Conference PaperPDF Available

Studying Cultural Differences in Emoji Usage across the East and the West

Authors:

Abstract and Figures

Global acceptance of Emojis suggests a cross-cultural, normative use of Emojis. Meanwhile, nuances in Emoji use across cultures may also exist due to linguistic differences in expressing emotions and diversity in conceptualizing topics. Indeed, literature in cross-cultural psychology has found both normative and culture-specific ways in which emotions are expressed. In this paper, using social media, we compare the Emoji usage based on frequency, context, and topic associations across countries in the East (China and Japan) and the West (United States, United Kingdom, and Canada). Across the East and the West, our study examines a) similarities and differences on the usage of different categories of Emojis such as People, Food & Drink, Travel & Places etc., b) potential mapping of Emoji use differences with previously identified cultural differences in users' expression about diverse concepts such as death, money emotions and family, and c) relative correspondence of validated psycho-linguistic categories with Ekman's emotions. The analysis of Emoji use in the East and the West reveals recognizable normative and culture specific patterns. This research reveals the ways in which Emojis can be used for cross-cultural communication.
Content may be subject to copyright.
Proceedings of the Thirteenth International AAAI Conference on Web and Social Media (ICWSM 2019)
Studying Cultural Differences in
Emoji Usage across the East and the West
Sharath Chandra Guntuku,1Mingyang Li,1Louis Tay,2Lyle H. Ungar1
1University of Pennsylvania, 2Purdue University
{sharathg@sas, myli@seas, ungar@cis}.upenn.edu, stay@purdue.edu
Abstract
Global acceptance of Emojis suggests a cross-cultural, nor-
mative use of Emojis. Meanwhile, nuances in Emoji use
across cultures may also exist due to linguistic differences
in expressing emotions and diversity in conceptualizing top-
ics. Indeed, literature in cross-cultural psychology has found
both normative and culture-specific ways in which emotions
are expressed. In this paper, using social media, we compare
the Emoji usage based on frequency, context, and topic asso-
ciations across countries in the East (China and Japan) and the
West (United States, United Kingdom, and Canada). Across
the East and the West, our study examines a) similarities and
differences on the usage of different categories of Emojis
such as People, Food & Drink, Travel & Places etc., b) poten-
tial mapping of Emoji use differences with previously iden-
tified cultural differences in users’ expression about diverse
concepts such as death, money emotions and family, and c)
relative correspondence of validated psycho-linguistic cate-
gories with Ekman’s emotions. The analysis of Emoji use in
the East and the West reveals recognizable normative and cul-
ture specific patterns. This research reveals the ways in which
Emojis can be used for cross-cultural communication.
Introduction
Emoji, a Japan-born ideographic system, offers a rich set of
non-verbal cues to assist textual communication. The Uni-
code Standard 11.0 specified over 2,500 Emojis1, ranging
from facial expressions (‘Smileys’ such as ) to everyday
objects (such as ). Starting as a visual aid for textual com-
munication, Emojis’ non-verbal nature has led to sugges-
tions that they are universal across cultures (Danesi 2016).
In this paper, we examine cross-cultural usages of Emojis
based on (1) linguistic differences across languages (and
cultures) in expressing emotions (Russell et al. 2013), and
(2) diversity in perceiving different constructs among cul-
tures (Boers 2003). Specifically, we compare Emoji use in
terms of frequency, context, and topic associations across
two eastern countries China and Japan and three western
countries United States of America (US), United Kingdom
(UK) and Canada. Hereafter, we refer to the collection of
Copyright c
2019, Association for the Advancement of Artificial
Intelligence (www.aaai.org). All rights reserved.
1http://unicode.org/Emoji/
US, UK, and Canadian cultures as ‘the Western culture’ (or
simply ‘the West’), the collection of Japanese and Mandarin-
speaking Chinese as ‘the East(-ern culture)’ with an ac-
knowledgment that there are several other countries which
can be added to each group (Hofstede 1983). We study the
differences using distributional semantics learned over large
datasets from Sina Weibo and Twitter, two closely related
microblog/social media platforms.
Past psychological research assessing emotional experi-
ence between the East and the West found both universal
and culture-specific types of emotional experience (Eid and
Diener 2009). If Emojis are a form of emotional experience
and expression as prior studies have shown (Kaye, Malone,
and Wall 2017), it is expected that we can find interpretable
and substantial similarity in Emoji usage and also distinct
cultural patterns. In other words, which Emojis are used, the
contexts where they are used, and what they semantically re-
fer to will bear resemblance across languages, even when no
common character is shared between their writing systems.
At the same time, there will also be unique cultural elements
in how certain types of Emojis are used and interpreted.
Research Questions
Due to the richness and diversity of the Emojis, it is diffi-
cult to hypothesize a priori how specific Emojis may differ.
Therefore, we undertake an abductive approach to construct
explanatory theories as patterns emerge from our analy-
sis (Haig 2005). Normatively, we fundamentally expect sim-
ilar patterns of Emoji usage to appear across both cultures.
We also seek to explore when there may be specific cultural
divergence. Therefore, we attempt to answer the following
research questions to explore and quantify the normativeness
and distinctiveness of Emoji usage across the two cultures:
1. How does frequency of different Emojis (as individuals
and in categories) vary across the East and the West?
2. How distinct is Emoji usage across cultures in terms of
validated psycho-linguistic categories they are often asso-
ciated with?
3. How do the semantics of Emoji usage vary when com-
pared against known universal emotion expressions emo-
tions (Ekman basic emotion categories (Ekman 1992))
across both cultures?
226
Background and Related Work
Weibo and Twitter: Analogs?
Weibo and Twitter have been studied to understand the
differences in content and user behaviors in multiple con-
texts (Ma 2013; Lin, Lachlan, and Spence 2016). Notwith-
standing the challenge of working with a non-random, non-
representative sample of social media users, several psycho-
logical traits and outcomes can be inferred from posts, in-
cluding users’ demographics (Sap et al. 2014; Zhang et al.
2016), personality (Li et al. 2014; Quercia et al. 2011), loca-
tion (Salehi et al. 2017; Zhong et al. 2015), as well as status
of stress (Guntuku et al. 2018; Lin et al. 2016), and men-
tal health (Guntuku et al. 2019; Tian et al. 2018) on both
platforms. Demonstrated in these prior studies, the empiri-
cal value suggests that the two platforms are representative
albeit to different countries. Several of the above mentioned
studies have used Linguistic Inquiry Word Count (LIWC)
(Pennebaker et al. 2015), which has psychometrically val-
idated categories, such as emotional valence, religion, and
money etc. in multiple languages.
Role of Culture in Emotion Expression and
Perception
Prior psychological research on emotion (Uchida and Ki-
tayama 2009; Bagozzi, Wong, and Yi 1999) suggests that
evolutionary and biological processes generate universal ex-
pressions and perceptions of emotions. For example, facial
expressions is one of such universal channels that convey
emotions across populations (Ekman 1993). On the other
hand, culture can play a significant role in shaping emotional
life. Specifically, different cultures may value different types
of emotions (e.g., Americans value excitement while Asians
prefer calm) (Tsai, Knutson, and Fung 2006), and there
are different emotional display rules across cultures (Mat-
sumoto 1990). Furthermore, besides representing emotions,
the Emoji system also explicitly contains cultural symbols
and thus potentially represents the distinctive values and
beliefs of cultures (Aaker, Benet-Martinez, and Garolera
2001). Prior works also suggest that culture plays a key role
in predicting perceptions of affect (Guntuku et al. 2015b;
Zhu et al. 2018; Guntuku et al. 2015a). In general, psycho-
logical research reveals both cultural similarities and differ-
ences in emotions (Elfenbein and Ambady 2003).
Emojis: A Proxy for Emotions?
From a methodological perspective, most large scale cross-
cultural psychology research projects have used a survey-
based approach to assess similarities and differences in emo-
tions (Tay et al. 2011). Analyzing use of Emojis between the
Eastern and the Western contexts provides an opportunity to
assess a plethora of behaviors related to emotion expression
and, arguably, emotional symbols that are culturally embed-
ded. This enhances our understanding of similarities and dif-
ferences in emotional life across cultures at a large scale but
fine-grain level.
Prior Studies on Emojis
Prior research on Emojis can be divided primarily into three
themes: (1) studying Emojis as a source of sentiment anno-
tation, (2) analyzing differences in Emoji perception based
on rendering, and (3) understanding similarities and differ-
ences in Emoji usage across different populations. We focus
on studying the similarities and differences in Emoji usage
across cultures.
