The Exceptional and the Everyday: 144 Hours in Kiev

Abstract

How can we use computational analysis and visualization of content and interactions on social media network to write histories? Traditionally, historical timelines of social and political upheavals give us only distant views of the events, and singular interpretation of a person constructing the timeline. However, using social media as our source, we can potentially present many thousands of individual views of the events. We can also include representation of the everyday life next to the accounts of the exceptional events. This paper explores these ideas using a particular case study – images shared by people in Kiev on Instagram during 2014 Ukranian Revolution. Using Instagram public API we collected 13208 geo-coded images shared by 6165 Instagram users in the central part of Kiev during February 17-22, 2014. We used open source and our own custom software tools to analyze the images along with upload dates and times, geo locations, and tags, and visualize them in different ways.

Full text

The Exceptional and the Everyday: 144 Hours in Kiev
Lev Manovich
The Graduate Center
City University of New York
New York, NY, U.S.A.
manovich.lev@gmail.com
Alise Tifentale
The Graduate Center
City University of New York
New York, NY, U.S.A.
atifentale@gc.cuny.edu
Mehrdad Yazdani
California Institute for
Telecommunication and Information
La Jolla, CA, U.S.A.
myazdani@ucsd.edu
Jay Chow
Web Developer
Katana
San Diego, CA, U.S.A.
jaychow0702@gmail.com
Abstract—How can we use computational analysis and
visualization of content and interactions on social media network
to write histories? Traditionally, historical timelines of social and
political upheavals give us only distant views of the events, and
singular interpretation of a person constructing the timeline.
However, using social media as our source, we can potentially
present many thousands of individual views of the events. We can
also include representation of the everyday life next to the
accounts of the exceptional events. This paper explores these
ideas using a particular case study – images shared by people in
Kiev on Instagram during 2014 Ukranian Revolution. Using
Instagram public API we collected 13208 geo-coded images
shared by 6165 Instagram users in the central part of Kiev
during February 17-22, 2014. We used open source and our own
custom software tools to analyze the images along with upload
dates and times, geo locations, and tags, and visualize them in
different ways.
Keywords—social media; photography; digital humanities,
Instagram, social movements
I. INTRODUCTION
Over a few days in February 2014, a revolution took place
in Ukraine. How was this exceptional event reflected in user-
generated media shared on Instagram in Kiev? What can visual
social media tell us about the experiences of thousands of
people and life in their city during a social upheaval?
While other researchers and journalists have used
quantitative analysis of social media during social protests and
uprisings, this is the first project to focus on Instagram. Using
Instagram public API (see Constructing the Dataset section for
details), we collected all publicly available geo-tagged
Instagram images shared during February 17-22, 2014 in the
central part of Kiev. The area was centered on Independence
Square, the key place of protests and confrontations with
government forces. Our dataset contains 13,208 images shared
by 6,165 Instagram users. The images were tagged with 5,845
unique tags.
We used open source and our own custom software tools to
analyze the images along with upload dates and times, geo
locations, and tags, and visualize them in different ways. To
compare the patterns during the days of the Revolution with a
more normal period, we also downloaded over 400,000 images
shared in the same area of Kiev between February 24 and May
14, 2014.
When news outlets report on events such as uprisings,
wars or revolutions, they typically show us only the events
taking place in particular areas – demonstrations, clashes with
police, etc. – as opposed to everything else that is going on
outside these areas. This makes sense since news is there to
inform us about things that have global political and economic
importance (for example, who will govern Ukraine next and
how this will play into European and global geopolitics). As a
result, when events of this scale are taking place, in media
reports they stand in for the whole city or a country. Nothing
else is visible at these times. For example, if we are to look at
news images of Kiev published during Maidan (the name used
to refer to the oppositional movement and 2014 Revolution)
events, the whole city reduced to what was taking place on
Independence Square.
But if we consider all Instagram photos shared during the
same days in central area of Kiev that includes the Square, a
very different picture emerges. Outside of the most intense
days and the areas of protests, you would not even know that
something political was happening. People post selfies and
other photos of their lives. They dress up getting ready to go
out, and take photos of cultural events. The images of Maidan
clashes, political slogans, and burned cars and buildings appear
right next to everything else. Most people continue their lives
and post their “likes” as on any other day. The exceptional co-
exists with the everyday. We saw this in the collected images,
and this was our motivation to begin this project.
