Content uploaded by Sofiane Abbar
Author content
All content in this area was uploaded by Sofiane Abbar on Nov 29, 2016
Content may be subject to copyright.
Fetishizing Food in Digital Age: #foodporn Around the World
Yelena Mejova and Sofiane Abbar
Qatar Computing Research Institute, Qatar
{ymejova,sabbar}@qf.org.qa
Hamed Haddadi
Queen Mary University of London, UK
hamed.haddadi@qmul.ac.uk
Abstract
What food is so good as to be considered pornographic?
Worldwide, the popular #foodporn hashtag has been used to
share appetizing pictures of peoples’ favorite culinary experi-
ences. But social scientists ask whether #foodporn promotes
an unhealthy relationship with food, as pornography would
contribute to an unrealistic view of sexuality (Rousseau
2014). In this study, we examine nearly 10 million Insta-
gram posts by 1.7 million users worldwide. An overwhelm-
ing (and uniform across the nations) obsession with chocolate
and cake shows the domination of sugary dessert over local
cuisines. Yet, we find encouraging traits in the association of
emotion and health-related topics with #foodporn, suggesting
food can serve as motivation for a healthy lifestyle. Social ap-
proval also favors the healthy posts, with users posting with
healthy hashtags having an average of 1,000 more followers
than those with unhealthy ones. Finally, we perform a demo-
graphic analysis which shows nation-wide trends of behavior,
such as a strong relationship (r= 0.51) between the GDP per
capita and the attention to healthiness of their favorite food.
Our results expose a new facet of food “pornography”, re-
vealing potential avenues for utilizing this precarious notion
for promoting healthy lifestyles.
Introduction
Gastro-porn was first coined by Alexander Cockburn in
1977 in a review of a cookbook: “True gastro-porn height-
ens the excitement and also the sense of the unattainable
by proffering colored photographs of various completed
recipes” (Cockburn 1977). Since then, the rise of diets and
fitness in the 80s, complemented by the rise in obesity rates
and eating disorders, has sustained the development of food-
related media (O’Neill 2003). With such a rich history, it is
no surprise the hashtag #foodporn is one of the most pop-
ular hashtags on social media, especially visual media such
as Instagram, often accompanied with a close-up of a tan-
talizing dish. In the age of social media, its users are now
defining what food “pornography” is to them.
#Foodporn hashtag is part of a vast lifestyle tracking
trend. “I Ate This” Flickr group, one of the largest and most
active on the site, hosts over 640K photos that have been
contributed by over 40K members. The practice is so perva-
Copyright c
2016, Association for the Advancement of Artificial
Intelligence (www.aaai.org). All rights reserved.
sive, manufacturers have released cameras with specific food
mode1, emphasizing the sharpness and saturation of colors.
Such daily food diaries provide an invaluable resource
for food culturalists and public health professionals. In par-
ticular, the public sharing of food “pornography” contextu-
alizes the favorite foods of social media users around the
world. As emotional state (Canetti, Bachar, and Berry 2002)
and social interaction (Christakis and Fowler 2007) have
been associated with the healthy weight maintenance, ob-
taining daily reflections on dietary experience becomes im-
perative for a holistic view of an individual’s relationship
with food. The case of #foodporn is especially interesting,
given the possibly negative connotations it introduces, per-
chance promoting an unhealthy relationship with food, as
pornography would contribute to an unrealistic view of sex-
uality (Rousseau 2014). In the United States, for example,
the most liked Instagram posts are from donut and cupcake
shops (Mejova et al. 2015). In this work, we seek to find
whether such attitudes are common across the world.
Instagram is a forum highly suitable for such food log
research. We find that a staggering 46% of the nearly 10
million Instagram posts mentioning #foodporn we collected
were geo-located – a proportion vastly outnumbering sim-
ilar collection of Twitter posts (which had only 5.8% with
available geo-location data). The 72 countries we examine
prove to have a wide range of integration with international
cuisine. Chocolate, cake, and other heavy foods dominate
the #foodporn conversation, with some nations sporting their
own unhealthy trends with #gordice (in Brazil) and #gour-
mandise (in France), meaning “fat” and “gluttony”, respec-
tively. However, we hesitate equating the #foodporn trend to
the effects of non-food pornography. Among the most pop-
ular foods is salad, followed by potentially healthy alterna-
tives of sushi and chicken. Furthermore, the social approval
(in terms of likes and comments), as well as user following is
the highest for hashtags associated with the healthy lifestyle.
Thus, the impact of the study presented below is two-fold:
an introduction of a fertile dataset for health research, and
a case study of a concept’s re-definition, as it is owned and
explored by the global social media community.
1http://www.cnet.com/news/what-are- all-
those-camera- modes-for- anyway/
arXiv:1603.00229v2 [cs.SI] 2 Mar 2016
Related Works
Use of social media for monitoring public health has been
increasingly popular in the last few years (for a system-
atic review, see (Capurro et al. 2014)). Popular Online So-
cial Networks (OSNs) such as Twitter, Instagram, and Face-
book, with over a billion users, have become a rich source
for social scientists, psychologists, and health professionals.
A range of behavioral health issues have been studied us-
ing OSNs, including depression and mental health (Park et
al. 2015), obesity and diabetes (Abbar, Mejova, and Weber
2015), tobacco use (Prier et al. 2011), and insomnia (Paul
and Dredze 2011). Complementary to physical activity data
and Electronic Health Records (EHRs), social media data
provides a record of daily social engagement necessary for a
holistic view on individuals’ health (Haddadi et al. 2015).