The studies by (Barbieri et al. 2016; Barbieri, Espinosa-
Anke, and Saggion 2016) are the closest to our work where
authors explore the meaning and usage of Emojis in social
media across four European languages, namely American
English, British English, Peninsular Spanish and Italian, and
across two cities in Spain respectively. They observe that the
most frequent Emojis share similar semantic usages across
these Western languages providing support for the norma-
tive claims of Emojis. However, to fully examine the issue of
normativeness, we need to go beyond examining only West-
ern contexts by also examining East-West similarity.
Since Emojis have been found to be very promising in
downstream applications such as sentiment analysis and
several techniques, utilizing Unicode descriptions (Eisner
et al. 2016; Wijeratne et al. 2017), multi-Emoji expres-
sions (L´
opez and Cap 2017), and including diverse Emoji
sets (Felbo et al. 2017) etc., have been proposed to improve
Emoji understanding, culture-specific norms and platform-
rending effects (Li et al. 2019; Miller Hillberg et al. 2018;
Miller et al. 2016) can be used for improving personaliza-
tion.
Methods
This work received approval from University of Pennsylva-
nia’s Institutional Review Board (IRB). Code repository is
available online2.
Data Collection
To obtain data for the US, UK, Canada and Japan, we
used Twitter data from a 10% archive from the TrendMiner
project (Preotiuc-Pietro et al. 2012), which used the Twitter
streaming API. Since Twitter is not widely used in China,
we obtained Weibo data. Since Weibo lacks a streaming in-
terface (as Twitter) for downloading random samples over
time, we queried for all posts from a given user. The list of
users were crawled using a breadth-first search strategy be-
ginning with random users.
Pre-processing
The count of posts for each country after each stage of pre-
processing and final user counts is shown in Table 1.
Geo-location: On Twitter, the coordinates or tweet coun-
try location (whichever was available) was used to geo-
locate posts. On Weibo, user’s self-identified profile location
was used to identify the geo-location of messages. We used
messages posted in the year 2014 in both corpora.
2https://github.com/tslmy/ICWSM2019
227
Culture Country # Posts crawled # Posts after lang. filter # Posts after geo-location # Users
West
USA 29.32M 18.99M 18.57M 4.39M
UK 6.74M 5.12M 4.83M 1.31M
Canada 1.6M 1.16M 1.12M 0.32M
East Japan 481.39M 82.56M 17.51M 2.06M
China 486.18M 205.22M 1.00M
Table 1: Number of posts in each corpora after each pre-processing stage and the final number of users in our analysis.
Language Filtering: To remove the confounds of bilin-
gualism (Fishman 1980), we filter posts by the languages
they are composed in. Language used in each Twitter post
(or ‘tweets’ hereafter) is detected via langid (Lui and Bald-
win 2012). Tweets written in languages other than English
in US, UK and Canada, and Japanese in Japan are re-
moved. Weibo posts are filtered for Chinese language us-
ing pre-trained fastText language detection models (Xu and
Mori 2011), due to its ability to distinguish between Man-
darin and Cantonese (we used only Mandarin posts in our
analysis). Further, traditional Chinese characters are con-
verted to Simplified Chinese using hanziconv Python pack-
age 3to conform with LIWC dictionary used in later sec-
tions. We also remove any direct re-tweets (indicated by
RT @USERNAME: on Twitter and @USERNAME// on
Weibo).
Tokenizing: Twitter text was tokenized using Social To-
kenizer bundled with ekphrasis4, a text processing pipeline
designed for social networks. Weibo posts were segmented
using Jieba5considering its ability to discover new words
and Internet slang, which is particularly important for a
highly colloquial corpus like Sina Weibo. Using ekphra-
sis, URLs, email addresses, percentages, currency amounts,
phone numbers, user names, emoticons and time-and-
dates were normalized with meta-tokens such as <url>’,
<email>’, <user> etc. Skin tone variation in Emo-
jis was introduced in 2015, and consequently no skin-toned
Emoji was captured in our corpora gathered from 2014.
Training Embedding Models
To study the lexical semantics across both cultures, we
trained a Word2Vec Continuous Bag-of-Words (CBoW)
model on each corpus/country (Mikolov et al. 2013). These
models were trained for 10 epochs with learning rates ini-
tialized at .025 and allowed to drop till 104. The dimen-
sion of learned token vectors was chosen to be 100 based
on previous work (Barbieri, Ronzano, and Saggion 2016).
To counter effects due to the randomized initialization in
the Word2Vec algorithm, each model was trained 5 times
independently. In all our analysis, we used the vector em-
beddings across the 5 instances for every analysis and then
averaged the resulting projections.
Measuring Topical Differences
In order to investigate topical differences across cultures,
we use LIWC dictionaries in Chinese (Huang et al. 2012)
3https://pypi.org/project/hanziconv/
4https://github.com/cbaziotis/ekphrasis
5https://github.com/fxsjy/jieba
and in English (Pennebaker, Francis, and Booth 2001) to
be consistent across languages. Since LIWC is not avail-
able in Japanese, we used methods from prior work (Shi-
bata et al. 2016) to translate the word lists from Chinese and
English into Japanese. The LIWC dictionary is a language-
specific, many-to-many mapping of tokens (including words
and word stems) and psychologically validated categories.
Each category (a curated list of words) is found to be corre-
lated with and also predictive of several psychological traits
and outcomes (Pennebaker, Francis, and Booth 2001).
We use the terms ‘tokens’, ‘Emojis’, ‘words’, etc. inter-
changeably with their corresponding vectorial representa-
tions. Next, we define our auxiliary term ‘category vectors’,
compute Emoji-category similarities, and analyze correla-
tions.
Preparing category vectors: In each corpus separately,
for each LIWC category i {Posemo,Family, ...}, all to-
kens (in vectorial representations ~
tl,ij) in this category are
averaged into one vector, which we term as “category vec-
tor” ~cl,i:
~cl,i =1
nl,i
nl,i
X
j=1
~
tl,ij
for corpus l {US,UK,Canada,Japan,China}where nl,i
is the amount of tokens in the LIWC category iin the corpus
l, and ~
tl,ijis the j-th token in the LIWC category iin the
corpus l.
Acknowledging that LIWC captures only verbal tokens,
and that Emojis, as non-verbal tokens, may have substan-
tial differences to verbal tokens captured during Word2Vec
training stage, we orthonormalize axial vectors using Gram-
Schmidt algorithm (Bj¨
orck 1994), to ensures they capture
more distinctive features between LIWC categories.
Computing cosine similarities: Separately in each cor-
pus l, for each pair of Emoji j { , , . . .}(in vectorial
representation ~
tl,j ) and category vector ~cl,i , a cosine similar-
ity is computed:
sl,i,j =sim ~cl,i,~
tl,j .
For clarity, we define ~sl,i ={sl,i,j for j}. Per-country co-
sine similarities are then averaged across each culture to re-
veal the western and the eastern cosine similarities:
~sW,i =1
3X
l∈{US,UK,Canada}
~sl,i,
and
~sE,i =1
2X
l∈{China,Japan}
~sl,i
for the category i {Posemo,Family, ...}.
228
Figure 1: Top 15 frequent Emojis in the East and in the West,
in percentage of total Emojis captured in the corresponding
corpora. East-West rank order correlation is .745.
Spearman Correlation Coefficients: For each of the
31 LIWC categories shared across all corpora, i
{Posemo,Family, ...}, we correlate the Western Emoji Us-
age vector ~sW,i = (sW,1, sW,2, ..., sW,J )and the Eastern
Emoji Usage vector ~sE,i = (sE,1, sE,2, ..., sE,J ), denoting
the Spearman correlation coefficient with ρi. Here, Jis the
total number of Emojis present in all corpus and appeared
for at least 1,000 times in total.
Results and Discussion
Frequency of Emoji Usage
Among the 1,281 Emojis defined in Emoji 1.06by Uni-
code7,602 Emojis appeared in all corpora. Only 528 of them
appeared more than 1,000 times. Figure 1 shows 15 most
frequently seen Emojis in each culture. Across the two cul-
tures, Spearman correlation coefficient (SCC) is 0.745 (two-
tailed t-test p-value <0.005). These statistics indicate a
strong correspondence in the types of Emojis favored across
these two cultures. This reveals normativeness in the types
of Emojis used between East and West.