In presenting this analysis, we have no intention of
downplaying the importance of 2014 Ukrainian Revolution, the
heroism of people who made it happen, and the work of
everybody else who supported Maidan movement. Nor are we
saying that most people in Kiev who shared images on
2014 IEEE International Conference on Big Data
978-1-4799-5666-1/14/$31.00 ©2014 IEEE
77

Instagram during the Revolution had no interest in politics, just
because they posted photos of something else. Intentions and
interests of a person can’t be guessed from a single image. And
the same image can mean different things depending on what
else is next to it. What we do want to do is to explore the things
that are usually left out from the brief news reports – the
everyday, and its relationship with the exceptional, as it is
reflected and staged on Instagram.
For journalists, social media is a window into what actually
took place. Thus it becomes important to identify who posted
what and who among them was a real participant. The web
users who can be verified are treated as additional sources of
the news. While we fully respect this approach and understand
its practical usefulness, in this project we approach social
media as its own reality, separate from the “real” reality on the
ground.
13,208 geo-tagged images from one area shared on
Instagram over six days with their 21,468 tags in three
languages (and hundreds of thousands of words used in
descriptions) paint their own fascinating picture. This picture is
not a “photo” of social reality. Instead, it can be compared to a
modern painting. It contains some references to the world
outside, but it is not its realistic copy. Our goal was to “see”
this picture dispersed between all images, tags, time stamps,
and geo-locations. To do this, we explored all this data in as
many ways as we could and then selected what we feel were
most interesting “views” of this picture.
This paper continues our previous investigations where we
visualize and interpret patterns in visual social media such as
2.3 million of Instagram images shared in 13 global cities
[1].All visualizations that appear in this paper are also available
in high resolution at the project web site http://www.the-
everyday.net/.
II. S
IX DAYS IN KIEV
To make the following discussion understandable, we
first briefly summarize the events of the 2014 Ukrainian
Revolution:
TABLE I.
Date Summary of the events
2/17/2014 The night before the 18th, oppositional parties called for
all concerned citizens to take a part in "peace offensive."
2/18/2014 In the morning, tens of thousands of demonstrators were
attacked by police. After a day of fighting in a few areas
in the city, police launched the attack on Independence
Square at 8pm. But by the midnight, 20,000 people still
remained on the square.
2/19/2014 Government closed the metro and blocked main roads.
According to one report, 30,000 were now at the square,
preparing for more confrontations.
2/20/2014 Another day of fighting between the protesters and the
police and Berkut (special government forces). During
these days, 103 protesters and 13 police were killed.
2/21/2014 An agreement between the protesters and the president
calling for constitutional reform and new elections was
reached. Soon thereafter President Yanukovych and most
of his ministers fled the city.
2/22/2014 Former President Yulia Tymoshenko was released from
prison and she addressed over 100,000 people on the
square. By February 23, transition towards new temporary
government was underway.
a.
This brief summary was compiled from: http://en.wikipedia.org/wiki/2014_Ukrainian_revolution;
http://en.wikipedia.org/wiki/List_of_people_killed_during_Euromaidan. The numbers of people at
Maidan Square are estimates by different journalists and agencies.
III. WHY INSTAGRAM?
The word euromaidan (the name for the popular movement
in Ukraine which led to the 2014 Revolution) was first used as
a hashtag on Twitter [2]. Protesters and their supporters
actively employed Twitter and Facebook to organize the
gatherings and demonstrations and communicate news to the
outside world. The protesters and other opposition parties made
most political announcements on Facebook, which had
approximately 3 million users in Ukraine in February 2014 [3].
“EuroMaydan” page became the most-liked Ukrainian page on
Facebook (with 304,590 likes as of 10/04/2014). Even more
people in Ukraine were using VKontakte - the largest social
network in Europe with English, Russian and Ukrainian as
official languages.
