Automatic Sarcasm Detection: A Survey

Transcription

1 Automatic Sarcasm Detection: A Survey Aditya Joshi 1,2,3 Pushpak Bhattacharyya 2 Mark James Carman 3 1 IITB-Monash Research Academy, India 2 IIT Bombay, India, 3 Monash University, Australia {adityaj,pb}@cse.iitb.ac.in, mark.carman@monash.edu Sarcasm as a linguistic phenomenon has been widely studied. Before we proceed to approaches for sarcasm detection, we present an introduction to sarcasm studies in linguistics. Campbell and Katz (2012) state that sarcasm occurs along several dimensions namely failed expectation, pragmatic insincerity, negative tension, and presence of a victim. Eisterhold et al. (2006) state that sarcasm is peculiar because of the response it elicits. They observe that the responses to sarcasm may be: laughter, zero response, smile, sarcasm (in return), a change of topic (because the listener was not happy with the caustic sarcasm), literal reply and non-verbal reactions. Camp (2012) show that there are four types of sarcasm: (1) Propositional: Such sarcasm appears to be a nonsentiment proposition but has an implicit sentiment involved, (2) Embedded: Such sarcasm has an embedarxiv: v1 [cs.cl] 10 Feb 2016 Abstract Automatic detection of sarcasm has witnessed interest from the sentiment analysis research community. With diverse approaches, datasets and analyses that have been reported, there is an essential need to have a collective understanding of the research in this area. In this survey of automatic sarcasm detection, we describe datasets, approaches (both supervised and rule-based), and trends in sarcasm detection research. We also present a research matrix that summarizes past work, and list pointers to future work. 1 Introduction Sarcasm is a peculiar form of sentiment expression, where surface sentiment differs from the implied sentiment. The Free Dictionary 1 defines sarcasm as a form of verbal irony that is intended to ridicule 2. Sarcasm is an often-quoted challenge for sentiment analysis (Liu, 2010) because sarcasm intends a negative sentiment, but a positive surface sentiment. This led to introduction of sarcasm detection as a research problem. Automatic sarcasm detection refers to computational approaches to detect sarcasm in text. This problem is hard because of nuanced ways in which sarcasm may be expressed. These nuances arise in several ways: general understanding ( I love to get ignored ), in terms of an author s background ( I love solving math problems all weekend ), in terms of a culture (popular use of honorifics in Chinese sarcasm), etc. Starting from the earliest known work by Tepperman et al. (2006) which deals with speech and text-related features, sarcasm detection research has seen wide interest. Following that, sarcasm detection from text has extended to different data forms (tweets, reviews, TV series dialogues), and spanned several approaches (rule-based, supervised, semi-supervised). This synergy has resulted in interesting innovations for automatic sarcasm detection. With such diversity of approaches, there is a need to have a collective understanding of the research in That sarcasm is a form of verbal irony explains the relationship between sarcasm and irony. Sarcasm is essentially hurtful or intended to be hurtful sarcasm detection. This paper collates past works in automatic sarcasm detection. Through tabular summarization, we provide views of the current state-of-art. One past work to summarize computational irony is by Wallace (2013). However, they focus on linguistic challenges of computational irony, and deliberate on possible future work. Their paper is, however, limited to linguistic challenges and theories in sarcasm. On the contrary, we focus on the computational angle, and present a survey of computational sarcasm detection techniques. Our paper looks at sarcasm detection research in terms of four parameters: datasets, approaches, trends and issues. We present an illustration that shows current research, and a detailed matrix that describes papers dealing with sarcasm detection. Both will serve as useful resources for future work. We also prescribe future directions based on current papers. The rest of the paper is organized as follows. Section 2 describes sarcasm studies in linguistics. Section 3 presents different problem definitions for sarcasm detection. Sections 4 and 5 discuss datasets and approaches reported for sarcasm detection. Section 6 highlights trends underlying sarcasm detection research, while Section 7 discusses key recurring issues in this research. Finally, Section 8 concludes the paper. 2 Sarcasm in Linguistics

