Emoticon Recommendation System Reﬂecting User Individuality

A Preliminary Survey of Emoticon Use

Taichi Matsui and Shohei Kato

Graduate School of Engineering, Nagoya Institute of Technology, Gokiso-cho, Showa-ku, Nagoya, 466-8555, Japan

Keywords:

Emoticon, Emotion, Recommendation, Text Messaging, Internet.

Abstract:

As the Internet has become widespread, t ext messaging has become a major means of communication. Be-

cause it is difﬁcult to express emotion through text, emoticons were developed. There are many kinds of

emoticons, and people often have difﬁculty ﬁnding one that conveys their meaning appropriately. This rese-

arch aims to propose an emoticon recommendation system that considers individual differences. To this end,

we conducted a survey about the use of emoticons. In this study, we report and analyze the results of this

survey.

1 INTRODUCTION

Along with the development of Intern e t tec hnology,

Internet communications have become popular. Many

types of Internet communications are text-based. Ty-

pical examples include e-mail, Twitter, Facebook and

LINE (Japans largest instant communicatio ns appli-

cation). These applications m ake it easy to use the

Internet f or communication. However, it is difﬁcult

to convey non-verbal information, such as facial ex-

pressions or tone of voice, using only text. There-

fore, emoticons are widely used to convey emotions

and facial exp ressions in text-based communications.

Emoticon s compr ise various punctuatio n marks and

are designed to convey an emotional state in plain text

messages(Riva, 2002)(Walther and DAddario, 2001).

(Arakawa et al., 2006) demo nstrated that emoti-

cons exist to controll the feelings of the communi-

cating parties and facilitate communication. Howe-

ver, emoticons are different from language vocabu-

lary because emoticons by themselves do not have a

clear meaning. Therefore, it is left to the user to de-

termine what m eaning their selected emoticon con-

veys, and its interpretation also remains ambiguous.

(Ono et al., 2003) d e ﬁned two types of emoticons:

emoticons that tend to have a similar interpr etation

and emoticons tha t are interpreted differently. (Naka-

maru, 2002 ) conﬁrmed that the degree of conﬁd e nce,

feelings, and evaluation increases whe n the meaning

of the sentence and emoticon match. On the other

hand, (Nakamaru, 2002) also noted that when an emo-

ticon is used that does not match the meaning of the

text, it is important to consider whether the intention

is conveyed correctly to the re c eiver.

There is a gender difference in emoticon use

(Wolf, 2000). There are also differences in emoticon

interpretation across cultures (Park et al., 2013). (Park

et a l., 2013) determined that an emoticon s meaning

can vary de pending on the identity of the speaker by

investigating a large-scale dataset of over one billion

tweets from different time periods and countries.

These studies suggested two important points.

One is that emoticons play an imp ortant role in com-

munication. The other is that if the sender does not

understand the receivers backgroun d, he or she may

not be able to convey the correct feelings if the emo-

ticons are incorrectly interpre te d. This research aims

to propose a system to help the sender select an emo-

ticon that is appropriate for the rece iver. In this study,

we report the results of a survey on emoticon use,

which will inform the design of the pro posed emo-

ticon recom mendation system.

2 RELATED STUDIES

Many studies have been conducted on emoticons

and face character recommendation systems. (Urabe

et al., 2013 ) created an emoticon database using a sur-

vey and proposed a system to recommend an emoti-

con that expresses a similar feeling as estimated from

the text. (Emura and Seki, 2012)proposed a method

to recommend emoticons by e stima ting the feeling,

type of communic a tion and movement from a text in-

Matsui T. and Kato S.

Emoticon Recommendation System Reﬂecting User Individuality - A Preliminary Survey of Emoticon Use.

