www.ijemr.

net ISSN (ONLINE): 2250-0758, ISSN (PRINT): 2394-6962

SPECIAL ISSUE (ACEIT-2018)

Second International Conference on Advancement in Computer Engineering
and Information Technology
Organized By: Department of Computer Science & Engineering,
Integral University, Lucknow, INDIA
Page Number: 26-30

Sarcasm Analysis Using Social Media: A Literature Review

Shagufta Masroor1, Dr. Mohd. Shahid Husain2
1,2
Integral University, Lucknow, INDIA

ABSTRACT Definition of sarcasm is defined differently in

An expression of resentment, criticism, and mockery different dictionaries, Macmillan English dictionary, defines
by using harsh words that intended to offend someone or sarcasm as “the act of saying or writing the opposite of what
something can be called Sarcasm statements. It is one of the you mean, or of speaking in a way intended to make
most difficult problems faced by authors while doing sentiment someone else feel stupid or show them that you are angry”
analysis. The most difficult sentences are those whose literal
[16], The Random House dictionary [17], defines sarcasm as
meaning differs from the emotion of the concerned individual.
Sarcastic comments can be done vocally as well; the “a harsh or bitter derision or irony” or “a sharply ironical
work is done under Audio mining Techniques. In studying taunt; sneering or cutting remark”. The Collins English
social media websites, Twitter comes up as a core site for all dictionary [18] states it as “mocking, contemptuous, or ironic
types of users. The sentiments of these people are studied in language intended to convey insults or scorn”. There are
order to extract the real emotions from their non-sarcastic as many definitions of sarcasm defined in different dictionaries,
well as sarcastic tweets. Since this is a survey paper, we are including Merriam-Webster dictionary, [19] which define
trying our best to show up the valuable work of various authors sarcasm as “a mode of satirical wit depending for its effect
who have invested their time in studying the sentiments of on bitter, caustic, and often ironic language that is usually
different people and their sarcastic comments.
directed against an individual”. Also, there are various
Keywords-: Sarcasm Detection, Twitter, Tweets, sentiment
approaches which are being used in recognizing sarcasm.
analysis, Hashtags Such as various researchers used behavioral modeling
approach [4], pattern-based approach [12], rule-based
approach [13] etc. sarcasm detection dataset used in the
research papers varies and includes Facebook, Twitter,
I. INTRODUCTION Blogs, movie reviews, feedbacks by customers, etc. and
types of techniques used includes machine learning [1], [2],
Sarcasm is a type of emotion which people share [6], [5], supervised learning [7] etc.
with each other and whose literal meaning differs from the In this paper, we discuss the sarcasm detection
actual meaning. If people make sarcastic comments openly in mechanism proposed by different researchers. The rest of the
front of one another, it can be easily understood and paper is organized as follows: in Section II, related work is
evaluated, but digitally recognizing the actual meaning of the presented. Types of sarcasm are given in Section III. Feature
context is quite a difficult task. As social media is a free set analysis is discussed in Section IV. Various approaches
platform available for everyone to express their true for sarcasm detection are given in Section V. Comparative
emotions, they feel and write their thoughts in the best table of some research papers is given in Section VI.. Finally,
possible way. Example: “Nothing I love more than a in Section VII, we conclude with some final remarks and
crowded library with no vacant seats” :-) #sarcasm. Here the present some future works.
example has no negative word yet the feelings of the person
are captured correctly by everyone except the digitally which
will convey the literal meaning of the sentence. In this II. TYPES OF SARCASM
example, people use to express the positive sentiment (love)
but overall tweet reflects negative sentiment toward the The study of work done by the researchers in this
library. domain reflects that there are different categories of sarcastic
comments. The types of sarcasm being studied by