In comparison, Instagram had significantly less users in
Ukraine. According to the current Alexa.com data, VKontakte
is third top web site in Ukraine, right after Google’s
international page and Ukrainian page. Facebook is no. 6,
Twitter is no. 2, and Instagram is no. 56 [4]. However, despite
its smaller size in Ukraine, Instagram gives us a unique view
into the events in Kiev during 2014 Ukrainian revolution. As a
global network organized around photography, it offers a
different picture than other social networks: a visual account of
the life in a city, the desires and imaginations of its people (or
at least, people in their 20s), and their actions and thoughts
during important social and political events. (The data on
demographics of Instagram users around the world is not
available. Our research on Instagram selfies in New York,
Berlin, Sao Paulo, Bangkok and Moscow showed that at least
among people posting selfies in these cities, the largest
numbers were in their mid-20s [5].)
Examining images and data we collected, it appears that
unlike Facebook and Twitter, Instagram was not used
systematically for communication by protesters, oppositional
parties or the government. Our image set is not dominated by a
few power users posting disproportional numbers of images.
And the images themselves are quite varied - we do not see a
few images repeating endlessly. Therefore, although we will
never claim that Instagram picture of the Ukrainian Revolution
days is “objective,” it is at least representative of interests and
experiences of significant numbers of people. Thus, its
relatively small user base in Ukraine in February 2014 makes it
more (as opposed to less) useful for research.
IV. C
ONSTRUCTING THE DATASET
Using Instagram API, we downloaded all available geo-
tagged Instagram images and video publically shared in the
central area of Kiev between 02/02/2014 and 05/15/2014. The
collection area is a rectangle centered on Independence Square:
3.9 miles by 6.2 miles (6.3 km by 9.9 km). The dataset contains
463,989 media files (%3.6 are video and the rest are still
images).
78

We focused our analysis on the days of 2014 Ukrainian
revolution: February 18-21. In order to better understand
Instagram patterns during these exceptional days, we also
included one day before and one day after. Consequently, our
final period for the project is February 17-February 22, 2014.
During this six-day period, 6,165 Instagram users shared
13,208 images in the central part of Kiev. This media has 5,845
unique tags; the total number of tags is 21,468.
V. CONSTRUCTING
THE PICTURE
What can we see in the world if we only use social media
content such as Instagram photos and their metadata
(descriptions, tags, locations)? Analysis and visualization of
large samples of social media can provide an alternative to
summaries of the events presented by historians, individual
journalists, or groups of writers (e.g., collaboratively authored
Wikipedia articles). This is especially true for the visual
summaries of the events. Instead of only a few views we can
now have thousands or event millions of separate views.
Of course, often these are only fragments and bits too short
to articulate a full statement - but often they are not. And with
images, the results can be particularly interesting, since even a
single image can contain much more information than many
Twitter posts put together.
While media outlets also personalize their reports by
interviewing some of the participants and then including parts
from these interviews into the reports, this is not the same
thing. The diversity of perspectives by tens of thousands of
participants can be much larger than that of only a few who
were interviewed.
Combination of computer data analysis and visualization
can help us to juxtapose these perspectives. We can find
commonalities and differences, and discover typical as well as
unique perspectives. But we have to remember that as any
other visual media, data visualization is not neutral. By
organizing images and data in particular way, we can tease this
or that pattern. Some patterns may be given too much attention,
while others may remain hidden.
For example, organizing all 13,208 images shared by 6,165
Instagram users strictly by their date and time creates
visualization where the exceptional and the everyday are
dramatically juxtaposed (figures 1 and 2). At the same moment
as one person shares a photo of the demonstrations, another
person is posting her portrait, and yet another is posting a
photo from a party the night before. Such a “film” created by
projecting thousands of images taken by thousands of people
over a large city area onto a single linear time dimension
creates a picture of extreme fragmentation which is perhaps
even more intense than the modernist collages of cities created
hundred years earlier. But it is important to remember that this
particular picture is not “native” or “natural” to Instagram in
general, or the use of Instagram by people in Kiev during
Maidan effect. It is the result of our systematic decision to
organize Instagram images shared by thousands of individuals
in a particular way for the visualization.
VI. THE
EXCEPTIONAL AND THE EVERYDAY
Having discussed our motivations and goals, the
construction of the dataset, and the construction of its “views”
though computer analysis and visualization, we will now
present some of these views and their interpretations. They are
selected from hundreds of graphs and visualizations we
generated while working with the data.
A. Flow
We start with a visualization of all 13,208 images shared
by 6,165 Instagram (Fig. 1).
