majority of the
clusters from the same meme (note coloration in Figure 6) are
hierarchically connected with a hig her-level cluster at a dis
tance close to 045.
of variants. In particular, we focus on "frog" memes (c.g., Pepe
the Frog [45]): as discussed later in Section 4.2, this is one of
the most popular memes in our datasets
To case readability, we filter out nodes and edges that have
a sum of in- and out-degree less than 10, which leaves 40 % of
the nodes and 92% of the edges. Nodes are colored according
to their KYM annotation. NB: the graph is laid out using the
OpenOrd algorithm [61] and the distance between the compo-
nents in it does not exactdly match the actual distance metric.
The dendrogram in Figure 6 shows the hierarchical rela
tonship between groups of clusters of memes related to frogs.
Overall, there are 525 clusters of frogs, belonging to 23 differ
ent memes. The se clusters can be grouped into four large cat
egories, dominated by Apu Apustaja [27], Feels Bad Man/Sad
Frog [35]. Pepe the Frog [45], and Smug Frog [52]. The dif
ferent memes express different ideas or messages: e.g., Apu
Apustaja depicts a simple-minded non-native speaker using
broken English, while the Feels Bad ManSad Frog (ironically)
expresses dismay at a given situation, often accompanied with
We observe a large set of disconnected components, with each
component containing nodes of primarily one color. This indi
cates that our distance metric is indeed capturing the peculiar
ities of different memes.
Finally, note that an interactive version of the full graph is
publicly availabk from [
ave X" The dendrogram
"You will never do/be
text
shows a variant of Smug Frog (smueg-frog-b) related to a vari
ant of the Russ ian Anti Meme Law [51] (anti-meme) as well
as relationships between clusters from Pepe the Frog and Isis
meme [38], and between Smug Frog and Brexit-related clus-
ters [56], as shown in Appendix C
The distance metric quantifies the similarity of any two vari
4.2 Web Community-based Analysis
We now present a macro-perspective analysis of the Web
communities through the lens of memes. We assess the pres-
ence of different memes in each community, how popular they
are, and how they evolve. To this end, we examine the posts
from all four communities (Twitter, Reddit, /pol/, and Gab) that
contain images matching memes from fringe Web communities
/pol, The Donald, and Gab)
ants of different memes; however, recall that two clusters can
be close to each other even when the medoids are perceptually
different (see Section 2.3), as in the case of Smug Frog variants
in the smug-frog-a and smug-frog-b clusters (top of Figure 6)
8
aouesa
On the Origins of Memes by Means of Fringe Web Communities](https://pics.onsizzle.com/pepe-sad-frog-savepepe-smug-frog-b-anti-meme-smug-frog-a-apustaja-0-6-0-4-02-52734648.png)