2 ded sentiment incongruity in the form of words and phrases itself, (3) Like-prefixed: A like-phrase provides an emphatic implied denial of the argument being made before the sarcasm is expressed, and (4) Illocutionary: This sarcasm involves non-textual clues that indicate an attitude opposite to a sincere utterance. Therefore, in such cases, prosodic variations play a role in sarcasm expression. Several theories have been proposed to represent sarcasm. Some of them are: 1. Situational disparity theory: According to Wilson (2006), sarcasm arises when there is situational disparity between text and a contextual information. 2. Negation theory of sarcasm: Giora (1995) state that irony/sarcasm is a form of negation in which an explicit negation marker is lacking. In other words, when one expresses sarcasm, a negation is intended, without putting a negation word like not tuple representation: Ivanko and Pexman (2003) define sarcasm as a 6-tuple consisting of <S, H, C, u, p, p > where: S = Speaker H = Hearer/Listener C = Context u = Utterance p = Literal Proposition p = Intended Proposition The tuple can be read as Speaker S generates an utterance u in Context C meaning proposition p but intending that hearer H understands p. Consider the following example. If a teacher says to a student, That s how assignments should be done! and if the student knows that (s)he has barely completed the assignment, the student would understand the sarcasm. In context of the 6-tuple above, the properties of this sarcasm would be: S: Teacher H: Student C: The student has not completed his/her assignment. u: That s how assignments should be done! p: You have done a good job at the assignment. p : You have done a bad job at the assignment. This linguistic background helps to establish the challenges for sarcasm detection. In the context of the theories described here, some challenges are: (1) Identification of common knowledge, (2) Intent to ridicule, (3) Speaker-listener context. As we will see in the next section, these challenges have been partly handled by different reported works in automatic sarcasm detection. 3 Problem Definition We now look at how computational sarcasm detection has been defined, in past work. The most common formulation for sarcasm detection is as a classification task. Given a piece of text, the goal is to predict whether or not it is sarcastic. However, alternate definitions are also possible. Understanding the relationship between sarcasm, irony and humor, Barbieri et al. (2014b) consider labels for the classifier as: politics, humor, irony and sarcasm. Reyes et al. (2013) use a similar formulation and provide pair-wise classification performance for these labels. Even in the context of classification, there have been interesting variations in the data units that will be annotated and hence, classified. Tepperman et al. (2006) is an early paper in sarcasm detection that looks at occurrences of a common sarcastic phrase yeah right and classifies each occurrence of yeah right as sarcastic or not. Veale and Hao (2010) annotate similes such as as excited as a patient before a surgery with sarcastic or not labels. Other formulations for sarcasm detection have also been reported. Ghosh et al. (2015b) model sarcasm detection as a sense disambiguation task. They state that a word may have a literal sense and a sarcastic sense. Their goal is to identify the sense of a word in order to detect sarcasm. Wang et al. (2015) formulate the problem as a sequence labeling problem. For a sequence of tweets in a conversation, they estimate the most probable sequence of three labels: happy, sad and sarcastic. Table 1 shows a matrix that summarizes sarcasm detection papers. The matrix is a useful resource to understand current work in sarcasm detection. The rest of this paper elaborates different columns of this matrix, i.e., we compare past work across five parameters: datasets, approach, annotation technique, features and extra-textual context, in the next sections. A note on languages Most research in sarcasm detection exists for English. Liu et al. (2014) use Chinese and English social media content. Barbieri et al. (2014a) present a first approach to detect sarcasm in Italian tweets. Ptácek et al. (2014) deal with Czech along with English tweets. Liebrecht et al. (2013) explore sarcasm detection for Dutch. 4 Datasets 4.1 Short text Social media makes available several forms of data. Due to availability of the Twitter API and popularity of twitter as a medium, sarcasm-labeled datasets of tweets are popular. One approach to obtain labels for tweets is manual annotation. Riloff et al. (2013) use a dataset of tweets, manually annotated as sarcastic or not. Maynard and Greenwood (2014) study sarcastic tweets and their impact to sarcasm classification. They experiment with around 600 tweets which are marked for subjectivity, sentiment and sarcasm. Ptácek et al. (2014) present a dataset of 7,000 manually labeled tweets in Czech.