26 Copyright © 2018. IJEMR. All Rights Reserved.

www.ijemr.net ISSN (ONLINE): 2250-0758, ISSN (PRINT): 2394-6962

researchers are based on distinct features and structure of the taking place from the last tweet is included in the current
text. tweet (for eg, positive ! Negative, negative! Positive) as a
Reganti et al [10] have discussed four types of feature (1 feature).Sarcasm as a complex form of expression
satire that are present in the English language that are: Readability
Exaggeration: It is to enlarge, accentuate or portray As sarcasm is widely standardized to be hard to read
something beyond mundane limits so as to highlight faults, and understand, the author adapts identical readability tests
e.g.: “I’m super exhilarated today!! So much that I’d kill to measure the degree of complexity and understandability of
myself” the tweet.
 Incongruity:  Sarcasm as a means of assigning emotion
It signifies to present things that are malapropos or 1. Mood
preposterous in cognation to its circumventions, e.g.: “The Mood represents the utilizer's state of emotion.
back camera of the phone is so good that I can capture every Naturally, the mood of the individual may be indicative of
atom of scenery”. his propensity to utilize mordancy; if the utilizer is in a
 Reversal: deplorable (negative) mood, he could opt to express it in the
Which is presenting diametrical to what is to form of a sarcastic tweet. Consequently, his mood is judged
authentically convey by the user, e.g.: “I’m profoundly utilizing sentiment expressed in his past tweets... They
disappointed. Not as expected! It’s just astounding how the captured the mood utilizing last tweets.
flash works!” 2. Affect and sentiment
 Parody: Sarcasm is a coalescence of affect and sentiment
Which is to imitate the demeanor/slang and/or style expression, and consequently, the effect and the sentiment
of some person, place or thing. expressed in mordant tweets are examined.
Jena et al [2] defines Six types of sarcasm which are  Sarcasm as the possible function of ease
discussed that occur in the text- 1. Familiarity with the language
T1- Contrasting comments between positive view and Naturally, one would expect a utilizer who utilizes a
negative view. form of language as intricate as sarcasm to have good
T2- Contrasting comments between negative view and command over the language. Consequently, the author
positive view. measures the utilizer’s language skills with features that are
T3- Fact Negation – i.e. text contradicting a fact. inspired by standardized language proficiency Cloze tests. In
T4- Likes and Dislikes Prediction – i.e. behavior based. cloze tests, proficiency is evaluated predicated on lexicon,
T5- Lexical Analysis – i.e. sarcasm hashtag based. grammar, dictation, and reading levels.
T6- Temporal Knowledge i.e. Extracting tweets contradicting 2. Familiarity with the environment
facts about the event. The users can express sarcasm better when they are
Liu et al [4] introduced the concept of Sarcasm which is well acquainted with their environment. Just like people are
discussed below- less liable to utilize sarcasm in an incipient, unfamiliar
 Contrast of various sentiments setting, users take time to get habituated with Twitter afore
 A complex form of expression posting mordant tweets. Author measure a utilizer's
 Means of assigning emotion familiarity with Twitter in terms of his utilization familiarity,
 Possible function of ease parlance familiarity, and convivial activity.
 A form of written manifestation  Sarcasm as a form of written manifestation
 Sarcasm as a contrast of various sentiments 1. Prosodic variations
1. Contrasting connotations The users often reiterate letters in words to stress
A mundane denotes of expressing mordancy is to and over accentuate certain components of the tweet (for
utilize words with distinct construals within the same tweet. example, sooooo, awesomeeee) to denote that they mean
In the example, I dote getting spam emails!, spam diametrical to what is indited.
conspicuously has a negative connotation while love is 2. Structural variations
Overwhelmingly positive and to model such occurrences, It is observed that mordant tweets sometimes have a
they construct features predicated on affect and sentiment certain construction wherein the commentor's views are
scores. verbalized in the first few words of the tweet, while in the
2. Contrasting present with the past later components, an explication of a particular scenario is
Sometimes, the utilizer may set up a contrasting put forth, e.g., I dote it when my friends ignore me.
context in his antecedent tweet and then, opt to utilize a
mordacious verbalization in his current tweet. To model such
demeanor, they obtain the sentiment expressed by the utilizer
(i.e., positive, negative, neutral) in the antecedent tweet and
the current tweet. Then, the type of sentiment transition