Completing the information cycle, social media has also
been illustrated to be a valuable tool for wellness and health
promotion (Neiger et al. 2012a).
Despite being in use for over 5 years, Instagram has re-
ceived little attention from the research community. With a
rich source of data around the picture posts, including social
network of the users, a folksonomy of hashtags (Yamasaki,
Sano, and Mei 2015), and an association with a hierarchy
of locations, it provides a detailed view of individuals’ daily
lives and habitual activities. Hu et al. (Silva et al. 2013)
present a first study of users’ posting behavior on Instagram,
showing that food pictures are a major feature of Instagram
posts. In recent work, we associate health-related activity
captured on Instagram to regional obesity, diabetes and other
census statistics in the US (Mejova et al. 2015), finding dis-
tinctions between the behavior of users who are more, or
less, likely to be healthy.
Dietary behavior research calls for a more global view,
since much of diet-related behavior is cultural (Counihan
and Van Esterik 2013). As Fischler (Fischler 1988) puts it,
by selecting and cooking food, one “transfers nutritional raw
materials from the state of Nature to the state of Culture”.
Culture-specific ingredient connections have been discov-
ered by Ahn et al. (Ahn et al. 2011) who mine recipes to
create a “flavor network”. Temporal nature of food con-
sumption has been explored by West et al. (West, White,
and Horvitz 2013), who mine logs of recipe-related queries.
They illustrate the yearly and weekly periodicity in food
density of the accessed recipes, with different trends in
Southern and Northern hemispheres, suggesting a link be-
tween food selection and climate. In this research, we quan-
tify the context around the favorite foods of nations around
the world in terms of perceived healthiness or unhealthiness,
social situations, and sentiment. We further correlate these
behaviors with nation-wide demographics.
The wider use of tantalizing food imagery has been stud-
ied by social scientists, often drawing parallels between
“food porn” and the non-food pornography. Anne McBride
recently asked academics and chefs about the term, finding
that many associate it with unrealistic, “sexy” photographs
of food, mostly used for advertisement (McBride 2010). The
word itself is meant to attract attention, but may have other
connotations, such as it being “indecent [...] when there is
so much hunger in the world”. Our research validates the
0 5000 15000 25000
# of posts
2015−03−24 00
2015−03−24 12
2015−03−25 00
2015−03−25 12
2015−03−26 00
2015−03−26 12
2015−03−27 00
2015−03−27 12
2015−03−28 00
2015−03−28 12
2015−03−29 00
2015−03−29 12
2015−03−30 00
2015−03−30 12
2015−03−31 00
2015−03−31 12
2015−04−01 00
2015−04−01 12
2015−04−02 00
2015−04−02 12
2015−04−03 00
2015−04−03 12
Instagood Food Foodporn
Figure 1: Volume of Instagram posts mentioning #insta-
good, #food, and #foodporn hashtags.
opinions of these experts, in finding the attitudes every-day
social media users associate with this term.
Data
Our main source of data is Instagram, a social platform for
sharing photo and video content via mobile devices (by de-
sign containing more geo-centric data compared to desktop-
oriented social sites). Using the Tags endpoint of Instagram
API, we collected all posts containing the keyword #food-
porn, resulting in a collection spanning 6 Nov 2014 – 6 Apr
2015 containing 9,378,193 posts from all over the world.
To put it in context of other popular streams, we collected
two additional datasets for a smaller overlapping time in-
terval, one to compare to a general food-related conversa-
tion, with hashtag #food (consisting of 1,460,226 posts), and
similarly to general discussion on Instagram not necessarily
about food, with hashtag #instagood (3,368,416 posts). Fig-
ure 1 shows the volume of the three streams, which decrease
with more topical specificity.
The posts both in large #foodporn dataset and smaller
ones were then geolocated using the World Borders shape
file2by converting the (latitude, longitude) pair to the coun-
try code from which the post is published. It is worth notice
that the geolocation module returns “None” when the post
does not contain geo-coordinates or when they fall close
to some disputed/unclear borders. The results of the geo-
location can be seen in Table 1. The proportion of success-
fully located posts increases from #instagood at 28.6%, to
#food at 31.6%, and to #foodporn at 42.8%. We observe
that not only do users geo-tag food-related posts more than
generic ones, but especially so the food they particularly en-
joy.
As Twitter is another popular micro-posting platform, we
have collected a similar dataset of #foodporn mentions span-
ning 10 days 02 Jun 2015 – 12 Jun 2015. Similarely to Insta-
gram posts, we used our geolocation module to map tweets
into countries (see Figure 2 for volume statistics). However,
Twitter provided much fewer geo-located posts, with only
2http://thematicmapping.org/downloads/world borders.php
Table 1: Dataset statistics for Instagram datasets for the overlapping period (25 Mar 2015 – 03 Apr 2015).
Hashtag posts users avg posts/user % geotagged posts unique locations
#instagood 3,368,416 920,495 3.65 28.58% 915,122
#food 1,460,226 619,340 2.35 31.56% 696,640
#foodporn 675,145 322,939 2.09 42.77% 361,653
0 1000 3000 5000
# of posts
2015−06−03 00
2015−06−03 12
2015−06−04 00
2015−06−04 12
2015−06−05 00
2015−06−05 12
2015−06−06 00
2015−06−06 12
2015−06−07 00
2015−06−07 12
2015−06−08 00
2015−06−08 12
2015−06−09 00
2015−06−09 12
2015−06−10 00
2015−06−10 12
2015−06−11 00
2015−06−11 12
2015−06−12 00
2015−06−12 12
Instagram Twitter
Figure 2: Volume of Instagram and Twitter posts mentioning
#foodporn.