3 Datasets Approach Annotatn. Features Context Short Text Long Text Other Rule-based Semi-superv. (Kreuz and Caucci, 2007) (Tsur et al., 2010) (Davidov et al., 2010) (Veale and Hao, 2010) (González-Ibánez et al., 2011) (Reyes et al., 2012) (Reyes and Rosso, 2012) (Filatova, 2012) (Riloff et al., 2013) (Lukin and Walker, 2013) (Liebrecht et al., 2013) (Reyes et al., 2013) (Reyes and Rosso, 2014) (Rakov and Rosenberg, 2013) (Barbieri et al., 2014b) (Maynard and Greenwood, 2014) (Wallace et al., 2014) (Buschmeier et al., 2014) (Barbieri et al., 2014a) (Joshi et al., 2015) (Khattri et al., 2015) (Rajadesingan et al., 2015) (Bamman and Smith, 2015) (Wallace, 2015) (Ghosh et al., 2015a) (Hernández-Farías et al., 2015) (Wang et al., 2015) (Ghosh et al., 2015b) (Liu et al., 2014) (Bharti et al., 2015) (Fersini et al., 2015) (Bouazizi and Ohtsuki, 2015) (Muresan et al., 2016) Superv Manual Hashtag-based Other Unigram Sentiment Pragmatic Spc. Sarc. Patterns Other Author Conversation Other Table 1: Matrix summarizing sarcasm detection research; short text means there was a word limit imposed by the platform where the text was created (such as twitter s word limit of 140 characters); context refers to situations where context beyond target text and general understanding, was considered The second approach is the use of hashtag-based supervision. Many approaches use hashtags in tweets as indicators of sarcasm. The popularity of this approach can be attributed to various factors: (a) No one but the author of a tweet can determine if it was sarcastic. A hashtag is a label provided by authors themselves, (b) The approach allows creation of large-scale datasets. In order to create such a dataset, tweets containing particular hashtags are labeled as sarcastic. Davidov et al. (2010) use a dataset of tweets, which are labeled with hashtags such as #sarcasm, #sarcastic, #not, etc. González-Ibánez et al. (2011) also use hashtag-based supervision for tweets. However, they eliminate cases where the hashtag is a part of the running text. For example, #sarcasm is popular in india is eliminated. Reyes et al. (2012) use hashtag-based supervision for tweets. Reyes et al. (2013) use a dataset of tweets labeled as sarcastic or not, using hashtags. Barbieri et al. (2014a) introduce a dataset of 25K Italian tweets. Joshi et al. (2015) present a dataset of tweets

4 Text form Tweets Related Work Manual: (Riloff et al., 2013; Maynard and Greenwood, 2014; Ptácek et al., 2014) Hashtag-based: (Davidov et al., 2010; González-Ibánez et al., 2011; Reyes et al., 2012; Reyes et al., 2013; Barbieri et al., 2014a; Joshi et al., 2015; Ghosh et al., 2015a; Bharti et al., 2015; Liebrecht et al., 2013; Bouazizi and Ohtsuki, 2015; Wang et al., 2015; Barbieri et al., 2014b; Bamman and Smith, 2015; Fersini et al., 2015; Khattri et al., 2015; Rajadesingan et al., 2015) Reddits (Wallace et al., 2014; Wallace, 2015) Long text (Reviews, etc.) Other datasets (Lukin and Walker, 2013; Reyes and Rosso, 2014; Reyes and Rosso, 2012; Buschmeier et al., 2014; Liu et al., 2014; Filatova, 2012) (Tepperman et al., 2006; Kreuz and Caucci, 2007; Veale and Hao, 2010; Rakov and Rosenberg, 2013; Ghosh et al., 2015b) Table 2: Summary of Sarcasm-labeled Datasets labeled with hashtag-based supervision. Ghosh et al. (2015a) use hashtag-based dataset of tweets trial, 4000 development and 8000 test tweets were introduced. Bharti et al. (2015) use a dataset of 50K tweets marked using hashtags. Liebrecht et al. (2013) identify how the hashtag #not is a popular modern form of expressing sarcasm. The dataset involves tweets that contain such