27 Copyright © 2018. IJEMR. All Rights Reserved.

www.ijemr.net ISSN (ONLINE): 2250-0758, ISSN (PRINT): 2394-6962

III. FEATURE EXTRACTION and tweet based feature. However exclaimed on some other
PROCESSES particular type of social graph-based features such as local
clustering coefficient, between centrality, and distance,
To analyze the sarcasm in texts posted on social media, according to them it is not possible for now. Tayal et al [7]
researchers suggested different features which can be used to worked on features such as used punctuation marks! ,?
distinguish the sarcastic comments from normal text. Characters in the sentence, #sarcasm tag and #irony,
Lunando et al[5] have given several features which emoticons and adjective and verb in conjunction with ! on
are taken from the pre-processed text which are Twitter. Along with these, they also considered emoticons
Unigram(According to the author unigram is more suitable with verbs and adjectives since the mere use of emoticons
for Indonesian social media text since the grammars used in like “:P” and “;)”. According to the author the emoticons can
Indonesian social media texts are various and informal.), also be used for humor sentiment analysis.
Negativity (This feature represents the percentage of the Liu et al[4] obtained a set of sarcastic tweets using
negative sentiment in the topic of the text message.) ,Number keywords #sarcasm and #not, altering out non-English tweets
of interjection words (This feature shows the number of and retweets they limited their analysis to tweets which
interjection words from the text such as “aha”, “bah”, “nah”, contain more than three words as they found that tweets with
“wew”, “wow”, “yay”, “uh”, etc.). fewer words were very noisy or clichéd (e.g., yeah, right!
Bouazizi et al [3] extract 2 sets of features: one #sarcasm). Along with these various kinds of features, Rosso
qualified as “non-textual”( Non-textual features: From the et al[6] focused on various tweets which have #irony , not,
“raw” tweets we first extract 6 features by counting the and #sarcasm words in their context. The main concern of
number of positive and negative Hashtags, that of positive the author is to differentiate the ironic tweets from sarcastic
and negative Emoticons, and that of positive and negative ones.
slang words.) and one qualified as “textual”( the tag “NOT ”
(e.g., “not”, “never”, etc.) , positive words such as love, IV. VARIOUS APPROACHES FOR
happy, and negative emotional content such as hate, sad etc.), SARCASM ANALYSIS
Jena et al [2] used these features, lexical, pragmatics and
hyperbole features to recognize sarcasm in text and also A. Time-Based Approach-
discussed the role of different lexical factors, such as lexical This is an approach which chao et al [1] used in
feature includes text properties such as uni-gram, bi-gram, order to predict the behavior of the sarcastic tweets done
tri-gram, and n-gram. before and after a certain fixed time and studied the behavior.
Chao et al[1] worked on various types of features B. Classification Based Approach-
such as age, no. of followers, following/friends, favorites, This approach is commonly used to identify the
lists, tweets sent, retweets done by n twitter users, hashtags target word or sentence as sarcastic or literal. Different
included, user mentions included, URLs included, no. of classifiers mainly machine learning classifiers such as SVM,
characters and no. of digits in a particular tweet also they Naïve Bayes, Maximum entropy, etc. are used, Authors [3],
broadly classified into two categories i.e. user based feature [5], [6] who are using this classification approach.

Table I Different Features extracted in some papers-

Sr. No. Author Types of features being worked on

1. Lunando et al [5] Unigram, Negativity, No. of interjection words.

Non textual(positive and negative (Hashtags, emoticons, slang words)), textual(the tag
2. bouazizi et al [3] “NOT ” (e.g., “not”,
“never”, etc.) , positive words such as love, happy, and negative emotional content such
as hate, sad etc. )

3. Jena et al [2] lexical, pragmatics and hyperbole

age, no. of followers, following/friends, favorites, lists, tweets sent, retweets done by n
4. Chao et al [1] twitter users,
hashtags included, user mentions included, URLs included, no. of characters and no. of
digits in a particular
Tweet

28 Copyright © 2018. IJEMR. All Rights Reserved.

www.ijemr.net ISSN (ONLINE): 2250-0758, ISSN (PRINT): 2394-6962

Used punctuation marks ! , ? , Characters in the sentence, #sarcasm tag and #irony,
5. Tayal et al[7] emoticons and adjective
and verb in conjunction with ! in Twitter

6. Liu et al[4] a set of sarcastic tweets using keywords #sarcasm and #not

7. Rosso et al[6] #irony ,not, and #sarcasm

Table II Glimpse of Algorithms and Techniques used for classification of tweets-

Author Classification Technique/Algorithm used

Lunando et al[5] Naïve Bayes, Maximum Entropy, and Support Vector Machine.