DOI: 10.5220/0006186104590464

In Proceedings of the 9th International Conference on Agents and Artiﬁcial Intelligence (ICAART 2017), pages 459-464

ISBN: 978-989-758-220-2

459

put by the user. The type of communication inclu-

des e moticons that assume communication like apo-

logy. The type of movement includes emoticons that

convey the movement like sleep. However, the goal

of these studies is to recommend emoticons that suit

the users text from a wid e range of emoticon s. Cer-

tainly, these emoticon r e commendation systems are

useful bec ause the number and type of available e mo-

ticons are in creasing. However, as (Arakawa et al.,

2006) suggested, the role of emoticons is not merely

to emphasize the emotion of the text but to modify the

overall mean ing of a sentence and facilitate commu-

nication. It is possible for a sender to use an emoticon

that expresses a feeling tha t is different from that of

text. For example, a smiley emoticon may be used af-

ter the text expressing anger to modify the expressed

anger. Thus, we believe it is not sufﬁcient for an emo-

ticon re c ommendation system to simply recommend

an emoticon that suits th e text. (Ono et al., 2003) sug-

gested, it is also important to consider individual dif-

ferences when using emoticons because there are in-

dividual differences in re cognizing the meaning of an

emoticon.

3 THE PROPOSED SYSTEM: AN

OVERVIEW

Figure 1 shows an overview of the proposed system.

In this system, the intended feeling of the senders

emoticon is ﬁrst estimated from the text and e moti-

con by the sender (a database that re la te s each users

feelings with emoticons is c reated in advance). Se-

cond, emoticons that generate feelings in the receiver

that are similar to those intended by the sender are se-

lected. Finally, the emoticon c a ndidates are displayed

to the sender. The sender selects a new emoticon from

those displa yed emoticons and completes the text. We

believe this system will make it easier for a sender to

select an effective emoticon.

In Figure 1, the sender selects the emoticon

(ˆ_ˆ)

. The system extracts the evaluation of

(ˆ_ˆ)

from the senders database and estimates certain emo-

ticons that express a similar emotion from the recei-

vers datab ase. The sender th en selects emoticon (·∀·).

4 PRELIMINARY EXPERIMENTS

AND RESULTS

It is necessary to obtain an individuals im pression of

each em oticon to complete the proposed system. The-

refore, we conducte d two surveys: on e was about how

users use em oticons, and the other was about the va-

rious em otions users ascribe to an e moticon. In this

study, we used the emoticons selected by ( Kawakami,

2008) a nd common emoticon s determined by a web

questionn aire.

4.1 Survey 1: How Users Use Emoticons

We condu cted a survey about how users comm only

use emoticons. User s were asked 1) what c ommun i-

cation apps they usually used, 2) who their communi-

cation partners were and how frequently they commu-

nicated with them, 3) the average number of messages

sent pe r day, and the emoticons most fre quently used.

4.1.1 Subjects

We collected the answers to the survey from 37 pe-

ople with an average age of 22.6 years. The youngest

person was 22 years old, the oldest was 27 years old,

and the median age was 22 years. Further, the survey

sample com prised 15 males and 22 Females. In terms

of education, 28 had an engineering education and 9

had a humanities education. Twenty-four respondents

were students and 13 were employed. Thirty-two Ja-

panese, 4 Indian s, and 1 Chinese participated in the

study.

4.1.2 Results

Figure 2 shows the results for commonly used com-

munication apps, and Figure 3 shows the results for

the types of communicatio n partners and frequency of

communication. Table 2 shows the average number of

messages per day sent by the survey participants, and

Table 1 shows exam ples of their frequently used emo-

ticons. In this ta ble, colored c e lls indicate emoticons

used by more the n one subject.

4.1.3 Communication Partners and Frequency

As Figure 2 shows, all subjects use L INE, which indi-

cates that LINE is widely used as a common commu-

nication app lication. Table 2 shows that the majority

messages are sent between real friends and then inter-

net friends. Figure 3 shows that the frequency of com-

munication between real frien ds is every day. In con-

trast, the most common f requency of communication

between Internet friends is not at all and the next mo st

common frequency is every 2 or 3 days. This means

that there are two types of subjects, those who com-

municate with Internet friends frequently and those

who do not.