5.76%, compared to Instagram’s 46.2% in the same time
span.
The above statistics show #foodporn Instagram collection
to be unique in both the volume (compared to Twitter) and
the proportion of geo-located data (compared to, say, #insta-
good). Thus, in this study we use the complete #foodporn
Instagram dataset encompassing 5 months, and containing
9.3 million posts.
This data spans 222 countries, with a characteristic long
tail distribution. Out of these, we select 72 countries having
at least 500 unique users in our dataset for further examina-
tion. At their head are the United States, Italy, and United
Kingdom, with the tail including Malta, Iran, and Pakistan.
Prominence and Use
The reach of #foodporn hashtag across the globe is diffi-
cult to understate. In the 150 days of our dataset, over 1.7
million individual users tagged their posts with it, making
the average rate at 62K posts per day. As with most so-
cial media, United States dominates the conversation, with
Italy being the second-highest user. Figure 3 plots the in-
ternet penetration-adjusted country population statistics3to
the unique number of users posting at least once with #food-
porn from that country, along with a linear regression line
(curved, due to log/log axes). We find China (CN) to be
an outlier, with disproportionately few users in our dataset
compared to its population. Although the top right corner
is dominated by European countries, notable exception are
Australia (AU), Malaysia (MY), and Singapore (SG) – the
3Statistics for 2013 using World Development Indicators data,
as described in Demographic Correlation Section.
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
5e+05 2e+06 5e+06 2e+07 5e+07 2e+08 5e+08
2e+02 5e+02 1e+03 2e+03 5e+03 1e+04 2e+04 5e+04 1e+05
Population x internet penetration
#foodporn users
AE
AR
AT
AU
BE
BR
BY
CA
CH
CL CN
CO
CR
CY
CZ
DE
DK
DO
EC EG
ES
FI
FR
GB
GR
HK
HR
HU
ID
IE
IL
IN
IR
IT
JP
KH
KR
LB
LK
LT
MA
MT
MX
MY
NL
NO
NZ
PA
PE
PH
PK
PL
PR
PT
QA
RO
RS
RU
SA
SE
SG
SI
SK
SV
TH TR
UA
US
VE
VN
ZA
Figure 3: Country population versus the unique number of
users posting at least once with #foodporn hashtag.
latter especially unusual considering its smaller population.
As we show later, Asian countries have by far the largest
proportion of users dedicated to tracking their #foodporn ex-
periences.
Next, we focus on the users. We query the Users end-
point of Instagram API to request the profiles of 436,472
(out of 1.7Mtotal users) having at least one geo-tagged post
with #foodporn hashtag. Each user profile comes with ba-
sic information such as id,username,full name,bio, and
the numbers of total shared media, and their followers and
friends. We compute then for each user the set of countries
associated with all her posts on #foodporn. To differentiate
the users by their contribution to the dataset, we define three
categories: Singletons, Residents, and Traveleres. The first
category contains users who have posted only once in the
time frame of our study (Singletons). The second encom-
passes users who posted more than once, but whose posts are
associated with only one country (Residents). The third con-
tains users posting from several countries (Travelers). Note
that we cannot tell where a user resides permanently, thus
the above definitions are shortcuts to characterize behavior
as captured through the use of #foodporn hashtag. A more
thorough examination of user’s activity and content is left
for future work.
Table 2 provides summary statistics of the three user
groups. While Travelers constitute only 5.30% of the en-
tire population, they contribute 15% of the posts. In fact,
a Traveler user posts in average 2.3times more than a Resi-
(a) Singletons (b) Residents (c) Travelers
Figure 4: Biographies of user groups: singletons (posting once), residents (posting several times from same country), and
travelers (posting from several countries).
Table 2: Summary statistics of user groups.
Category users % users posts % posts p/u
Singletons 122,581 28.09% 122,581 2.76% 1
Residents 290,741 66.61% 3,625,784 82.24% 12.47
Travelers 23,150 5.30% 687,390 15% 29.69
All 436,472 100% 4,435,755 100% -
dent user. Note that Singletons may also “reside” in the same
country as a Resident user, but they are not heavy users of
#foodporn hashtag.
Among the countries we consider, we find Asia to have
the greatest dominance of Residents over Singletons, that is,
its users are determined to use #foodporn habitually. These
countries include Hong Kong, Korea, Singapore, Taiwan,
Thailand, Japan, etc, and their populations have under 20%
singletons. The countries on the opposite side of the spec-
trum are Norway and Finland, who have over 40% single-
tons. In terms of proportion of travelers, Cambodia and
China stand out at 27 and 20%, respectively, suggesting their
native populations are not as involved in the conversation.
Next, we examine the profiles of users, particularly their
self-identified biographies. Figure 4 shows the most fre-
quent words mentioned by users of each category (with stop
words removed). Interestingly, we find that Singleton users’
interest is about general topics such as life and love – these
users only occasionally use #foodporn among their many
other interests. Residents and Travelers share the same top
interest of food, while differing in their top second interest
which is travel for Travelers and life for Residents.
We further find support to the fact that tourism may drive
the experiences users associate with #foodporn when we re-
late the number of Travelers in our dataset to the number of
tourists entering the country. Figure 5 shows the relation-
ship, which has strong positive correlation.
As we move forward with the content analysis, in most
cases we do not distinguish between the groups above.
