Fracking Sarcasm using Neural Network
Aniruddha Ghosh
University College Dublin
aniruddha.ghosh@ucdconnect.ie
Tony Veale
University College Dublin
tony.veale@ucd.ie
Abstract
Precise semantic representation of a sentence
and definitive information extraction are key
steps in the accurate processing of sentence
meaning, especially for figurative phenom-
ena such as sarcasm, Irony, and metaphor
cause literal meanings to be discounted and
secondary or extended meanings to be inten-
tionally profiled. Semantic modelling faces
a new challenge in social media, because
grammatical inaccuracy is commonplace yet
many previous state-of-the-art methods ex-
ploit grammatical structure. For sarcasm de-
tection over social media content, researchers
so far have counted on Bag-of-Words(BOW),
N-grams etc. In this paper, we propose a
neural network semantic model for the task
of sarcasm detection. We also review se-
mantic modelling using Support Vector Ma-
chine (SVM) that employs constituency parse-
trees fed and labeled with syntactic and se-
mantic information. The proposed neural net-
work model composed of Convolution Neu-
ral Network(CNN) and followed by a Long
short term memory (LSTM) network and fi-
nally a Deep neural network(DNN). The pro-
posed model outperforms state-of-the-art text-
based methods for sarcasm detection, yielding
an F-score of .92.
1 Introduction
Figurative language, such as metaphor, irony and
sarcasm, is a ubiquitous aspect of human communi-
cation from ancient religious texts to modern micro-
texts. Sarcasm detection, despite being a well-
studied phenomenon in cognitive science and lin-
guistics (Gibbs and Clark, 1992; gib, 2007; Kreuz
and Glucksberg, 1989; Utsumi, 2000), is still at its
infancy as a computational task. Detection is diffi-
cult because literal meaning is discounted and sec-
ondary or extended meanings are instead intention-
ally profiled. In social contexts, one’s ability to
detect sarcasm relies heavily on social cues such
as sentiment, belief, and speaker’s intention. Sar-
casm is mocking and often involves harsh delivery
to achieve savage putdowns, even though it can be
also crafted more gently as the accretion of polite-
ness and the abatement of hostility around a criti-
cism (Brown and Levinson, 1978; Dews and Win-
ner, 1995). Moreover, sarcasm often couches crit-
icism within a humorous atmosphere (Dews and
Winner, 1999). (Riloff et al., 2013) addressed one
common form of sarcasm as the juxtaposition of a
positive sentiment attached to a negative situation,
or vice versa. (Tsur et al., 2010) modeled sarcasm
via a composition of linguistic elements, such as
specific surface features about a product, frequent
words, and punctuation marks. (Gonz
´
alez-Ib
´
anez et
al., 2011) views sarcasm as a conformation of lex-
ical and pragmatic factors such as emoticons and
profile references in social media. Most research ap-
proaches toward the automatic detection of sarcasm
are text-based and consider sarcasm to be as a func-
tion of contrasting conditions or lexical clues. Such
approaches extract definitive lexical cues as features,
where the linguistic scale of features is stretched
from words to phrases to provide richer contexts for
analysis. Lexical feature cues may yield good re-
sults, yet without a precise semantic representation
of a sentence, which is key for determining the in-
tended gist of a sentence, robust automatic sarcasm