ICAART 2017 - 9th International Conference on Agents and Artiﬁcial Intelligence

460

㻴㼑㼘㼘㼛

(^_^)

㻱㼙㼛㼠㼕㼏㼛㼚㻌㼐㼍㼠㼍㼎㼍㼟㼑

㼒㼛㼞㻌㼠㼔㼑㻌㼟㼑㼚㼐㼑㼞

㻱㼟㼠㼕㼙㼍㼠㼑㻌㼠㼔㼑㻌㼑㼙㼛㼠㼕㼛㼚

㼛㼒㻌㼠㼔㼑㻌㼑㼙㼛㼠㼕㼏㼛㼚

㻱㼙㼛㼠㼕㼏㼛㼚㻌㼐㼍㼠㼍㼎㼍㼟㼑

㼒㼛㼞㻌㼠㼔㼑㻌㼞㼑㼏㼑㼕㼢㼑㼞

㻱㼟㼠㼕㼙㼍㼠㼑㻌㼑㼙㼛㼠㼕㼏㼛㼚㼟

㼑㼤㼜㼞㼑㼟㼟㼕㼚㼓㻌㼟㼕㼙㼕㼘㼍㼞㻌㼑㼙㼛㼠㼕㼛㼚

(䞉Ѧ䞉)

(*^_^*)

䠄䠺ν䠺䠅

㻱㼟㼠㼕㼙㼍㼠㼑㼐㻌㼑㼙㼛㼠㼕㼏㼛㼚㼟

㻴㼑㼘㼘㼛

(䞉Ѧ䞉)

㻿㼑㼘㼑㼏㼠㼕㼛㼚

㼎㼥㻌㼠㼔㼑㻌㼟㼑㼚㼐㼑㼞

㻿㼑㼚㼐㼑㼞

㻾㼑㼏㼑㼕㼢㼑㼞

Figure 1: Emoticon recommendation system.

Table 1: Part of frequently used emoticons. In this table, colored cells indicate emoticons used by more than one subject.

Table 2: Average number of messages per day(SD) .

Real friends Family Internet friends St rangers

20.6(23.7) 8.0(15.8) 11.5(34.1) 2.6(12.4)

4.1.4 Emoticon Use

The average number of emoticons subjects usually

used is 8.6. This number is extremely low comp ared

with the approximately 100 million types of emoti-

cons registered in the e moticon dictionary Minna no

Kaomoji (Minna no Kaomoji, 2016). In addition, as

Table 1 shows, there are a few emoticons that are used

by more than on e subjects. In total, 234 varieties of

emoticons were collected in this survey and 197 vari-

eties of emoticons (almost 84 %) were unique to on e

user. Hence, we conclude that subje c ts select emo-

ticons from an enormous range of emoticons depen-

ding on their preference s and characteristics. In o t-

her words, the em oticons selected by a subject could

represent individuality of that person. This suggests

that e moticons are used as a way of not only expres-

sing emotion but also describing perso nality.

4.2 Survey 2: User Interpretatio n of

Emoticon Emotions

It is necessary to evaluate individual emotions con-

veyed by emoticons to implement the proposed sy -

stem. Therefo re, we conducted a survey to determine

the various emo tions u sers ascribe to an emoticon.

Emoticon Recommendation System Reﬂecting User Individuality - A Preliminary Survey of Emoticon Use

461

0.0

10.0

20.0

30.0

40.0

50.0

60.0

70.0

80.0

90.0

100.0

㼚㼡㼙㼎㼑㼞㼟㼞㼍㼠㼕㼛

Figure 2: Frequently used ommunication apps.

Figure 3: Communication partners and the frequency.

4.2.1 Affective Evaluation Method for

Emoticons

Using a ﬁve-point scale, (Kawakami, 2008) studied

the degree of six emotions for 31 emoticons. (Ka-

wakami, 2008) used the basic emotions: happiness,

sadness, anger, amusement, impatience, and surprise

to evaluate the emoticons and created a database. We

also used the same ba sic emotions for our evalua tion.