First, it is impracticable (and often impossible) to locate
the “home” country of 436K users. Second, as we show in
Demographic Correlation Section, countries with expat cul-
tures provide their own characterization of local #foodporn
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
2e+05 5e+05 1e+06 2e+06 5e+06 1e+07 2e+07 5e+07 1e+08
10 20 50 100 200 500 1000 2000
Tourists per year
#foodporn travelers
AR
AT
AU
BE
BR
BY
CA
CH
CL
CN
CO
CR
CY
CZ
DE
DK
DO
EC
EG
ES
FR
GB
GR
HK
HR
HU
ID
IE
IL
IN
IR
IT
JP
KH
KR
LB
LK
LT
MA
MT
MX
MY
NL
NZ
PA
PE
PH
PL
PR
PT
QA
RO
RS
RU
SA
SE
SG
SI
SV
TH
TR
UA
US
VE
VN
ZA
Figure 5: Number of tourists entering country per year ver-
sus number of Travelers users posting from a country.
cuisine. Finally, due to sparsity, it is often impossible to
study all three cohorts for all countries. However, the inter-
action between local and outside perspective on a country’s
cuisine is an exciting future research direction.
Food Preferences
Instagram is a media (photo & video) sharing platform, but it
is the hashtags which make it searchable, and which provide
valuable annotation of the post’s content and context. Our
analysis combines qualitative and quantitative insights using
this textual information.
Global View
First, we ask, which foods do people in various countries en-
joy so much as to associate them with the #foodporn hash-
tag? We begin by computing the frequencies of all hashtags
other than #foodporn. To prevent prolific users from domi-
nating the rankings, we count the use of each hashtag only
once per user. Similarly, we aggregate hashtags per coun-
try, first normalizing the frequencies (to probabilities), such
that over-represented countries (most notably United States)
cake
salmon
coffee
eggs
pasta
cheesecake
saladfruit
bread
strawberry
icecream
chocolate
soup
sushi
wine
tea fish
fruits
nutella
cheese
banana
pancakes
vegetables pizza
steak
meat
beef
burger
chicken
bacon
Figure 6: Top foods mentioned in the dataset, normalized by
user and country.
does not dominate the others. Thus, we find at the top of
such a list the tags #food, #instafood, #yummy, #delicious,
and #foodie. Out of the meals, #dinner precedes #lunch,
suggesting #foodporn is more often experienced during the
evening meal.
In particular, we are interested in the foods mentioned in
these posts. To identify such foods, we use Crowdflower4to
label the top 100 hashtags of each country as either food or
not. To assist labelers with foreign foods, links to Google
Translate5and Google Search6were provided. Aggregating
over 3 labels for each term, 3,870 terms were labeled, out of
which 972 mentioned food (with high inter-rater agreement
of 91% label overlap).
To compile a global ranking of foods, not dominated
by any one country, we normalize hashtag frequencies per
country and then aggregate globally. The resulting top 30
foods are shown in Figure 6. Globally, we find that users are
most excited about sweets and fast food. In the 72 countries
we consider, when we look at the top 3 foods, chocolate ap-
pears in 56 (78%), and cake in 29 (41%). The top non-sweet
foods are pizza,salad,sushi, and burger – representing both
Western and Eastern cuisines. Top drink is coffee, top alco-
holic one is wine, and top fruit is strawberry. Although the
vast majority are generic ingredients and dishes, a brand ap-
pears at the 18th spot – Nutella, a hazelnut chocolate paste –
which has sold 365,000 tonnes worldwide in 20137.
As one can see from Figure 6, the terms most used along-
side #foodporn are in English (even when volume is nor-
malize per-country such that, for example, United States or
Great Britain, do not dominate the ranking). This could be
due to some fraction of our users being English-speaking
tourists (recall that “travelers” contribute 15% of the posts).
However, the fact that English remains the predominant lan-
guage throughout the nations in our dataset, once again
points to it being the lingua franca of social media (as dis-
4http://crowdflower.com/
5https://translate.google.com/
6https://www.google.com/
7http://www.bbc.com/news/magazine-27438001
Table 3: Top most distinct five foods co-occurring with
#foodporn for select countries.
cussed, for instance, in (Tagg and Seargeant 2012)). Inci-
dentally, we manually labeled the top 50 tags used by the
three groups in each country for language, and found a sur-
prisingly consistent proportion of English (at 85% for Sin-
gletons and 88% for the other two groups). This further em-
phasizes the prominence in the use of English in both heavy
and light social media users.
In order to find foods most used in each country, we com-
pute a score by subtracting the probability a food is men-
tioned in our dataset from the probability it is mentioned
in a particular country. Due to space, we direct the reader
to 8to explore the top foods of the countries. A selection
is listed in Table 3. At the top we often find national spe-
cialties – Canadian poutine, Ecuadorian ceviche, Japanese
ramen, Swiss fondue. Major diasporas can be found, such
as the Asian one in Canada (the largest minority at 15% of
population9). Yet other countries show the international na-
ture of their residents, such as Qatar and USA, which have a
wide array of cuisines. Finally, we find local alphabets and
words, although even those are often transliterated into latin
script (as in the case of Iran).
National Dietary Behavior
Next, we expand our analysis to themes possibly signify-
ing the dietary health and habits of the populations around
the world. Emotions related to food intake have long been
hypothesized to be associated with healthy weight mainte-
nance (Canetti, Bachar, and Berry 2002), and social psychol-
ogists have studied food choice, weight control, and self-
presentation (O’Connor and O’Connor 2004). Using the
textual context of Instagram posts, we attempt to quantify
emotional and social aspects of the favorite cuisines around
the world.