Further, we sum up the usage frequencies by Unicode Cat-
egory of Emojis. Figure 2 shows the frequency of these cat-
egories, denoted with SCCs for Emoji frequencies within
each category, representing the similarity between West-
erners and Easterners in using the Emojis in each cate-
gory. While the SCC values range from moderate (.383) to
high (.807), suggesting a high correspondence of Emoji us-
age patterns, drilling down by categories of Emojis is elu-
cidating. The lowest correlations in Emoji usage occur in
the ‘Symbols’, ‘Food & Drink’, and Activities’ categories.
This is not surprising, because cultures often have their
own meaning symbols that are representative of specific val-
ues (Aaker, Benet-Martinez, and Garolera 2001). Moreover,
culture is often instantiated in cuisines representing dietary
preferences, identities, and ecology (Van den Berghe 1984).
Further, culture also influences the time spent across the
world on work, play, and development activities (Larson and
Verma 1999).
Semantic Similarity of Emojis
Vector representations allow for mathematical projections,
which essentially serve as a measure of similarity. We com-
pute a pairwise similarity for each pair of Emojis in each
6Published in August 2015, Emoji 1.0 is closest to 2014, the
year from which our corpora were gathered.
7http://unicode.org/Public/emoji/1.0/emoji-data.txt
Figure 2: Normalized frequency of Emojis grouped by Uni-
code categories. SCCs across East and West, r, are denoted
below each.
Figure 3: Pairwise similarities (measured by Pearson r) of
countries in terms of Emojis learned from Word2Vec mod-
els. East-West rof .59 indicates some level of normativ-
ity, though lower than previous findings across Western lan-
guages.
country, and use the vectors of per-country pairwise Emoji
similarities as the basis of generating a country-level pair-
wise similarity matrix (shown in Figure 3). The Pearson cor-
relation coefficient between the West and the East is 0.59,
indicating similarity in the semantics of Emoji usage even
across two different cultures. While this supports some level
of normativity, we find that this level of East-West similar-
ity is lower than previous findings (Barbieri et al. 2016) of
Emoji semantics across four Western languages where simi-
larity matrices of Emojis were correlated > .70. Our within
Western nation correlations were similar to past findings,
ranging from .65 to .79. Altogether, these results reveal that
there is still normativity in Emoji usage across East-West
with the moderate positive Pearson correlation, though there
is less similarity than if we were to compare across Western
nations.
229
Figure 4: Association with Psycholinguistic categories represented by LIWC with top 5 Emojis in the Eastern countries and in
the West, ranked by their similarity with each LIWC category. The SCC for all Emojis in each LIWC category is also presented
on the right indicating a measure of corresponding (dis-)similarity. In each cell, the SCC is computed for the shown Emoji
between the two cultures. The two arrays (on which this SCC is calculated) contain the cosine similarities of the Emoji vector
and each LIWC category vector.
Association with Psycholinguistic Categories
Figure 4 demonstrates top 5 Emojis similar to each LIWC
category in the East and in the West. Specifically, these re-
sults reveal the correspondence between the LIWC category
(and all its related words) and a set of Emojis. The extent that
SCCs are high shows that the same set of words across two
cultures relate to the same types of Emojis; low SCCs reveal
that the same category of words is associated with different
Emojis across the two cultures.
There is overall evidence for normativeness between East
and West in how concepts captured by LIWC are represented
by Emojis. Almost all the LIWC categories have positive
SCCs and the median SCC is .38. At the same time, there are
also specific categories that reveal more distinctiveness be-
tween the two cultures. In the next paragraphs, we describe
specific findings in an exploratory manner.
Substantively, LIWC categories can be represented from
words into Emoji expressions; the rank-order correlations
reveal if these Emoji expressions overlap in reflecting the
specific category.
East-West Similarities. LIWC categories that are most
similar in terms Emojis are ‘Ingest’, ‘Death’, Anger’,
‘Money’, ‘Home’, and ‘Family’. Prima facie, many of these
categories are recognized as universal and the choice of
Emojis to represent these categories are the most similar.
Given that money is a medium of exchange in almost all so-
cieties of the world given global capitalism (Berger and Dore
230
Figure 5: Top and Bottom 5 Most Universal Emojis in Terms of Similarities to LIWC Categories, grouped by Unicode Category.
Emoji icon differences measured by Spearman correlation across East and West. Correlation is computed on similarities of the
Emoji vectors to each of the LIWC category vectors in both corpora. Top 10 and bottom 10 correlated Emojis across both
platforms are shown for each category in the Unicode Consortium, along with the mean and variance of each category.
1996), ‘Money’ is a category that is universally understood
and regarded in a similar way and this is represented as such
with Emojis ( , , etc.). This also applies to the Emoji ex-
pressions in the category of ‘Death’ ( and in the East;
, , etc. in the West) which is the ultimate issue all hu-
mans face (K¨
ubler-Ross 1973). Similarly, the categories of
emotion Anger’ ( , etc.) is tied to the basic emotions of
anger which has been found to be universally expressed and
recognized facially (Ekman 1992). The category of Ingest
( , etc.) and how people imbibe food as expressed in Emo-
jis are also similar based on the rank-order correlations.
East-West Differences. At the same time, amid similarity
overall, we also observe that there are some cultural dimen-
sions that emerge from the plots as the Emojis for rice bowl
( ) and ramen ( ) dominated the East, while meat-related
Emojis ( , , etc.) take the majority in the West (Prescott
and Bell 1995; Ahn et al. 2011). On the other hand, the
LIWC categories that have lower correlations, and indeed
even, inverse correlations show that Emojis used to express
these constructs are likely different overall. ‘Insight’, ‘Dis-
crepancy’, ‘Quantitative’, ‘Number’, ‘Time’, ‘Friend’, and
‘Work’ have small or near-zero correlations. This seems to
be in line with categories that are linked to cultural influ-
ence. In terms of the grammatical categories of ‘Quantifiers’
and ‘Number’, given that there are differences in the gram-
mar and syntax of Chinese, Japanese, and English, this differ-
ence is also understandable. Similarly, ‘Time’ is often viewed
symbolically and is laden with cultural meaning; moreover,
there are differences in the importance of keeping time and
the timing of events (Brislin and Kim 2003). East Asians
and Westerner also have differences in interpersonal deal-
ings. With regard to the former, Confucianism places a pre-
mium on harmony and proper relationships as the basis for
Asian society whereas Westerners often place greater impor-
tance in outcomes and direct communication (Yum 1988).
This is revealed in differences in the category of ‘Friend’ in
how Emojis are used to express that idea. To a larger extent,
it also extends to the broader category of social processes
where we also find a low correlation on Emojis expressing
the concept of ‘Family’. Emojis also seem to reflect govern-
mental policies. For example, the Chinese government had
been banning game consoles till 2015, and in our dataset
collected from 2014 the game controller Emoji that
dominates West in ‘Leisure’ is nowhere to be seen in the East.
With regard to the categories of Discrepancy and Work
having lower correlations, our explanations are at best spec-
ulative. The category of ‘Discrepancy’ (containing the tex-
tual tokens ‘should’, ‘could’, ‘would’) may be expressed
differently with Emojis due to the deferential culture and
higher power-distance in the Eastern context as opposed
to the Western context (Farh, Hackett, and Liang 2007;
Schwartz 1994).
Icon Differences
Given that Emojis are based on an ideographic system where
each symbol represents a specific concept, it is also impor-
tant to examine how Emojis are similar or different between
the Eastern and Western contexts. For this, we transpose the
results from association with psycholinguistic categories to
determine the extent Emojis are similar based on how the
different LIWC categories are projected onto the Emojis.
Figure 5 shows the top 5 and bottom 5 Emojis (ranked by
SCC) across both platforms for each Unicode category. The
mean and std. dev of each of these categories are also pre-
sented. Social scientific theories emphasizing on the univer-
sality of basic emotions suggests that emotion-expressing
Emojis tend to show high convergence even between dis-
231
Figure 6: SCCs of LIWC categories and Ekman emotion words using similarities with all Emojis as underlying values. The
three axis of the scatter plot represent SCCs between the three western countries, SCCs between the the eastern countries, and
SCCs between the West and the East, respectively. Both adjective and noun forms of the Ekman emotion words are considered
and labeled in red. Points with blue labels are LIWC categories. Multi-view projection is shown as quiver plots, where arrows
are colored under the same rules.
tinct cultures such as the East and the West (Ekman 1992;
2016). Therefore, it is expected that categories of ‘Smi-
leys’ and ‘People’ would likely be more convergent com-
pared to other categories. Consistent with this expectation,
the Emoji categories of ‘Smileys’ and ‘People’ display rela-
tively higher mean correlations between both cultures (ρ=
.31 and .32 respectively).