Fig. 1. A visualization showing 13,208 images shared on Instagram central
part of Kiev during February 17-22, 2014. The images are organized by
shared date/time (top to bottom, left to right).
Six light-to-dark “waves” correspond to the six days
(lighter images during the day, darker images at night). From
this bird’s eye view, we don’t see any obvious reflections of
the exceptional events that took place during this period. It
seems as though the Revolution never took place.
Fig. 2 is the close-up view of the visualization, with the
addition of time stamp right above each image. We also added
a dark background and a space around every image. Each row
shows all photos shared within a short period during the night
and morning of February 18. Looking at this scale, we now see
photos of the events at Independence Square (fires and crowds
of people in first and second row) next to the photos showing
other subjects.
Fig. 2. The close-up view of the visualization in Fig. 1, with the addition of
time stamp right above each image.
B. Time
Now we will visualize the data as a graph, plotting the
number of shared images over time (Fig 3). Grey part of the
graph shows all shared images between February 17 and 22.
Red part corresponds to images shared only around
Independence Square (1,900 images, or %14 of the total).
79

Fig. 3. The numbers of Instagram images shared in th
e
during February 17-22 (left to right). Each “mountain” c
o
hour period. Grey: all shared images. Red: im
a
Independence Square.
If we look at the grey part of the graph,
quite similar – few images in early morni
afternoon, and then gradual decrease in the
e
day in our period (February 22) is somewhat
d
because it is Saturday, so people are waking
u
out later as well. Again, looking only at t
h
shared images, you would not know that
a
place.
Filtering the data to select only image
Independence Square (red area) tells a differe
n
almost nothing on the first day before the e
v
The next day there are confrontations betwe
and government forces in a few areas (2/18
)
b
ig attack on Independence Square at 8 pm,
for hours. In the graph, we see a big jump alr
e
The next day there is no fighting on the
s
people are arriving on the square in prepar
a
fight, and many also coming to see and take
p
phones (2/19). This high level of activity con
t
of our selected period.
C. Space
Having separated time into two streams –
e
in the central part of Kiev and the part of the
d
images shared around the area of Independe
n
made the exceptional clearly visible. We
a
temporal shape (red part in Fig. 3).
We will now switch from time to spa
c
locations of images. Figure 4 shows two
displays locations of images shared durin
g
confrontations start (February 17). The secon
d
of images shared during next five days, i.e.
(Note that since our collection was limite
d
centered on Independence Square, there are
n
of this rectangle).
e
central part of Kiev
o
rresponds to one 24-
a
ges shared around
all the days are
ng, more in the
e
vening. The last
d
ifferent probably
u
p later and going
h
e volume of all
a
revolution took
s shared around
nt story. There is
v
ents start (2/17).
en the protestors
)
. Police stages a
and it continues
e
ady around 6pm.
s
quare, but more
a
tion for the next
p
hotos with their
t
inues for the rest
e
verything shared
d
ataset containing
n
ce Square – we
a
lso revealed its
c
e, and map the
maps. The first
g
the day before
d
shows locations
February 18-22.
d
to a rectangle
n
o points outside
Fig. 4. The locations of images shared
February 18-22 (bottom).
There is no spike in activity on
I
17
th
, but over next five days the ex
c
p
lace around the square, and th
e
afterwards – clearly stand out.
We can “slice” these map
a
(latitude or longitude). The graphs
shared images (vertical axis) a
g
(horizontal axis) for February 17
story told by these maps and slices
i
time plot in Fig 3. –
b
ut now the
e
dramatically. On the top graph
Independence is just a part of over
a
center. But during the next five da
y
all other areas (the bottom graph).
during February 17 (top) and
I
ndependence Square on the
c
eptional – the fights taking
e
massive meetings there
a
cross a single coordinate
in fig. 5 show volume of
g
ainst latitude coordinate
and February 18-22. The
i
s similar to wha
t
we saw in
e
xceptional stands out more
in Fig. 6 the area of
a
ll busy activity in the city
y
s it completely dominates
80

Fig. 5. Volumes of shared images (Y) against their latitude coordinate (X).
Top: February 17. Bottom: February 18-22.