In the questionnaire, subjects evaluated 131 emoti-

cons using the six e motions on a ﬁve-point scale from

1 ( You cannot feel th e emotion) to 5 (You can feel the

emotion very well). The em oticons were display ed

randomly. The subjects were instruc te d to not ima-

gine the comm unication partner to eliminate th e pos-

sibility of the type of commu nication partner a ffecting

the evalua tion.

4.2.2 Consideration of Obtained Evaluation

Figure 4 shows combined results for ha ppiness obtai-

ned by (Kawakami, 2008) and this study. It also

shows how the results o f b oth studie s are similar. This

similarity is also seen for the ﬁve other emotions. we

Table 3: Emoticons divided into two clusters(partial results

shown).

next compared the sevaluation values of males and fe-

males. Figure 5 shows that females tend to give lower

emotion scores a nd this tendency is also seen in the

ﬁve other emotions. Furthermore, we investigated the

evaluation values o f each emoticon. Figure 6 shows

an emoticon to wh ic h males anger the highest score

and fe males gave sadness the highest score . This re-

sult indicate s a difference in the emotion evaluatio n

of males and fem ales.

We conduc te d a cluster analysis u sin g a six-

dimensional vector to represent each emoticon. We

used the h clust func tion of R and ca lculated it using

Wards criterion. Figure 7 shows the result, which in-

dicates two clusters. Table 3 shows example emoti-

cons for two clusters. The emoticons are divided into

positive and negative emoticons.

5 CONCLUSION

In this study, we focused on computer-mediated com-

munication and the use of emoticons used as a way

to express emotion in text. Our aim is an e moticon

recommendation system that depends on individua-

lity. We thus conduc ted two surveys to determine how

users use and interpret emoticons. The results show

that users may select emoticons dep ending on their

preferences and characteristics and sometimes they

may have different inte rpretations of the same emo-

ticon. However, the number o f subjects in this study

is small, so more sub je cts are needed to verify the re-

sults. In the future, we plan to implement the pro-

posed system and have sub jects evalu a te the system.

There are many cha llenges to implementing the pro-

posed system, which include determining the emoti-

cons that are best suited to express emotions, how to

guaran tee the individuality of users and how to com-

pare em oticons. Further, we need to determine the

best interface for our system and how to implement

it. We pla n to tackle these all challen ges in the future.

ICAART 2017 - 9th International Conference on Agents and Artiﬁcial Intelligence

462

1.5

2.5

3.5

4.5

䠄^-^䠅

䠄䞉Ѧ䞉䠅

䠄㼿 Ѧ 䠼䠅

䠄^_^䠅

䠄^ 0 ^䠅

䠄^䕱^䠅

䠄>Ѧ<䠅

䠄^0^䠅/

䠄Ӎ䕱ӌ䠅

䠄o^o^o䠅

䠄T_T䠅

䠄>_<䠅

䠄;_;䠅

䠄ToT䠅

䠄> <䠅

(䠼Д㼿)

䠄-_-#䠅

䠄¯_¯#䠅

(䠼䜈㼿)

䠄䢟䕕䢟;䠅

(䢛Д䢛;)

䠄^_^;䠅

䠄¯䕕¯;䠅

(䢛Д䢛)

䠄䢟䕕䢟䠅

(䢛䕕䢛)

䠄*_*䠅

䠄䛴Д 䠼䠅

Σ䠄䢟䕕䢟䠅

䠄䢟ω䢟䠅

m(_ _)m

⿕㦂⪅ᖹᆒ

ඛ⾜◊✲

Figure 4: Evaluation values for happiness from the results of a previous study (31 emoticons from Kawakami (2008)) and this

study.