A significant cultural difference component is language
itself as it forms the basis of cultural expression. The use
of emics, from within the social group, in anthropology and
psychology where cultural behaviors and ideas are under-
stood from the context of the culture itself emphasizes the
specificity and distinctiveness of language rather than its
commonality (Harris 1976). From this view, the ‘Symbols’
category is likely to converge less. This was borne out from
the relatively low average correlations (.17) between Eastern
and Western cultures for the ‘Symbols’ category.
It is again important to note that there appears to be ev-
idence for the universality of Emojis from this analysis as
there is a positive correlation across all the different Emoji
categories. Further categories such as ‘Objects’ and ‘Travel’
had similar levels of correlations as ‘Smileys’ and ‘People’.
232
Association with Universal Ekman Emotions
We further investigated the semantics of Emoji usage and
how they vary when compared against expression of univer-
sal basic emotions (specifically Ekman categories). If Emoji
representations are normative, we would find similar levels
of SCCs with basic emotion categories. As LIWC does not
cover all 6 Ekman basic emotions (Ekman 1992), we looked
at specific emotion words such as ‘anger’ and ‘happy’. To
overcome the selection bias in part of speech, we considered
both nouns and adjective forms. We consider 12 individual
word vectors learned from each Word2Vec model. Hence,
we extend the previous definition of ‘category vector’ ~cl,i to
include also 12 Ekman emotion words. For each LIWC cat-
egory and Ekman emotion word i, SCCs between country
pairs {l1, l2}are computed. Represented by ~sdl1,l2,i, they
are then averaged with respect to whether l1and l2are both
from the East (‘In-East SCCs’), both from the West (‘In-
West SCCs’), or different cultures (‘Cross-Cultural SCCs’).
The 3 vectors are plotted as coordinates in the 3D scatter
graph in Figure 6. We find that, based on SCC magnitudes,
there is a greater similarity in Emoji representation among
Western nations compared to the East. LIWC categories
such as ‘Friend’, ‘Insight’, ‘Motion’, ‘Work’, and ‘Number’
had relatively low similarities within Western and Eastern
contexts and also did not have substantial East-West similar-
ity. However, categories such as Anger’, ‘posemo’ (positive
emotion), ‘Death’, and ‘Family’ had relatively higher simi-
larities within Western and Eastern contexts and also higher
similarities across the two cultures. Ekman emotion word
terms were also included in the quiver plots to assess the
degree to which basic emotions are similar within and also
across Eastern and Western cultures. We found that the most
universal terms were with regard to anger and happiness
(i.e., similarity within and between cultures). However, with
regard to surprise, disgust, sadness, and fear there was less
relative similarity across cultures. We emphasize relative be-
cause this also confirms our findings that the Emoji represen-
tations instantiated in LIWC categories have a substantial
degree of normativeness; therefore, we find that even ba-
sic emotion categories (e.g., surprise, disgust, sad/sadness,
fear/terrified) do not uniquely distinguish themselves to have
much higher cross-cultural SCCs, although we do see a
trend that some categories like ‘Quant’, ‘Time’, ‘Work’, and
‘Space’ have lower cross-cultural convergence.
Limitations and Future Work
In this paper, we find, amid similarity, that there are some
cultural dimensions that emerge from how the semantics
of Emoji vary across both cultures. While these differences
were studied primarily from the perspective of Emoji use,
a large portion of it could potentially also be attributed to
the text of the post. By training Word2Vec models with both
text and Emoji tokens across the East and the West, and by
analyzing the Emoji associations with word categories as
captured by LIWC, we attempt to uncover the interactions
between both. However, it would be interesting to study
the cross-cultural variation in text and Emoji usage inde-
pendently to quantify each. Approaches from recent work
on understanding Emoji ambiguity in English (Miller et al.
2017) could be coupled with ours to achieve this goal.
Even though we attempted to compare Emoji usage in 2
Eastern countries (Japan and China) and 3 Western coun-
tries (US, UK, and Canada) respectively, we nevertheless
used two different platforms, namely Weibo and Twitter
to represent each. We also used LIWC from Chinese and
English, and obtained Japanese version by translating the
Chinese LIWC. This could potentially introduce confounds
around platform differences, over and above cultural differ-
ences. To minimize this, we restricted posts based on geo-
location, dropped bilingual posts, and analyzed posts in the
primary language of communication in each country. Prior
studies also found a lot of similarities in user demograph-
ics, intention of use and topical differences (Gao et al. 2012;
Ma 2013; Lin, Lachlan, and Spence 2016). However, it
would be promising to look at data from other sources (such
as smart phone users (Lu et al. 2016)) where social desir-
ability and censorship confounds might be lower to further
validate the findings in our study.
All our analyses were based on correlation rather than
causality. Because of the richness and diversity of the Emo-
jis, it is difficult to hypothesize a priori how specific Emojis
may differ. Therefore, we undertook an abductive approach
to construct explanatory theories as patterns emerged from
our analysis (Haig 2005). For social science research, these
methods offer data-driven insights into group and user be-
haviors which can be used to generate new hypotheses for
testing and can be used to unobtrusively measure large popu-
lations over time. Commercial applications include improv-
ing targeted online marketing, increasing acceptance of Hu-
man Computer Interaction systems and personalized cross-
cultural recommendations for communication.
Future work should investigate similarities and differ-
ences in other socio-psychological constructs. Also, consid-
ering the promise of Emojis in downstream application tasks
such as sentiment analysis, studies should explore the contri-
bution of Emojis in multi-modal and cross-lingual sentiment
analysis and transfer learning tasks.
Conclusion
In this paper, we compared Emoji usage based on frequency,
context, and topic associations across countries in the East
(China and Japan) and the West (United States, United King-
dom, and Canada). Our results offer insight into cultural
similarities and differences at several levels. In general, we
found evidence for the normativeness, or the universality,
of Emojis. While there are relative differences in that West-
ern users tend to use more Emojis than Eastern users, the
relative frequencies in different types of Emojis are cor-
related across cultures. Moreover, distributional semantics
found that the Emoji expressions were clustered in a similar
manner across cultures. Even when we used universal basic
emotions as a benchmark, we found that Emojis were rep-
resented in a cross-culturally similar manner compared to
these basic emotion expressions.
At the same time, we found that there appear to also in-
terpretable distinctions between Emoji use based on topical
analyses. Emojis were culturally specific as certain types of
233
Emojis such as rice-based dishes had the highest projection
on the LIWC category of ‘Ingest’ in the East while a mix of
meat and spaghetti had the highest project on the same cate-
gory in the West. Analysis at the icon level reveal support for
general social scientific theories of cultural similarities and
differences where relative similarities were found more in
terms of the ‘Smileys’ and ‘People’ icons whereas relative
differences were found for ‘Symbols’ icons. Nevertheless,
these findings need to be construed from the perspective that
there appears to be a robust thread of cross-cultural similar-
ity in Emoji patterns.
References
Aaker, J. L.; Benet-Martinez, V.; and Garolera, J. 2001. Consump-
tion symbols as carriers of culture: A study of japanese and spanish
brand personality constucts. Journal of personality and social psy-
chology 81(3):492.
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
1:196.
Bagozzi, R. P.; Wong, N.; and Yi, Y. 1999. The role of culture
and gender in the relationship between positive and negative affect.
Cognition & Emotion 13(6):641–672.
Barbieri, F.; Kruszewski, G.; Ronzano, F.; and Saggion, H. 2016.
How cosmopolitan are emojis?: Exploring emojis usage and mean-
ing over different languages with distributional semantics. In Pro-
ceedings of the 2016 ACM on Multimedia Conference, 531–535.
ACM.
Barbieri, F.; Espinosa-Anke, L.; and Saggion, H. 2016. Revealing
patterns of twitter emoji usage in barcelona and madrid. Frontiers
in Artificial Intelligence and Applications. 2016;(Artificial Intelli-
gence Research and Development) 288: 239-44.
Barbieri, F.; Ronzano, F.; and Saggion, H. 2016. What does this
emoji mean? a vector space skip-gram model for twitter emojis. In
LREC.
Berger, S., and Dore, R. P. 1996. National diversity and global
capitalism. Cornell University Press.
Bj¨
orck, ˚
A. 1994. Numerics of gram-schmidt orthogonalization.
Linear Algebra and Its Applications 197:297–316.