D. What is #euromaidan?
Our images have seven different Maidan tags, spelling the
word differently in English, Russian and Ukrainian. These tags
are: #майдан, #maidan, #euromaidan, #євромайдан,
#евромайдан, #euromaydan, #Euromaidan. Overall, 1,340
images have at least one of these tags (%10 of the total).
Figures 6 and 7 show the visualization of these 1,340 images
organized by upload date and time (left to right, top to bottom).
Fig. 6. 1340 images which have Maidan tags organized by time (left to right,
top to bottom).
Zooming into this visualization we find that while most of
the images with Maidan tags are indeed directly related to
Maidan events, some are not (fig. 7). (We also visualized the
images that do not have any Maidan tags, and discovered that
occasionally they show the events on the square. And when we
visualized the images shared around the square, we found that
some of them have no obvious relation to Maidan events.) So
we cannot exclusively rely on tags to predict the subjects of
images. For a computer scientist concerned with detecting
social upheavals in social media, this finding would indicate a
problem that needs to be solved. Such images may be
considered as “noise,” only to be removed from the “signal.”
But for us, they are the real “finding.” They show that the
everyday and the exceptional do not simply “co-exists” side-
by-side (as presented in Figures 1-2). Instead, they “leak” into
each other, so to speak.
As we discovered, this “leakage” has its own patterns. Out
of 1,340 images that have one or more Maidan tags, %30 also
have some other tag(s). Images with only Maidan tags typically
indeed refer to Maidan events. But if it also has other tags,
there is a chance that it shows another subject (such as the
selfie in upper left corner of the close-up in Fig. 7).
Fig. 7. A close-up of the visualization in Fig. 6.
81

E. “Speaking the images”
When a person assigns multiple tags to a single image, this
image literally multiplies - because it will now show up in
searches for any of these tags. For example, let’s say an image
has #euromaidan tag and also #майдан tag (the word
“maidan” spelled in Russian). If you use Instagram to search
for these tags, this image will show up in both results.
When a person applies multiple Maidan tag versions to the
same image, every tag potentially says something else.
“Maidan” spelled in English, in Ukrainian, or in Russian is not
the same thing. So the reason behind using more than one tag
for the same word is not only to have wider dissemination of
images, or to address different linguistic communities.
Assigning a new tag to an image is like saying something else
about this image.
Our next visualization (Fig. 7) is based on this idea. Like
the visualization in Fig 6, it shows all images that have one or
more Maidan tags, but now every image is repeated for each of
its Maidan tags. For example, if an image has #euromaidan
and #майдан tags, it is repeated twice. As a result, 1,340
images turn into 2,917.
Figure 8. 1340 unique images with Maidan tags. Each image is repeated for
each of its tags. The images are organized by date and time (left to right, top
to bottom).
Figure 9. A close-up of the visualization in Fig. 8.
F. Clusters
The words in the tags and the number of times the tags with
with the same concept are applied to the same images often can
tells what images represent, but they also can be misleading.
But even if the tags and image content match perfectly, they
usually can’t completely describe everything an image
represents, and how it represents it (composition, focus, etc.). If
it is true that “an image equals a thousand words,” we certainly
are out of luck - no single Instagram image in our dataset has
as many tags.
Let’s take an Instagram photo of a person as an example.
The tags can tell us about the gender of a person (#girl, #guy),
her or his mood (#happy), and perhaps location (#beach) - but
not about the composition, or colors, or all other visual
dimensions.
For our next stage of the analysis, we will bypass language.
Instead, we will use computer algorithms to automatically
separate images into groups based only on visual similarity.
Each group contains some images that have something in
common. We use digital image analysis to extract visual
characteristics of images, and disregard tags, locations and date
and time information.
Using cluster analysis, we divided all images into 60
clusters. A few of these clusters consist from mostly Maidan
related images; most others are not. These latter reveal the
types of “everyday” in Kiev as it presented on Instagram:
double portraits, objects against a white background, city views
with light sky and darker lower parts, etc.
Fig. 10 show three examples of our clusters. The first
cluster contains mostly city views with light sky and darker
lower part. The images in the second cluster show a single
subject in the center framed by a light background. These
examples illustrate how a large image collection may contain
groups of visually similar images that are not visible through
direct examination of a collection. While we can notice
separate instances of types of images shown in these clusters, it
is much harder to see how many such instances a collection
contains.