Bouazizi et al[3] Naive Bayes, Support Vector Machine (SVM), and Maximum Entropy classifiers.

Jena et al[2] PBLGA(Parsing Based Lexicon Generation Algorithm),

Chao et
al[1] Trend Micro’s WRS, manual inspection.

Liu et al[4] SCUBA, Random Classifier, Majority Classifier, n-gram classifier.

Table III. Comparative analysis of some research papers mentioned.

Author Approach used problem solved Tools used Advantages

Trend Micro’s Web Found that classifiers’ ability

Chao et al Machine Learning Detection of Spam Reputation to
detect Twitter spam reduced
Tweets System when
in a near real-world scenario
since
the imbalanced data has
brought
bias.

Natural Language Sarcasm Detection

Jena et al Processing in TEXTBLOB(python based Achieved 0.89, 0.81 and 0.84
Tweets package) precision, recall and f score
,respectively in tweets with
sarcastic hashtag and 0.64,
0.75
and 0.69 precision, recall and
f
score respectively in tweets
without sarcastic hashtag.

SCUBA (n-gram,
Liu et al Streaming API, Behavioral Sarcasm Detection behavioral Using SCUBA, social media
29 Copyright © 2018. IJEMR. All Rights Reserved.
www.ijemr.net ISSN (ONLINE): 2250-0758, ISSN (PRINT): 2394-6962

teams can better detect

Modeling, modeling Augmented sarcasm
and deliver appropriate
Approaches- Framework) responses
- Contrast to sarcastic tweets.
- Hybrid

Such Linguistic Analysis of

Rosso et Machine Learning Differentiation N-grams, TwitIE uses Penn not
involving automatic
Al between Irony and Treebank Project tagset. techniques is
a crucial step in works that
Sarcasm Tweets deals
with distinguishing between
irony
and sarcasm.

By using the sentiment score,

Lunando Machine Learning Indonesian Sarcasm SentiWordNet(Unigram the
Feature extraction accuracy of the general
et al and sentiment Component) sentiment
analysis analysis is improved for about
4%.And then, the result
shown
that the negativity feature for
detecting sarcasm are quite
effective since it increased the
accuracy by 6%.

The accuracy of negative

Bouazizi Machine Learning Effect of Sarcastic Key Performance tweets
increases after adding
et al tweets on sentiment Indicator(KPI) Sarcasm
analysis Related feature.

Detection of By Using a supervised

Tayal et al Supervised Learning polarity Supervised Learning approach
we can determine the polarity
of sarcastic Political and
Tweets predict results.

V. CONCLUSION AND FUTURE Micros WRS, manual inspection etc. It has been found that
WORK various techniques applied for sarcasm analysis are domain and
language specific.
In this paper we did a detailed study of sarcasm Taking a finite data set and perform sarcasm detection
detection from various Research papers, we found that the analysis is one of the limitation found, and work on dynamic
data set selected by them varies in each research. Giving data should be focused in future also we found that the sarcasm
the possibility to focus more on a different data set and detection does not limit till text mining but other areas can also
gives a great opportunity in data mining field. Secondly, be focused on such as Audio mining, video mining and
we have discussed various techniques to carry out sarcasm recognition etc. and work in this field can be focused further.
analysis on Twitter data including Naïve Bayes, Maximum
Entropy, and Support Vector Machine, PBLGA (Parsing
Based Lexicon Generation Algorithm), SCUBA, Trend