Boers, F. 2003. Applied linguistics perspectives on cross-
cultural variation in conceptual metaphor. Metaphor and Symbol
18(4):231–238.
Brislin, R. W., and Kim, E. S. 2003. Cultural diversity in people’s
understanding and uses of time. Applied Psychology 52(3):363–
382.
Danesi, M. 2016. The semiotics of emoji: The rise of visual lan-
guage in the age of the internet. Bloomsbury Publishing.
Eid, M., and Diener, E. 2009. Norms for experiencing emotions
in different cultures: Inter-and intranational differences. In Culture
and Well-Being. Springer. 169–202.
Eisner, B.; Rockt¨
aschel, T.; Augenstein, I.; Boˇ
snjak, M.; and
Riedel, S. 2016. emoji2vec: Learning emoji representations from
their description. arXiv preprint arXiv:1609.08359.
Ekman, P. 1992. An argument for basic emotions. Cognition &
emotion 6(3-4):169–200.
Ekman, P. 1993. Facial expression and emotion. American psy-
chologist 48(4):384.
Ekman, P. 2016. What scientists who study emotion agree about.
Perspectives on Psychological Science 11(1):31–34.
Elfenbein, H. A., and Ambady, N. 2003. Universals and cultural
differences in recognizing emotions. Current directions in psycho-
logical science 12(5):159–164.
Farh, J.-L.; Hackett, R. D.; and Liang, J. 2007. Individual-level
cultural values as moderators of perceived organizational support–
employee outcome relationships in china: Comparing the effects of
power distance and traditionality. Academy of Management Jour-
nal 50(3):715–729.
Felbo, B.; Mislove, A.; Søgaard, A.; Rahwan, I.; and Lehmann,
S. 2017. Using millions of emoji occurrences to learn any-
domain representations for detecting sentiment, emotion and sar-
casm. arXiv preprint arXiv:1708.00524.
Fishman, J. A. 1980. Bilingualism and biculturism as individual
and as societal phenomena. Journal of Multilingual & Multicul-
tural Development 1(1):3–15.
Gao, Q.; Abel, F.; Houben, G.-J.; and Yu, Y. 2012. A comparative
study of users’ microblogging behavior on sina weibo and twitter.
In International Conference on User Modeling, Adaptation, and
Personalization, 88–101. Springer.
Guntuku, S. C.; Lin, W.; Scott, M. J.; and Ghinea, G. 2015a. Mod-
elling the influence of personality and culture on affect and enjoy-
ment in multimedia. In 2015 International Conference on Affective
Computing and Intelligent Interaction (ACII), 236–242. IEEE.
Guntuku, S. C.; Scott, M. J.; Yang, H.; Ghinea, G.; and Lin, W.
2015b. The cp-qae-i: A video dataset for exploring the effect of per-
sonality and culture on perceived quality and affect in multimedia.
In 2015 Seventh International Workshop on Quality of Multimedia
Experience (QoMEX), 1–7. IEEE.
Guntuku, S. C.; Buffone, A.; Jaidka, K.; Eichstaedt, J.; and Ungar,
L. 2018. Understanding and measuring psychological stress using
social media. arXiv preprint arXiv:1811.07430.
Guntuku, S. C.; Preot¸iuc-Pietro, D.; Eichstaedt, J. C.; and Ungar,
L. 2019. What twitter profile and posted images reveal about de-
pression and anxiety.
Haig, B. D. 2005. An abductive theory of scientific method. Psy-
chological methods 10(4):371.
Harris, M. 1976. History and significance of the emic/etic distinc-
tion. Annual review of anthropology 5(1):329–350.
Hofstede, G. 1983. National cultures in four dimensions: A
research-based theory of cultural differences among nations. Inter-
national Studies of Management & Organization 13(1-2):46–74.
Huang, C.-L.; Chung, C. K.; Hui, N.; Lin, Y.-C.; Seih, Y.-T.; Lam,
B. C.; Chen, W.-C.; Bond, M. H.; and Pennebaker, J. W. 2012.
The development of the chinese linguistic inquiry and word count
dictionary. Chinese Journal of Psychology.
Kaye, L. K.; Malone, S. A.; and Wall, H. J. 2017. Emojis: Insights,
affordances, and possibilities for psychological science. Trends in
cognitive sciences 21(2):66–68.
K¨
ubler-Ross, E. 1973. On death and dying. Routledge.
Larson, R. W., and Verma, S. 1999. How children and adoles-
cents spend time across the world: work, play, and developmental
opportunities. Psychological bulletin 125(6):701.
Li, L.; Li, A.; Hao, B.; Guan, Z.; and Zhu, T. 2014. Predicting
active users’ personality based on micro-blogging behaviors. PloS
one 9(1):e84997.
Li, M.; Guntuku, S. C.; Jakhetiya, V.; and Ungar, L. H. 2019. Ex-
ploring (dis-)similarities in emoji-emotion association on twitter
and weibo. In EMOJI2019 colocated with The WebConf. ACM.
Lin, H.; Jia, J.; Nie, L.; Shen, G.; and Chua, T.-S. 2016. What does
social media say about your stress? In Proceedings of IJCAI.
234
Lin, X.; Lachlan, K. A.; and Spence, P. R. 2016. Exploring extreme
events on social media: A comparison of user reposting/retweeting
behaviors on twitter and weibo. Computers in Human Behavior
65:576–581.
L´
opez, R. P., and Cap, F. 2017. Did you ever read about frogs
drinking coffee? investigating the compositionality of multi-emoji
expressions. In Proceedings of the 8th Workshop on Computational
Approaches to Subjectivity, Sentiment and Social Media Analysis,
113–117.
Lu, X.; Ai, W.; Liu, X.; Li, Q.; Wang, N.; Huang, G.; and Mei,
Q. 2016. Learning from the ubiquitous language: an empirical
analysis of emoji usage of smartphone users. In Proceedings of
the 2016 ACM International Joint Conference on Pervasive and
Ubiquitous Computing, 770–780. ACM.
Lui, M., and Baldwin, T. 2012. langid. py: An off-the-shelf lan-
guage identification tool. In Proceedings of the ACL 2012 system
demonstrations, 25–30. Association for Computational Linguis-
tics.
Ma, L. 2013. Electronic word-of-mouth on microblogs: A cross-
cultural content analysis of twitter and weibo. Intercultural Com-
munication Studies 22(3).
Matsumoto, D. 1990. Cultural similarities and differences in dis-
play rules. Motivation and emotion 14(3):195–214.
Mikolov, T.; Sutskever, I.; Chen, K.; Corrado, G. S.; and Dean, J.
2013. Distributed representations of words and phrases and their
compositionality. In Advances in neural information processing
systems, 3111–3119.
Miller, H.; Thebault-Spieker, J.; Chang, S.; Johnson, I.; Terveen,
L.; and Hecht, B. 2016. Blissfully happy” or “ready to fight”:
Varying interpretations of emoji. Proceedings of ICWSM 2016.
Miller, H. J.; Kluver, D.; Thebault-Spieker, J.; Terveen, L. G.; and
Hecht, B. J. 2017. Understanding emoji ambiguity in context: The
role of text in emoji-related miscommunication. In ICWSM, 152–
161.
Miller Hillberg, H.; Levonian, Z.; Kluver, D.; Terveen, L.; and
Hecht, B. 2018. What i see is what you don’t get: The effects of
(not) seeing emoji rendering differences across platforms. Proceed-
ings of the ACM on Human-Computer Interaction 2(CSCW):124.
Pennebaker, J. W.; Boyd, R. L.; Jordan, K.; and Blackburn, K.
2015. The development and psychometric properties of liwc2015.
Technical report.
Pennebaker, J. W.; Francis, M. E.; and Booth, R. J. 2001. Linguistic
inquiry and word count: Liwc 2001. Mahway: Lawrence Erlbaum
Associates 71(2001):2001.
Preotiuc-Pietro, D.; Samangooei, S.; Cohn, T.; Gibbins, N.; and Ni-
ranjan, M. 2012. Trendminer: An architecture for real time analysis
of social media text.
Prescott, J., and Bell, G. 1995. Cross-cultural determinants of food
acceptability: recent research on sensory perceptions and prefer-
ences. Trends in Food Science & Technology 6(6):201–205.
Quercia, D.; Kosinski, M.; Stillwell, D.; and Crowcroft, J. 2011.
Our twitter profiles, our selves: Predicting personality with twitter.