Hashtag Categories
We begin by extracting the top hashtags of the posts asso-
ciated with each of the 72 selected countries, and process
them similarly to the previous section. First we get the top
8http://cdb.io/1jSHJTr
9http://www12.statcan.gc.ca/nhs-enm/2011/
dp-pd/dt- td/Index-eng.cfm
1000 hashtags for each country, these are then converted to
a probability distribution (by normalizing the term frequen-
cies by their sum within the list). A similar vector is created
for all countries by summing up all of the probability vectors
and again normalizing by the sum to get a distribution of top
1000 hashtags. Then we normalize the top hashtag probabil-
ities by the vector of most popular terms, to get the adjusted
scores which would signify how prominent the terms are for
that particular country. To do this, we subtract the probabil-
ity of a term appearing in “all” vector from its probability
in the country vector. Note that here some scores may be 0,
and some may end up being negative. We then sort the terms
by this score 10. The top term is invariably either the name
of the country or its capital (or most famous, in case of New
York or Barcelona, city).
Now, we convert this qualitative data to quantitative. We
use CrowdFlower to label the hashtags into categories deal-
ing with social, health, and emotional aspects of dietary ex-
perience. We take the top 50 most distinguishing hashtags
of each country, and exclude numbers and words of 1 or 2
characters. For each hashtag, links to Google Translate and
Google Search were again provided, with instructions to se-
lect the last resort option of “Don’t understand” only after
attempting to translate the term. The following options were
provided (as determined via content coding by the authors):
•Sentiment (emotions, opinions, etc)
•Healthy (food, lifestyle, veggies, etc)
•Unhealthy (food, lifestyle, desserts, etc)
•Social (other people, events, holidays, etc)
•Location (place, city, country, etc)
•Food / drink (particular foods/drinks)
•Time / date / time of the day / mealtime
•None of the above
•Don’t understand
To ensure high quality of responses, we requested 5 dif-
ferent labels per hashtag. Twenty-three gold-standard ques-
tions were used to discard careless labelers and bots. Anno-
tator agreement, as measured in the number of exact label
overlaps, is 78.8%, which is satisfactory for our task, con-
sidering there are 9 choices, and such that several can be
selected at the same time.
The resulting 2060 labeled hash tags (available at 11) have
the distribution of labels shown in Table 4, along with some
examples. Most tags are in English, but the set also in-
cludes many other languages and alphabets, including Ko-
rean, Cyrillic, and Chinese.
National & Regional Behavior
Now that the hashtags have been categorized, the interests
expressed through them in countries and their larger regions
can now be quantified. For example, Figure 7 shows the
shares of each category in the top 50 hashtags of each part
of the world.
10the term distributions are available at http://scdev5.
qcri.org/sabbar/tags/foodporn.html
11https://tinyurl.com/foodporn- hashtags
Table 4: Categories of top distinguishing hashtags.
Category # Examples
Sentiment 115 yum, nomnom, blessed
Healthy 106 lowcarb, gym, goodeats
Unhealthy 30 fatty, cheatmeal, cake
Social 114 life, igers, girls
Location 810 italia, home, london
Food/drink 657 paella, pic food, mojito
Time/date 52 monday, evening, dinner
None of the above 176 vsco, tao, sun
Australia−N. Zealand
Caribbean
Central America
Eastern Europe
Northern Africa
Northern America
Northern Europe
South America
South−Central Asia
South−Eastern Asia
Southern Africa
Southern Europe
Western Asia
Western Europe
Eastern Asia
food/drink healthy location none sentiment social time/date unhealthy
Figure 7: Sub-regional hashtag category statistics
The most health-conscious regions are Northern and
Western Europe, as well as Australia and New Zealand.
Sentiment is most associated with #foodporn in Southern
Europe (Greece, Italy, Spain, etc.) and West Asia (Turkey
and Middle East), and social events mentioned in Northern
Africa (Egypt, Morocco), South America (Argentina, Brazil,
etc.), South-Central Asia (India, Malaysia, etc.), and South-
ern Europe.
The highest rate of unhealthy tags came from Brazil, Ar-
gentina, and France. Examples of such popular tags are
#gordice (in Brazil), which derives from “gordo” (“fat”), and
#gourmandise (in France), which translates as “gluttony”.
These three countries have 5 healthy tags in their top 50 be-
tween them.
The most healthy-conscious country is the Netherlands,
with 25 out of the top 50 hashtags dealing with fitness and
health, including #fitgirl, #fitspo, #eatclean, etc. Ireland is a
close second, with more bodybuilding-related tags, such as
#proteinpancakes, #girlswholift, and #fitness.
As mentioned earlier, some countries display a higher
population of tourists, and we find a sign of this trend in the
social tags used in the two countries found to have the most
of these: in El Salvador with #family, #friends, and #sunday-
funday and Cambodia with #wanderlust, #instatravel, and
#holiday, among others. Similarly, United Arab Emirates is
famous for Dubai and Abu Dhabi, destinations mentioned in
no fewer than 17 out of 50 tags, but due to a high number
amazing
hungry
delish
foodlover
beautiful
sweet
tasty
lecker
nomnomnom
miam
nomnom
yum
happiness
delicious
good
happy
motivation
like4like
love
likeforlike
yummy
sogood
followme
bestoftheday
treat
favorite
hot
selfmade
spicy
yummi
Figure 8: Top emotions expressed in the context of #food-
porn hashtag.
of expatriate population, these may only partially indicate
transient tourism.