Clustering image by their visual characteristics allows us to
understand how representations of the exeptional and the
everydays on Instagram are related in yet another way. Each
cluster shown in Fig. 10 contains mostly images of the
everyday. But because of the similary in composition, the
clusters also “catch” a few Maidan related images. In the first
cluster, these are the landscapes manipulated to only contain
blue and yellow colors of Ukraine flag. In the second cluster,
this is a political text situated in front of a candle (second to
last row, first and second image from the left).
82

Fig. 9. Two among 60 clusters identified by cluster analysis of our image set.
G. Language
Instagram users assigned 5845 unique tags to images they
shared in the larger central area of KIev during February 17-22.
If we analyze these tags, would this tell us the same story as the
previous graphs and visualizations, or a different one?
Fig. 10 shows top 10 tags for each of the six days, sorted
from top to bottom in order of their frequency.
Fig. 10. Top ten tags for February 17-22. We used blue for the geography
tags, grey for the “everyday” tags, and “red” for the tags related to the
revoltuon (Maidan tags variations and #revolution).
On the first day (February 17) before confrontations start,
we see typical Instagram pattern common to numerous places
around the world. The geographic tags which identify the
location appear first, followed by “universal” Instagram tags
not specific to any location: #love, #follow, followme,
#instagood, #me, #photooftheday. These tags are among top 15
tags on Instagram around the world every day (To see the top
100 Instagram tags on any given day, consult
http://websta.me/hot/
.)
As confrontations begin, the words “EuroMaidan,”
“Maidan” and “revolution” immediately jump to the top. On
February 18, only two universal Instagram tags still make it to
the bottom of the top ten ranking (#love and #followme).
February 19 is truly exceptional: no generic Instagram tags are
in the list. And on February 20 and 21, only single generic tag
#love appears in the bottom. Finally, on the 22
nd
, as the
fighting is over and the revolution has succeeded, two generic
tags slip back in (#followme and #instagood).
Overall, we get a perfect arc. Before the exeptional events,
“the Instagram everyday” dominates the list (February 17
th
).
then it disappears completely (19
th
), and after that gradually
starts coming back (20
th
-22
nd
). The exceptional local events
push out the universal everyday - but only for a short time.
The pattern is remarkably clear, but it raises a question. The
proportion of images shared around Indepedence Square is
only %14 of all 13,208 images shared in the larger central part
of Kiev. But when we consider the tags of all these images,
Maidan tags dominate during the revolution days. Why? One
reasonable explanation is that people who did not come to
Maidan were still concerned with the events, and they used
Maidan tags to show their support of (or other attitudes
towards) the revolution.
VII. C
ONCLUSION
Humans are always looking for signals standing out against
noise. But as modern society developed techniques to generate
progressively more data, this became particularly important.
83

Modern “news” pick out what is important for us to know.
Claude Shannon defined information as the amount of
unpredictability in a message. Flickr pioneered the use
“interestingness” to filter the photos [6]. Google Analytics
“monitors your website's traffic to detect significant statistical
variations, and then automatically generates alerts,
or Intelligence Events, when those variations occur” [7]. These
are only two examples of the key technology of “data society”
- data mining – that uses automatic computational techniques
“with the intention of uncovering hidden patterns in large data
sets” [8].
The metaphor of mining suggests that you recover what is
valuable and discard the rest. But what if we reverse the
procedure?
In his 1973 text Species of Spaces and Other Pieces,
French writer Georges Perec noted how the news only talk
about the exceptional, but never the everyday: “What speaks
to us, seemingly, is always the big event, the untoward, the
extra-ordinary: the front-page splash, the banner headlines.
Railway trains only begin to exist when they are derailed, and
the more passengers that are killed, the more the trains exist…
How should we take account of, question, describe what
happens every day and recurs everyday: the banal, the
quotidian, the obvious, the common, the ordinary, the infra-
ordinary, the background noise, the habitual?” [9]
During a weekend in October 1974, Perec set out to realize
his idea of systematically capturing infra-ordinary. Over three
days, he described what he saw from a window of a café in
Place Saint-Sulpice: buses, cars, people passing by, ordering
coffee, the pigeons, and so on. The results were published as a
short book An Attempt to Exhaust a Place in Paris.