In Privacy, Security, Risk and Trust (PASSAT) and 2011 IEEE Third
Inernational Conference on Social Computing (SocialCom), 2011
IEEE Third International Conference on, 180–185. IEEE.
Russell, J. A.; Fern´
andez-Dols, J.-M.; Manstead, A. S.; and
Wellenkamp, J. C. 2013. Everyday conceptions of emotion: An in-
troduction to the psychology, anthropology and linguistics of emo-
tion, volume 81. Springer Science & Business Media.
Salehi, B.; Hovy, D.; Hovy, E.; and Søgaard, A. 2017. Huntsville,
hospitals, and hockey teams: Names can reveal your location. In
Proceedings of the 3rd Workshop on Noisy User-generated Text,
116–121.
Sap, M.; Park, G.; Eichstaedt, J.; Kern, M.; Stillwell, D.; Kosin-
ski, M.; Ungar, L.; and Schwartz, H. A. 2014. Developing age
and gender predictive lexica over social media. In Proceedings of
the 2014 Conference on Empirical Methods in Natural Language
Processing (EMNLP), 1146–1151.
Schwartz, S. H. 1994. Beyond individualism/collectivism: New
cultural dimensions of values.
Shibata, D.; Wakamiya, S.; Kinoshita, A.; and Aramaki, E. 2016.
Detecting japanese patients with alzheimer’s disease based on word
category frequencies. In Proceedings of the Clinical Natural Lan-
guage Processing Workshop (ClinicalNLP), 78–85.
Tay, L.; Diener, E.; Drasgow, F.; and Vermunt, J. K. 2011. Mul-
tilevel mixed-measurement irt analysis: An explication and appli-
cation to self-reported emotions across the world. Organizational
Research Methods 14(1):177–207.
Tian, X.; Batterham, P.; Song, S.; Yao, X.; and Yu, G. 2018. Char-
acterizing depression issues on sina weibo. International journal
of environmental research and public health 15(4):764.
Tsai, J. L.; Knutson, B.; and Fung, H. H. 2006. Cultural variation
in affect valuation. Journal of personality and social psychology
90(2):288.
Uchida, Y., and Kitayama, S. 2009. Happiness and unhappiness in
east and west: themes and variations. Emotion 9(4):441.
Van den Berghe, P. L. 1984. Ethnic cuisine: Culture in nature.
Ethnic and Racial Studies 7(3):387–397.
Wijeratne, S.; Balasuriya, L.; Sheth, A.; and Doran, D. 2017.
Emojinet: An open service and api for emoji sense discovery. In
Eleventh International AAAI Conference on Web and Social Media.
Xu, M., and Mori, N. 2011. Fast text character set recognition. US
Patent 7,865,355.
Yum, J. O. 1988. The impact of confucianism on interpersonal
relationships and communication patterns in east asia. Communi-
cations Monographs 55(4):374–388.
Zhang, W.; Caines, A.; Alikaniotis, D.; and Buttery, P. 2016. Pre-
dicting author age from weibo microblog posts. In LREC.
Zhong, Y.; Yuan, N. J.; Zhong, W.; Zhang, F.; and Xie, X. 2015.
You are where you go: Inferring demographic attributes from lo-
cation check-ins. In Proceedings of the eighth ACM international
conference on web search and data mining, 295–304. ACM.
Zhu, Y.; Guntuku, S. C.; Lin, W.; Ghinea, G.; and Redi, J. A. 2018.
Measuring individual video qoe: A survey, and proposal for fu-
ture directions using social media. ACM Transactions on Mul-
timedia Computing, Communications, and Applications (TOMM)
14(2s):30.
235
... In total, we obtained about 29 million posts posted in 2014 and 2015 from 859,054 users. Based on prior work, we used the selfreported location from user's profile information (Guntuku et al., 2019). Three prior studies have found that this method of locating participants is accurate and reliable on Twitter and Weibo (Giorgi et al., 2018;Jaidka et al., 2020;Cui et al., 2022). ...
... We analyzed 266 million terms from~8.8 million tweets posted between 2014 and 2019, collected using the Twitter API with a bounding box around the country of Japan. We further geolocated them to the 47 Japanese prefectures by looking for matches between each prefecture name and the Twitter location field obtained from user profile information based on prior work (Guntuku et al., 2019). ...
... We then tested whether the 10 LIWC word categories related to rice in China would also be related to rice farming in Japan. Since LIWC is not available in Japanese, we used methods from prior work to translate the word lists from Chinese and English into Japanese (Shibata et al., 2016;Guntuku et al., 2019). This method has been used in prior studies on social media across cultures (Guntuku et al., 2019) and in spoken language in Japan (Shibata et al., 2016). ...
Article
Full-text available
We used natural language processing to analyze a billion words to study cultural differences on Weibo, one of China’s largest social media platforms. We compared predictions from two common explanations about cultural differences in China (economic development and urban-rural differences) against the less-obvious legacy of rice versus wheat farming. Rice farmers had to coordinate shared irrigation networks and exchange labor to cope with higher labor requirements. In contrast, wheat relied on rainfall and required half as much labor. We test whether this legacy made southern China more interdependent, as measured by modern day language. Across all word categories, rice explained twice as much variance as economic development and urbanization. Rice areas used more words reflecting tight social ties, holistic thought, and a cautious, prevention orientation. We then used Twitter data comparing prefectures in Japan, which largely replicated the results from China. This provides crucial evidence of the rice theory in a different nation, language, and platform.
... Such pictographs purport to represent a type of "ubiquitous language, capable of being understood by people from varying linguistic, social and cultural backgrounds" (Wiseman and Gould 2018: 1), to the effect that the meaning of the most popular emoji are similar across countries or cultures, with "money"-emoji being the primary example (Völkel et al. 2019;Guntuku et al. 2019). However, variability and differences in emoji meaning also occur across cultures, e.g., in expressing notions such as 'time', 'positivity', 'praise', 'friends', 'work' or 'discrepancy' on the East-West axis (Togans et al. 2021;Guntuku et al. 2019). ...
... Such pictographs purport to represent a type of "ubiquitous language, capable of being understood by people from varying linguistic, social and cultural backgrounds" (Wiseman and Gould 2018: 1), to the effect that the meaning of the most popular emoji are similar across countries or cultures, with "money"-emoji being the primary example (Völkel et al. 2019;Guntuku et al. 2019). However, variability and differences in emoji meaning also occur across cultures, e.g., in expressing notions such as 'time', 'positivity', 'praise', 'friends', 'work' or 'discrepancy' on the East-West axis (Togans et al. 2021;Guntuku et al. 2019). For many users, creating a shared emoji with special meaning is a way of communicating important ideas in entirely non-verbal ways. ...
Article
Full-text available
Communication via digital media is characterized by growing use of graphic elements employed to convey emotions and concepts in a succinct form, especially in interpersonal interactions. Emoji are naturally ambiguous in meaning, which undergoes various modifications. One way of incorporating emoji into interpersonal communication is repurposing their meaning and use, so that a shared personal meaning develops between individual users or within small groups. The article explores novel and unique applications of certain emoji in an independent (conceptual) replication study of Wiseman and Gould’s work (2018). The present study analyzes the data obtained via a web-based survey in a group of 132 respondents aged 17 to 55+ to subsequently get to know their motivations behind the repurposed use of emoji. It is demonstrated how this adaptation of graphic icons helps people express their likes and preferences, romantic feelings, amusement, and, as a result, can aid users in building closer relationships in micro-communities. In essence, the article is intended to contribute to a better understanding of emoji use in multimodal interpersonal communication.
... Several studies have explored factors influencing human interpretation of emojis, including personality traits ) and demographics. Gender and age are frequently cited as having a profound impact on emoji comprehension, with additional associations identified between cultural background, religious beliefs, and emoji interpretation (Guntuku et al. 2019;Wang 2022). These demographic influences, particularly age and gender, are central to our investigation. ...
Preprint
Emojis have become a universal language in online communication, often carrying nuanced and context-dependent meanings. Among these, irony poses a significant challenge for Large Language Models (LLMs) due to its inherent incongruity between appearance and intent. This study examines the ability of GPT-4o to interpret irony in emojis. By prompting GPT-4o to evaluate the likelihood of specific emojis being used to express irony on social media and comparing its interpretations with human perceptions, we aim to bridge the gap between machine and human understanding. Our findings reveal nuanced insights into GPT-4o's interpretive capabilities, highlighting areas of alignment with and divergence from human behavior. Additionally, this research underscores the importance of demographic factors, such as age and gender, in shaping emoji interpretation and evaluates how these factors influence GPT-4o's performance.