Finally, we turn to the concern some social scientists
have expressed over the use of “porn” in context of food
consumption (McBride 2010). Although we find some un-
healthy associations (which are sometimes highly localized),
the emotions expressed around #foodporn are overwhelm-
ingly positive (see Figure 8). Generic emotions expressed
include #love, #sogood, #happiness, #happy, #good, etc.
Among the top 30 emotions we also see #motivation and
#selfmade, indicating association with healthy lifestyle.
Demographic Correlation
Whereas these hashtag categories provide a structured qual-
itative look at the associations with #foodporn around the
world, we ask whether there are quantitative relationships
between the attitudes we glimpse and economic and social
characteristics of each nation.
We use World Development Indicators (WDI)12 data to
enrich our understanding of these countries. The statistics
we gathered are as follows:
•gdppc - Gross Domestic Product per Capita
•gini - the Gini Index measuring income inequality
•unemployment - unemployment (% of total labor force)
•mobile - mobile phone subscribers (per 1,000 people)
•urban - urban population (% of total)
•tourists - international tourism, number of arrivals
•diabetes - diabetes prevalence (% of population ages 20 to 79)
•obesity - from World Health Organization13
Figure 9 shows the correlation of the above social, eco-
nomic, and health statistics with the hashtag use. The statis-
tics begin the triangle from the top, and the hashtag statis-
12http://wdi.worldbank.org/tables
13http://apps.who.int/gho/data/view.main.
2450A
−1 −0.9 −0.8 −0.7 −0.6 −0.5 −0.4 −0.3 −0.2 −0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
gdppc
gini
unemployment
mobile
urban
tourists
diabetes
obesity
#food/drink
#healthy
#unhealthy
#sentiment
#social
gini
unemployment
mobile
urban
tourists
diabetes
obesity
#food/drink
#healthy
#unhealthy
#sentiment
#social
#location
−0.34
−0.02
0.16
0.55
0.24
−0.09
0.34
0.24
0.51
0.19
0.2
0.22
0.4
0.02
0.11
0.1
−0.13
0.33
0.13
0.05
−0.14
−0.01
−0.04
0.01
−0.1
0.01
0.06
0.1
−0.03
0.31
−0.02
−0.09
−0.17
0.07
0.06
−0.13
0.33
−0.11
−0.01
0.4
−0.07
0.03
−0.03
0.11
0.06
−0.1
0.07
0.02
0.55
0.07
0.23
0.06
0.07
0.07
0.15
0.01
0.01
0.47
0.34
0.26
0.46
0.46
0.58
0.29
0.01
−0.1
−0.11
−0.04
−0.05
−0.08
0.11
0.27
−0.1
0.12
0.04
−0.09
0.62
0.46
0.65
0.74
0.76
0.62
0.27
0.34
0.64
0.12
0.28
0.63
0.94
0.63 0.72
Figure 9: Correlation matrix of country demographics with
hashtag categories (which are signified with #).
tics are aggregated per group and are signified with # sym-
bol. We see a few notable correlations – especially those
of GDP per capita (gdppc) with healthy (0.51) and loca-
tion (0.40) hashtags. Healthy hashtags are also slightly neg-
atively correlated with Gini index (at -0.14) such that the
more unequal the incomes are (and the higher Gini index be-
comes), the fewer healthy tags are used. Interestingly, there
is also a slight negative relationship between the use of un-
healthy tags and unemployment rate (at -0.17). Tourism, we
find, fairly strongly affects the use of all tags, but especially
the mention of locations, social situations, and sentiment, as
well as particular foods and drinks.
Surprisingly, we find a positive correlation between obe-
sity and mentions of healthy hashtags (at 0.27). This may be
due to wealthy nations both having a greater problem with
obesity, yet also enough disposable income and leisure time
to concern with healthy lifestyle (perhaps in different pop-
ulations). In order to discern the effects of different sub-
populations, a more detailed profiling, either by automatic
or standard survey techniques, is necessary.
Social Approval
Finally, we turn to the social approval and interactions
around #foodporn. Here, we examine which kinds of behav-
iors are most promoted by the Instagram community. Fig-
ure 10 shows the distribution of average number of likes
given to the posts containing a hashtag from one of the
classes. To avoid extreme outliers from misspellings and
singletons, we exclude all hashtags used by 4 or fewer
unique users in our dataset. We then average over the num-
●● ●●●● ●
●●● ●
● ● ●● ●●●● ●● ●● ●● ●●●● ●●● ●● ●●●● ●●●● ● ●●●● ●● ●●● ●
● ●● ●●● ●●●●● ●●● ●●●● ●● ●●●●● ● ●●●●
●●● ● ●●
●●● ● ●
●●● ●
●
0 50 100 150 200
Other
Time/date
Location
Food/drink
Social
Sentiment
Unhealthy
Healthy
Figure 10: Distribution of average likes for posts containing
hashtags from a certain group.
●● ● ●●
● ●●●
●● ●● ●● ●●● ●● ●● ●●● ●● ● ● ●● ●● ●● ● ●●●● ●●● ●● ● ●●●● ●●
● ● ●●● ●● ●● ● ●●● ● ●● ● ●● ●● ●● ●● ●●● ●●● ●●● ● ●●
●●
●● ●● ● ●●
●
0 2000 4000 6000 8000
Other
Time/date
Location
Food/drink
Social
Sentiment
Unhealthy
Healthy
Figure 11: Distribution of average followers of users whose
posts contain hashtags from a certain group.
ber of likes each post containing a hashtag received. Healthy
hashtags are by far the best liked, having an average of 87.6
likes (median 51), compared to 68.2 (median 35) of un-
healthy ones (similarly, 4.7 comments for healthy and 3.7
for unhealthy). In fact, the top posted and liked hashtags in
our dataset include #eatclean, #fresh and #fitness.