30 Copyright © 2018. IJEMR. All Rights Reserved.

www.ijemr.net ISSN (ONLINE): 2250-0758, ISSN (PRINT): 2394-6962

REFERENCES International Conference on Information Technology

(ICIT),pp. 703-709 IEEE.
[1] Chao chen, et al,” A Performance Evaluation of [12] Mondher Bouazizi; Tomoaki Otsuki Ohtsuki,” A Pattern-
Machine Learning Bsed Streaming SpamTweets Based Approach for Sarcasm Detection on Twitter”, IEEE
Detection” IEEE Transactions on Computational Social Access, , Vol: 4, pp. 5477 - 5488 ,IEEE Journals & Magazines,
Systems Vol. 2, no. 3 pp: 65 – 76, 2015. 2016
[2] Santosh Kumar Bharti; Korra Sathya Babu; Sanjay [13] Satoshi Hiai; Kazutaka Shimada,” A Sarcasm Extraction
Kumar Jena,“Parsing- based sarcasm sentiment Method Based on Patterns of EvaluationExpressions”, 2016 5th
recognition in Twitter data” IEEE/ACM International IIAI International Congress on Advanced Applied Informatics
Conference on Advances in Social Networks Analysis and (IIAI-AAI), pp. 31 – 36,2016
Mining (ASONAM) pp: 1373 – 1380; 2015 [14] Anukarsh,G.Prasad; S.Sanjana; Skanda, M. Bhat;
[3] Mondher Bouazizi; Tomoaki Ohtsuki, “Opinion mining B.S.Harish,” Sentiment analysis for sarcasm detection on
in Twitter: How to make use of sarcasm to enhance streaming short text data”, 2017 2nd International Conference
sentiment analysis”in 2015 IEEE/ACM International on Knowledge Engineering and Applications (ICKEA), pp.1-5,
Conference on Advances in Social Networks Analysis and IEEE publications, 2017
Mining (ASONAM) pp. 1594 – 1597, 2015 [15] Abhinav Mathur; Vikas Saxena; Sandeep K Singh,” Und
[4] Ashwin Rajadesingan, Reza Zafarani, Huan erstanding sarcasm in speech using mel-frequency cepstral
Liu,”Sarcasm Detection on Twitter: A Behavioral coefficent” 2017 7th International Conference on Cloud
Modelling Approach”, Proceedings of the Eighth ACM Computing, Data Science & Engineering-Confluence, pp
International Conference on Web Search and Data Mining, 728-732,IEEE Publications, 2017
pp. 97-106, ACM,2015 [16] http://www.macmillandictionary.com/.
[5]Edwin Lunando; Ayu Purwarianti,” Indonesian soci al [17] http://www.thefreedictionary.com/.
media sentiment analysis with sarcasm detection”, I [18] http://www.collinsdictionary.com/.
nternational Conference on Advanced Computer Science [19] http://www.merriam-webster.com/.
and Information Systems (ICACSIS),pp: 195 – 198 IEEE,
2013.
[6] Maria Khokhlova; Viviana Patti; Paolo Rosso,” Disti
nguishing between irony and sarcasm in social media
texts: Linguistic observations”In 2016 International
FRUCT Conference on Intelligence, Social Media and
Web (ISMW FRUCT)pp. 1 – 6, IEEE publications, 2016
[7] D. K. Tayal; Sumit Yadav; Komal Gupta; Bhawna
Rajput; Kiran Kumari,” Polarity detection of sarcastic
political tweets” International Conference on Computing
for Sustainabl Global Development (INDIACom),pp: 625
– 628, 2014 ,IEEE
[8] Dana Al-Ghadhban; Eman Alnkhilan; LammaTatwany;
Muna Alrazgan,” Arabic sarcasm detection in Twitter”,
2017 International Conference on Engineering & MIS
(ICEMIS), pp.1-7, 2017, IEEE
[9] Anand kumar D. Dave; Nikita P. Desai,” A
comprehensive study of classification techniques for
sarcasmdetection on textual data”, 2016
International Conference on Electrical, Electronics, and
Optimization Techniques (ICEEOT),pp, 1985 – 1991,
2016,IEEE
[10] Aishwary N. Reganti; Tushar Maheshwari; Upendra
Kumar; Amitava Das; Rajiv Bajpai,” Modeling Satire in
English Text for Automatic Detection”. 2016 IEEE 16th
International Conference on Data Mining Workshops
(ICDMW),pp.970 – 977,2016,IEEE
[11] Mohd Suhairi Md Suhaimin; Mohd Hanafi Ahmad
Hijazi; Rayner Alfred; Frans Coenen,” Natural language
processing based features for sarcasm detection: An
investigation using bilingual social media texts”, 2017 8th

31 Copyright © 2018. IJEMR. All Rights Reserved.