... However, our corpus did contain over a thousand emoji, which justifies our quantitative statistical analyses. Because all participants were Dutch, results may not be generalisable crosslinguistically or cross-culturally, since emoji use has been shown to differ across cultures (Barbieri et al. 2016;Freedman 2018;Guntuku et al. 2019). This invites future research to take a contrastive cross-cultural perspective to children's emoji use. ...
Article
Full-text available
Emoji (e.g., 🤪✈🧁) are increasingly used on social media by people of all ages, but little is known about the concept ‘emoji literacy’. To investigate different age groups’ emoji preferences, an exploratory corpus analysis was conducted using an innovative corpus-gathering method: children and adults were instructed to add emoji magnets to pre-constructed printed social media messages. The corpus (with 1,012 emoji) was coded for the number of emoji used per message, the type of emoji, their position and function in the message, and the sentiment they conveyed. Intuitions about emoji use turned out to be similar for children and adults, with greater use of facial emoji, emoji at the end of messages, emoji to express emotions, and emotional emoji to convey positive sentiment. Children’s emoji preferences were studied in more detail. Results revealed that their age, gender, smartphone ownership, and social media use related to differences in the number, position, and function of the emoji used. The data showed that older children, girls, children with their own smartphone, and children using social media exhibited a more advanced and sophisticated use of emoji than younger children, boys, and children without smartphones or social media experience. This study constitutes an important first step in exploring children’s emoji literacy and use.
... In addition, the same emoji might convey diverse meanings across diverse cultures [19]. Conversely, similar meanings can be conveyed by disparate appearances owing to cultural distinctions [20,21]. Even users from the same culture, such as those of different ages, often disagree on the usage and interpretation of emoji [17,22,23]. ...
Article
Full-text available
This study explored emoji-based sarcastic statements made by Chinese teenagers using the WeChat app. Two experiments (N = 597) were conducted to investigate both language production and comprehension. In Experiment 1 (N = 236), a free-response task was used to determine which emoji Chinese teenagers would use to signal their sarcastic intentions. The results showed that the smiling emoji (Image 1) and the tears-of-joy emoji (Image 2) were the most commonly used emoji to indicate sarcastic intent. Experiment 2 (N = 361) involved a rating task to investigate the role that two selected emoji played in the comprehension and emotional impact of WeChat messages (literal and sarcastic statements). Our findings revealed that both emoji generally increased the perceived sarcasm in literal comments; however, while the smiling emoji intensified the perceived sarcasm in sarcastic comments, the tears-of-joy emoji had no significant effect on such comments. Regarding emotional impact, the smiling emoji amplified the negativity of the statement, whereas the tears-of-joy emoji mitigated it. This indicates that both emoji and context are reliable cues for interpreting sarcasm, though their levels of reliability may differ. Specifically, the smiling emoji carries greater weight than context, while context is afforded stronger weight than the tears-of-joy emoji in the constraint-satisfaction process of sarcasm. Overall, the smiling emoji is perceived as more sarcastic and more negative than the tears-of-joy emoji.
... Facebook satisfies more hedonic needs, such as entertainment and social interaction, it may lead to a greater openness to humorous content when compared to platforms such as Twitter or LinkedIn which serve more informative needs (Tellis et al., 2019). By focusing on Facebook, the context is also somewhat limited to Western, Englishspeaking cultures, thus future research should explore the possibility of a cultural influence (Guntuku et al., 2019). For example, China is home to the largest gay social app in the world, Blued, thus determining the generalizability of our findings is warranted (Williams, 2021). ...
Article
Full-text available
LGBT+ brands, i.e., those supporting and catering for the LGBT+ community and primarily owned and operated by members of this community, have been standing out for their authenticity in a marketplace where brands are regularly accused of “rainbow‐washing.” Yet, despite the recent media spotlight these brands are still struggling to grow and gain visibility among the general public. This study aims to explore how LGBT+ brands can leverage their content to increase e‐Word‐of‐Mouth (eWOM) on social media. Specifically, we focus on humorous content because its impact on eWOM has been mixed in the literature and brand type has been overlooked as a potential moderator to date. Using brand anthropomorphism and schema congruity theory, our research proposes that LGBT+ brands that adopt humor in their social media content generate higher eWOM when compared to mainstream brands. We suggest that consumers attribute the expected characteristics of gay individuals (e.g., humorous, witty, …) to LGBT+ brands, thus leading to perceived brand‐post fit as the underlying mechanism. Two studies based on Facebook data from two different LGBT+ brands and a scenario‐based experiment validate these findings. Alongside conceptualizing LGBT+ brands, we provide valuable insights into how these brands can gain exposure and visibility on social media.
... Guntuku et al. [83] studied cultural differences in social media usage, suggesting that understanding these patterns can enhance marketing strategies across different demographics. ...
Preprint
Full-text available
Rationale: The pervasive influence of social media on consumer behavior and marketing strategies necessitates a comprehensive understanding of its impact on the hospitality industry. As hotels increasingly leverage digital platforms for engagement and reputation management, examining the effectiveness of various social media strategies and their implications becomes crucial for optimizing marketing efforts and enhancing brand identity. Objective: This study aims to investigate the impact of social media on hotel management, branding, engagement, and reputation in the hospitality sector. Method: A mixed-methods approach was employed, involving quantitative surveys of hotel customers and qualitative interviews with hotel marketing managers. Data on demographic characteristics, social media usage patterns, primary purposes of social media usage, perceptions of hotel branding, and reputation management practices were collected and analyzed to identify trends and insights. Results: The data reveals that promotional offers and discounts (100%), user reviews (90%), and visual content (80%) are critical factors influencing consumer hotel choices. Collaboration with influencers, user-generated content, live video content, and social media advertising all demonstrated 100% effectiveness in engaging audiences and building brand identity. Pre-social media, 40% of hotels promptly addressed issues, while post-social media, this decreased to 20%, with private resolutions increasing from 20% to 30%. Conclusions: Social media significantly influences consumer behavior in hotel selection and booking decisions. The shift in reputation management practices underscores the need for hotels to balance transparency with effective conflict resolution strategies. Social media strategies that leverage influencers, user-generated content, and dynamic visual content are crucial for enhancing engagement and brand identity. Recommendations: Hotels should prioritize influencer collaborations, encourage user-generated content, and invest in high-quality visual media. Implementing regular updates, engaging contests, and targeted promotional offers can drive bookings and enhance customer satisfaction. Proactive reputation management and transparent communication are essential for maintaining a positive public image. Significance Statement: This study provides valuable insights into the transformative impact of social media on the hospitality industry, offering evidence-based strategies and policy recommendations for optimizing digital marketing efforts. The findings underscore the importance of a strategic and multifaceted approach to social media, highlighting its role in shaping consumer behavior, enhancing brand identity, and driving business growth.
... The Rank of Top 10 Emojis in the East and West[8] ...
Article
Full-text available
In the digital era, Emojis have emerged as a critical component of online communication, transcending linguistic barriers while introducing new challenges in cross-cultural conversations. This paper explores the complex ways in which cultural backgrounds influence the selection and interpretation of Emojis, providing a nuanced understanding of digital nonverbal communication. By using a comparative analysis between Eastern and Western usage of popular Emojis, the research highlights significant cultural discrepancies. The study traces Emoji popularity trends from 2019 to 2021, a period notably affected by the COVID-19 pandemic, which catalyzed the global shift towards online interaction and promoted the usage of specific Emojis related to work, empathy, and emotional connection. Through a sociolinguistic lens, this study illuminates the role of Emojis as a modern pictographic language that encapsulates the inherent diversity of cultural emotion, nuance, and global digital communication. The paper's findings highlight the need to increase cultural awareness of Emoji interpretation and application to reduce misunderstandings in the increasingly connected world.
... The research shows that despite the normative use of emojis across cultures, there are still differences in the use and understanding of emojis due to different backgrounds. As an example, a study found significant differences in the frequency of emojis use, context, and thematic associations between East Asia (China and Japan) and Western countries (the United States, the United Kingdom, and Canada) [3]. In addition, another research has shown that Western cultures are more inclined to use emojis with negative facial expressions , whereas East Asian cultures are more inclined to use emojis that represent medically relevant objects, such as syringes and pills. ...