However, it is the individual celebrities that receive the
highest number of likes per post. At the top we find #ztf, a
hashtag referring to the fans of Zizan Razak, a Malaysian ac-
tor and comedian, with an average of 18,850 likes per post.
The most commented tag is #electrifynutrition, a supple-
ment for weight training. These large communities adopt the
#foodporn hashtag and create their own context around it,
redefining what their favorite food is (even when it is eaten
by their favorite celebrity).
Liking behavior is closely linked to the popularity of the
posting user – Pearson correlation between the likes of the
posts and the number of followers of the posting user is 0.74.
Indeed, the most popular Instagram users are those using
the healthy hashtags, at an average of 3,426 followers, com-
pared, for example, to 2,432 followers of users posting un-
healthy hashtags (see Figure 11). It is unclear whether the
prominence brings interest in health, or the health-related
content to popularity, we detect a social approval of health-
related content.
Discussion
In this work we present a global view of the #foodporn hash-
tag, as used in 72 countries around the world. We caution
the reader from making interpolations about any particular
country in this dataset. We show in Data Section that 15%
of posts are generated by users we dubbed as Travelers –
those posting in more than one country. As it was untenable
to collect the posts of all the users and attempt to estimate
their “home” country, we estimate that even more of these
users may be traveling, and perchance not “from” the coun-
try to which they have been mapped. Additional bias comes
from the English language of the query words (“food” and
“porn”) – albeit popular around the world, English use limits
our access to the native languages.
Regardless of query language, we do find a strong na-
tional cuisine popularity in many countries, including Japan,
China, and Iran, with the accompanying alphabets and
transliterations. The plurality of other nations, including
Canada and USA, speak to the historical developments
which resulted in a multi-cultural environment. Uniting
most nations in our dataset is chocolate, so much so that
a brand of chocolate spread – Nutella – has made it to the
list of top 20 foods mentioned. This finding contrasts that
obtained by the Oxfam survey (gro 2011) of 17 countries
around the world, who found pasta, Chinese, and pizza (in
that order) to be the favorite. Although we find both pizza
and pasta near the top of #foodporn associations (in 3rd and
8th place, to be precise), our data presents a different defini-
tion of “favorite” foods. These differences may be due to the
unique affordances social media presents to its users to form
“images” of both themselves and their favorite foods. These
may include the visual attractiveness of dishes, as well as the
social context of online media sharing.
Looking beyond chocolate and cake, we find that not
all associations with #foodporn are unhealthy. We find
a positive correlation between a nation’s obesity and the
presence of healthy tags (see Figure 9), suggesting that in
such communities there is a greater awareness of healthy
lifestyle. Further, healthy hashtags’ an association with
GDPPC (GDP per capita) suggests the wealthier countries
are developing the “taste” for healthy food, so much so that
it is considered “pornographic” (a contradiction to its im-
plied unhealthy connotations). Finally, the heightened so-
cial approval of healthy tags (Figure 10) suggests that the
community is already self-policing in promoting a health-
ier lifestyle. Stronger health- and fitness-oriented commu-
nities may play an important role, informing local policies
for community health promotion. These results are also
indicative of the ability to use social media for promoting
health (Neiger et al. 2012b). In particular, this analysis in-
forms the persuasive technologies which incorporate experi-
ence formation, behavior tracking and reinforcement via big
data and social media (Fogg 2002).
Although #foodporn may be considered playful or bom-
bastic, its use may have real-world health implications for
the individuals using it. Even though there is less stigma at-
tached to consuming “food porn” than the non-food kinds,
it may still impact individuals with eating disorders. Re-
cently, Signe Rousseau (Rousseau 2014) suggested that the
perception that “consuming food porn may be safer than
consuming real food” may especially effect the “sufferers of
eating disorders who rely on images of food as a substitute
(within limits) for eating”. Further, much like pornography
contributes to an unrealistic view of sexuality, an obsessive
fixation on #foodporn potentially promotes an “unhealthy”
relationship with food. Our findings suggest that there is a
diversity in the context in which the tag is used (including a
healthy one), but a finer-grained analysis is required to detect
the impact of #foodporn on the users with eating disorders.
Conclusion
In this paper we introduce a social media dataset capturing
the global use of the highly popular #foodporn hashtag on
Instagram. At a high rate of geo-location, it captures the
worldwide dietary trends much more thoroughly than the
popular Twitter platform. Our analysis of these 9.3M posts
reveals that while #foodporn is often associated with high-
calorie and sugary foods such as cake and chocolate, it also
often appears in a healthy context. The sentiment associated
with #foodporn indicates that it is used to motivate healthy
living, especially in countries with high GDPPC. This re-
definition of “gastro-porn” within the social media commu-
nity is another illustration of the effect social media has on
our views and conceptualization of our surroundings and
ourselves.
Forthcoming collaboration with nutrition experts in the
creation of rich, culturally diverse food libraries will be in-
strumental to a successful understanding of public health
using the vast amount of data in social media. Resources
similar to United States Department of Agriculture’s food
database14 should be extended to foods in other world
cuisines for a quantitative comparison of nutrient balance
(and imbalance) around the world. In ongoing efforts, we
are also exploring computer vision techniques for under-
standing the content of images, including their composition
and caloric value, for health and wellbeing research.