Similar to Perec, in our project we use a rectangle as a
frame, and only take into account what this frame captures. For
Perec, it was a window in a café; for us, it’s the rectangle on
the map defined by longitude and latitude coordinates passed to
Instagram API. But since we substituted a single human point
of view by the social media network, we can stretch our frame,
to capture over much larger area. And this what we did in our
project. This allowed us to observe both infra-ordinary and the
extra-ordinary, and reconstruct some of the ways in which they
interact.
VIII. CONCLUSION
“What can we see in the world if we only use social media
content such as Instagram photos and their metadata?” In this
paper we investigated this question using the images shared
during 2014 Ukrainian revolution as our case study. In the
news reports of this exceptional event, the whole city was
reduced to a single square where the confrontations before the
protesters and goverment were taken place. Instagram gives us
a very different picture. The images of the Revolution appear
right next to many other subjects. Using a number of
visualization techniques and different parts of the data (images,
tags, dates, etc.), we explored the various ways in which the
representations of the exceptional and the ordinery interact
with each other.
IX. A
PPENDIX: CLUSTER ANALYSIS DETAILS
To find clusters of similar images, we use the k-means
clustering algorithm provided by the R statistical programming
environment. All images are 150-by-150 pixels (the standard
Instagram thumbnail dimension). We extract images features in
Python using the scikit-image library [10]. From this library,
we use the Histogram of Oriented Gradients (HoG) features
[11] with 8 gradient orientations and non-overlapping 16-by-16
window cells. In addition, we also use the raw grayscale pixel
values of the images as features, as well as the latitude and
longitude of image locations. Since these feature have different
scales and units, we re-scale them to have zero mean and unity
standard deviation.
In k-means clustering, the number of clusters must be
specified a priori. We experimented with several values for k,
ranging from 10 to 100. For each value of k, we use 10 random
starts to ensure that we find clusters that converge to the same
cluster centers. When the number of clusters is large, there are
redundant clusters that can be merged based on our evaluation
of similarity. To evaluate the found clusters, we use the
ImageMontage [12] tool to visualize all images in each cluster
(see examples in Fig. 10). The primary advantage of using a
large number of clusters is that non-linear cluster boundaries
can be approximated with k-means. After a number of
experiments, we decided that k=60 provides best results, and
this is the number we used in this paper.
R
EFERENCES
[1] N. Hochman and L. Manovich, “Zooming into an Instagram city:
Reading the local through social media.” First Monday, 6/1/2013.
[2] http://en.wikipedia.org/wiki/Euromaidan#Name_history
[3] J. Dickinson, “The Revolution Will Be Live-streamed: Social Media and
the Maidan,” presented at SOYUZ Symposium, Havighurst Center,
3/1/2014. Unpublished.
[4] http://www.alexa.com/topsites/countries/UA, retrieved 10/04/2014.
[5] L. Manovich, M. Stefaner, M. Yazdani. D. Baur, D. Goddemeyer, A.
Tifentale, N. Hochman, J. Chow. “Selfiecity.” New York, February
2014. www.selfiecity.net
[6] Flickr, Interestingness Ranking of Media Objects, U.S. Patent
#US008732175, Filed February 8, 2006.
http://pdfpiw.uspto.gov/.piw?PageNum=0&docid=08732175&IDKey=B
8DCE9513A3F
[7] https://support.google.com/analytics/answer/1320491?hl=en&ref_topic=
1032994, accessed 11/8/ 2014
[8] http://en.wikipedia.org/wiki/Data_mining
[9] G. Perec, Species of Spaces and Other Pieces, transl. J. Sturrock.
London; New York: Penguin Books, 1997. First published in 1973 in
Cause Commune.
[10] van der Walt S, Schönberger JL, Nunez-Iglesias J, Boulogne F, Warner
JD, Yager N, Gouillart E, Yu T, the scikit-image contributors, “scikit-
image: image processing in Python,” PeerJ 2:e453
http://dx.doi.org/10.7717/peerj.453
, 2014.
[11] N. Dalal B. Triggs, B., “Histograms of Oriented Gradients for Human
Detection,” IEEE Computer Society Conference on Computer Vision
and Pattern Recognition, 2005, San Diego, CA, USA.
[12] https://github.com/culturevis/imagemontage
84