Article
Full-text available
This study explores the differences in emoji interpretation between Chinese and Japanese users, focusing on the influence of culture on digital communication. Responses from 25 participants were collected through a questionnaire distributed on Twitter and WeChat, which included personal information, social media use and understanding of emoji. The results of the study showed that although Chinese and Japanese users perceived certain emoji similarly, there were differences in the interpretation of other emoji. These results emphasize the importance of cultural context in emoji use and provide valuable insights into cross-cultural communication.
Article
Full-text available
Previous work has found strong links between the choice of social media images and users’ emotions, demographics and personality traits. In this study, we examine which attributes of profile and posted images are associated with depression and anxiety of Twitter users. We used a sample of 28,749 Facebook users to build a language prediction model of survey-reported depression and anxiety, and validated it on Twitter on a sample of 887 users who had taken anxiety and depression surveys. We then applied it to a different set of 4,132 Twitter users to impute language-based depression and anxiety labels, and extracted interpretable features of posted and profile pictures to uncover the associations with users’ depression and anxiety, controlling for demographics. For depression, we find that profile pictures suppress positive emotions rather than display more negative emotions, likely because of social media self-presentation biases. They also tend to show the single face of the user (rather than show her in groups of friends), marking increased focus on the self, emblematic for depression. Posted images are dominated by grayscale and low aesthetic cohesion across a variety of image features. Profile images of anxious users are similarly marked by grayscale and low aesthetic cohesion, but less so than those of depressed users. Finally, we show that image features can be used to predict depression and anxiety, and that multitask learning that includes a joint modeling of demographics improves prediction performance. Overall, we find that the image attributes that mark depression and anxiety offer a rich lens into these conditions largely congruent with the psychological literature, and that images on Twitter allow inferences about the mental health status of users.
Conference Paper
Full-text available
Emojis have gained widespread acceptance, globally and cross-culturally. However, Emoji use may also be nuanced due to differences across cultures, which can play a significant role in shaping emotional life. In this paper, we a) present a methodology to learn latent emotional components of Emojis, b) compare Emoji-Emotion associations across cultures, and c) discuss how they may reflect emotion expression in these platforms. Specifically, we learn vector space embeddings with more than 100 million posts from China (Sina Weibo) and the United States (Twitter), quantify the association of Emojis with 8 basic emotions, demonstrate correlation between visual cues and emotional valence, and discuss pairwise similarities between emotions. Our proposed Emoji-Emotion visu-alization pipeline for uncovering latent emotional components can potentially be used for downstream applications such as sentiment analysis and personalized text recommendations. CCS CONCEPTS • Human-centered computing → Social networks.
Conference Paper
Full-text available
A body of literature has demonstrated that users' mental health conditions, such as depression and anxiety, can be predicted from their social media language. There is still a gap in the scientific understanding of how psychological stress is expressed on social media. Stress is one of the primary underlying causes and correlates of chronic physical illnesses and mental health conditions. In this paper, we explore the language of psychological stress with a dataset of 601 social media users, who answered the Perceived Stress Scale questionnaire and also consented to share their Facebook and Twitter data. Firstly, we find that stressed users post about exhaustion, losing control , increased self-focus and physical pain as compared to posts about breakfast, family-time, and travel by users who are not stressed. Secondly, we find that Facebook language is more predictive of stress than Twitter language. Thirdly, we demonstrate how the language based models thus developed can be adapted and be scaled to measure county-level trends. Since county-level language is easily available on Twitter using the Streaming API, we explore multiple domain adaptation algorithms to adapt user-level Facebook models to Twitter language. We find that domain-adapted and scaled social media-based measurements of stress outperform sociodemographic variables (age, gender, race, education, and income), against ground-truth survey-based stress measurements, both at the user-and the county-level in the U.S. Twitter language that scores higher in stress is also predictive of poorer health, less access to facilities and lower socioeconomic status in counties. We conclude with a discussion of the implications of using social media as a new tool for monitoring stress levels of both individuals and counties.
Article
Full-text available
This research argues that the meaning embedded in consumption symbols, such as commercial brands, can serve to represent and institutionalize the values and beliefs of a culture. Relying on a combined emic-etic approach, the authors conducted 4 studies to examine how symbolic and expressive attributes associated with commercial brands are structured and how this structure varies across 3 cultures. Studies 1 and 2 revealed a set of "brand personality" dimensions common to both Japan and the United States (Sincerity, Excitement, Competence, and Sophistication), as well as culture-specific Japanese (Peacefulness) and American (Ruggedness) dimensions. Studied 3 and 4, which extended this set of findings to Spain, yielded brand personality dimensions common to both Spain and the United States (Sincerity, Excitement, and Sophistication), plus nonshared Spanish (Passion) and American (Competence and Ruggedness) dimensions. The meaning of these brand personality dimensions is discussed in the context of cross-cultural research on values and affect, globalization issues, and cultural frame shifting.
Article
Full-text available
Within- and between-nations differences in norms for experiencing emotions were analyzed in a cross-cultural study with 1,846 respondents from 2 individualistic (United States, Australia) and 2 collectivistic (China, Taiwan) countries. A multigroup latent class analysis revealed that there were both universal and culture-specific types of norms for experiencing emotions. Moreover, strong intranational variability in norms for affect could be detected, particularly for collectivistic nations. Unexpectedly, individualistic nations were most uniform in norms, particularly with regard to pleasant affect. Individualistic and collectivistic nations differed most strongly in norms for self-reflective emotions (e.g., pride and guilt). Norms for emotions were related to emotional experiences within nations. Furthermore, there were strong national differences in reported emotional experiences, even when norms were held constant.
Article
Full-text available
The next generation of multimedia services have to be optimized in a personalized way, taking user factors into account for the evaluation of individual experience. Previous works have investigated the influence of user factors mostly in a controlled laboratory environment which often includes a limited number of users and fails to reflect real-life environment. Social media, especially Facebook, provide an interesting alternative for Internet-based subjective evaluation. In this article, we develop (and open-source) a Facebook application, named YouQ1, as an experimental platform for studying individual experience for videos. Our results show that subjective experiments based on YouQ can produce reliable results as compared to a controlled laboratory experiment. Additionally, YouQ has the ability to collect user information automatically from Facebook, which can be used for modeling individual experience.
Article
Full-text available
The prevalence of depression has increased significantly over the past few years both in developed and developing countries. However, many people with symptoms of depression still remain untreated or undiagnosed. Social media may be a tool to help researchers and clinicians to identify and support individuals who experience depression. More than 394,000,000 postings were collected from China’s most popular social media website, Sina Weibo. 1000 randomly selected depression-related postings was coded and analyzed to learn the themes of these postings, and a text classifier was built to identify the postings indicating depression. The identified depressed users were compared with the general population on demographic characteristics, diurnal patterns, and patterns of emoticon usage. We found that disclosure of depression was the most popular theme; depression displayers were more engaged with social media compared to non-depression displayers, the depression postings showed geographical variations, depression displayers tended to be active during periods of leisure and sleep, and depression displayers used negative emoticons more frequently than non-depression displayers. This study offers a broad picture of depression references on China’s social media, which may be cost effectively developed to detect and help individuals who may suffer from depression disorders.
Article
Emoji are popular in digital communication, but they are rendered differently on different viewing platforms (e.g., iOS, Android). It is unknown how many people are aware that emoji have multiple renderings, or whether they would change their emoji-bearing messages if they could see how these messages render on recipients' devices. We developed software to expose the multi-rendering nature of emoji and explored whether this increased visibility would affect how people communicate with emoji. Through a survey of 710 Twitter users who recently posted an emoji-bearing tweet, we found that at least 25% of respondents were unaware that the emoji they posted could appear differently to their followers. Additionally, after being shown how one of their tweets rendered across platforms, 20% of respondents reported that they would have edited or not sent the tweet. These statistics reflect millions of potentially regretful tweets shared per day because people cannot see emoji rendering differences across platforms. Our results motivate the development of tools that increase the visibility of emoji rendering differences across platforms, and we contribute our cross-platform emoji rendering software to facilitate this effort.
Conference Paper
Recent studies have found that people interpret emoji characters inconsistently, creating significant potential for miscommunication. However, this research examined emoji in isolation, without consideration of any surrounding text. Prior work has hypothesized that examining emoji in their natural textual contexts would substantially reduce potential for miscommunication. To investigate this hypothesis, we carried out a controlled study with 2,482 participants who interpreted emoji both in isolation and in multiple textual contexts. After comparing the variability of emoji interpretation in each condition, we found that our results do not support the hypothesis in prior work: when emoji are interpreted in textual contexts, the potential for miscommunication appears to be roughly the same. We also identify directions for future research to better understand the interplay between emoji and textual context.