1.5

2.5

3.5

4.5

䠄^-^䠅

䠄㼿 Ѧ 䠼䠅

䠄^ 0 ^䠅

䠄>Ѧ<䠅

䠄T_T䠅

䠄;_;䠅

(䠼Д㼿)

䠄¯_¯#䠅

䠄䢟䕕䢟;䠅

䠄^_^;䠅

(䢛Д䢛)

(䢛䕕䢛)

䠄^0^䠅/

䠄> <䠅

䠄o^o^o䠅

(`䡡ω䡡㼿)䢅

(㼿䡡ω䡡`)

((((䠗䢛Д䢛)))))))

( ^ω^ )

m(_ _)m

䠄䠺ω䠺䠅

(㼿Д` )

( *䠼ω㼿)

(䢟䞊䢟)

( 䞉ω䞉)

(-_-)

:-D

:-!

-@--@-

:-S

:-*

_(┐䛂ε:)_

_(-ω-`_)՜)_

:'-(

:-)

:-@

@_@

:-P

m9(^Д^)

0(:3 )ࠥ

O.o?

(>_<;)

(*.*)

( ˘ω˘)

䠄䖒೅䖓䠅

(^ω^;;;;;)

:")

(㼿䠗ω䠗䠼)

(*㼿Ѧ䠼)

(∗ 䙠⃘ 䕳䙡⃘ ∗)

㻔㼿и•ω•и 䠼

(฀˃ᴗ˂฀)

(ˊᗜˋ)

󲔜 ⍵󲔜 

ʕ•̫•ʔ

䂄 ꒰*´Ⴣ`*꒱̌

Ò8 ũ8 㼌䡝㻕

󰗚(󲔕󳙽/ᾥ’ 󲔕󳙽󲔕󳙾󰗚

:;(∩㼿ũ`∩);:

(σ 䡡Ѧ䡡)σ

(@_@)

 



Figure 5: Comparison of happiness evaluation values given by males and females (131 emoticons).

Figure 6: Example of Evaluation value to a emoticon.

REFERENCES

Arakawa, A., Takehara, T., and Suzuki, N. (2006). Effect of

messages with emoticons on regulating the receivers’

emotional intensity. Japanese Journal of Research on

Emotions, 13(1):22–29.

Emura, Y. and Seki, Y. (2012). Facemark recommendation

based on emotion, communication, and motion type

Figure 7: Clustering of the emoticons.

estimation in text (in Japanese). IPSJ SIG Technical

Report, 2012(1):1–7.

Kawakami, M. (2008). Database about emotion and emp-

hasis of emoticons (in Japanese). The Human Science

Research Bulletin, 7:67–82.

Nakamaru, S. (2002). The effect of emoticon t o degree of

reliability of text (in Japanese). Komazawa Sociology

Research, 34:91–114.

Ono, S., Harada, S., and Tokuda, K. (2003). The use of

emoticon among student at college i: Meaning recog-

nition of emoticon (in Japanese). Bulletin of Japanese

Emoticon Recommendation System Reﬂecting User Individuality - A Preliminary Survey of Emoticon Use

463

Association of Educatinal Psychology, (45):267.

Park, J., Barash, V., Fink, C., and Cha, M. (2013). Emoti-

con style: Interpreting differences in emoticons across

cultures. In ICWSM.

Riva, G. (2002). The sociocognitive psychology of

computer-mediated communication: The present and

future of technology-based interactions. CyberPsy-

chology & Behavior, 5(6):581–598.

Urabe, Y., Rafal, R., and Araki, K. (2013). Emoticon re-

commendation for japanese computer-mediated com-

munication. In Semantic Computing (ICSC), 2013

IEEE Seventh International Conference on, pages 25–

31. IEEE.

Walther, J. B. and DAddario, K. P. (2001). The impacts

of emoticons on message interpretation in computer-

mediated communication. Social Science Computer

Review, 19(3):324–347.

Wolf, A. (2000). Emotional expression online: Gender dif-

ferences in emoticon use. CyberPsychology & Beha-

vior, 3(5):827–833.

ICAART 2017 - 9th International Conference on Agents and Artiﬁcial Intelligence

464