References
[Abbar, Mejova, and Weber 2015] Abbar, S.; Mejova, Y.; and We-
ber, I. 2015. You tweet what you eat: Studying food consumption
through twitter. ACM SIGCHI’15.
[Ahn et al. 2011] Ahn, Y.-Y.; Ahnert, S. E.; Bagrow, J. P.; and
Barab´
asi, A.-L. 2011. Flavor network and the principles of food
pairing. Scientific reports 1.
[Canetti, Bachar, and Berry 2002] Canetti, L.; Bachar, E.; and
Berry, E. M. 2002. Food and emotion. Behavioural processes
60(2):157–164.
[Capurro et al. 2014] Capurro, D.; Cole, K.; Echavarr´
ıa, M. I.; Joe,
J.; Neogi, T.; and Turner, A. M. 2014. The use of social networking
sites for public health practice and research: a systematic review.
Journal of medical Internet research 16(3).
[Christakis and Fowler 2007] Christakis, N. A., and Fowler, J. H.
2007. The spread of obesity in a large social network over 32 years.
New England journal of medicine 357(4):370–379.
[Cockburn 1977] Cockburn, A. 1977. Gastro-porn. In New York
Review of Books.
14http://ndb.nal.usda.gov/ndb/search/list
[Counihan and Van Esterik 2013] Counihan, C., and Van Esterik, P.
2013. Food and culture: A reader. Routledge.
[Fischler 1988] Fischler, C. 1988. Food, self and identity. Social
science information 27(2):275–292.
[Fogg 2002] Fogg, B. J. 2002. Persuasive technology: using com-
puters to change what we think and do. Ubiquity 2002 (Dec):5.
[gro 2011] 2011. Grow campaign 2011 global opinion research
– final topline report. https://www.oxfam.org/sites/
www.oxfam.org/files/grow-campaign- globescan-
research-presentation.pdf.
[Haddadi et al. 2015] Haddadi, H.; Ofli, F.; Mejova, Y.; Weber, I.;
and Srivastava, J. 2015. 360 quantified self. In IEEE International
Workshop on Smart and Connected Health (SCH 2015). IEEE.
[McBride 2010] McBride, A. E. 2010. Food porn. The Journal of
Food and Culture 10(1):38–46.
[Mejova et al. 2015] Mejova, Y.; Haddadi, H.; Noulas, A.; and We-
ber, I. 2015. # foodporn: Obesity patterns in culinary interactions.
ACM Conference on Digital Health (DH).
[Neiger et al. 2012a] Neiger, B. L.; Thackeray, R.; Burton, S. H.;
Giraud-Carrier, C. G.; and Fagen, M. C. 2012a. Evaluating social
media’s capacity to develop engaged audiences in health promotion
settings use of twitter metrics as a case study. Health promotion
practice 1524839912469378.
[Neiger et al. 2012b] Neiger, B. L.; Thackeray, R.; Van Wagenen,
S. A.; Hanson, C. L.; West, J. H.; Barnes, M. D.; and Fagen, M. C.
2012b. Use of social media in health promotion purposes, key
performance indicators, and evaluation metrics. Health promotion
practice 13(2):159–164.
[O’Connor and O’Connor 2004] O’Connor, D. B., and O’Connor,
R. C. 2004. Perceived changes in food intake in response to stress:
The role of conscientiousness. Stress and Health 20(5):279–291.
[O’Neill 2003] O’Neill, M. 2003. Food porn. Columbia Journalism
Review 5:38–45.
[Park et al. 2015] Park, S.; Kim, I.; Lee, S. W.; Yoo, J.; Jeong, B.;
and Cha, M. 2015. Manifestation of depression and loneliness on
social networks: A case study of young adults on facebook. In
ACM CSCW’15.
[Paul and Dredze 2011] Paul, M. J., and Dredze, M. 2011. You are
what you tweet: Analyzing twitter for public health. In Interna-
tional Conference on Web and Social Media (ICWSM), 265–272.
[Prier et al. 2011] Prier, K. W.; Smith, M. S.; Giraud-Carrier, C.;
and Hanson, C. L. 2011. Identifying health-related topics on twit-
ter. In Social computing, behavioral-cultural modeling and predic-
tion. Springer. 18–25.
[Rousseau 2014] Rousseau, S. 2014. Food “porn” in media. In
Encyclopedia of Food and Agricultural Ethics. Springer.
[Silva et al. 2013] Silva, T. H.; Vaz de Melo, P. O.; Almeida, J. M.;
Salles, J.; and Loureiro, A. A. 2013. A comparison of Foursquare
and Instagram to the study of city dynamics and urban social be-
havior. SIGKDD International Workshop on Urban Computing.
[Tagg and Seargeant 2012] Tagg, C., and Seargeant, P. 2012. The
role of english as an online lingua franca: code choice and translo-
cality on social network sites. Sociolinguistics Symposium.
[West, White, and Horvitz 2013] West, R.; White, R. W.; and
Horvitz, E. 2013. From cookies to cooks: Insights on dietary
patterns via analysis of web usage logs. In WWW’13.
[Yamasaki, Sano, and Mei 2015] Yamasaki, T.; Sano, S.; and Mei,
T. 2015. Revealing relationships between folksonomy and social
popularity score in image/video sharing services. In IEEE Con-
sumer Electronics-Taiwan (ICCE-TW).