a hashtag. Bouazizi and Ohtsuki (2015) use Sentiment140 dataset of tweets. Wang et al. (2015) use a dataset of tweets marked as happy, sad and sarcastic. The tweets are labelled using a supervised classifier since the main task is to detect sequence of labels. Barbieri et al. (2014b) create a dataset using hashtag-based supervision to create multiple labels: politics, sarcasm, humor and irony. Use of distant supervision using hashtags presents challenges, and may require quality control. To ensure quality, Bamman and Smith (2015) use hashtag-based dataset of tweets. The positive tweets are the ones containing #sarcasm the negative tweets are assumed to be the one not containing these labels. Fersini et al. (2015) use a dataset of 8K tweets marked using hashtags. To ensure quality, these tweets are additionally labelled by annotators. Twitter also provides access to additional context. Hence, in order to predict sarcasm, datasets may look at tweets by the same author. Khattri et al. (2015) use a dataset of tweets, introduced in the past. They look up the complete twitter timeline (limited to 3200 tweets, by Twitter) to establish context. Rajadesingan et al. (2015) use a dataset of tweets, labeled by hashtagbased supervision along with a historical context of 80 tweets per author. Using tweets in an ongoing conversation in order to predict sarcasm has not been explored yet. Other social media text includes reddits. Wallace et al. (2014) create a corpus of reddit posts of 10K sentences, from 6 reddit topics. Wallace (2015) present a dataset of reddit comments: 5625 sentences. 4.2 Long text Reviews and discussion forum posts have also been used for sarcasm-labeled datasets. Lukin and Walker (2013) use Internet Argument Corpus that marks a dataset of discussion forum posts with multiple labels one of the labels is related to sarcasm. Reyes and Rosso (2014) use a dataset of movies, book reviews and news articles marked with sarcasm and sentiment. Reyes and Rosso (2012) deal with products that saw a spate of sarcastic reviews all of a sudden. They consider reviews in total. Filatova (2012) use a dataset of around 1000 reviews, marked as sarcastic or not. Buschmeier et al. (2014) use 1254 Amazon reviews, out of which 437 are ironic. Tsur et al. (2010) consider a large dataset of amazon reviews. Liu et al. (2014) use a dataset from multiple sources such as Amazon, Twitter, Netease and Netcena. 4.3 Other datasets Several other novel datasets have been used. Tepperman et al. (2006) use 131 call center transcripts. Each occurrence of yeah right is marked as sarcastic or not. The goal is to find features in a transcripts that identify which yeah right is sarcastic. Kreuz and Caucci (2007) use 20 sarcastic excerpts and 15 non-sarcastic excerpts, which are marked by 101 students. The goal is to identify lexical indicators of sarcasm. Veale and Hao (2010) focus on identifying which similes are sarcastic. Hence, they first use google for the pattern * as a *. This results in 20,000 distinct similes which are then marked as sarcastic or not. Rakov and Rosenberg (2013) create a crowdsourced dataset of sentences from a MTV show, Daria. Ghosh et al. (2015b) use a crowdsourcing tool to obtain a non-sarcastic version of a sentence if applicable. For example Who doesn t love being ignored is expected to be corrected to Not many love being ignored. 5 Approaches Figure 1 shows key milestones in sarcasm detection. Only initial work in each area are indicated in the figure. Following foundational studies, there were many reported works on supervised/semi-supervised sarcasm classification that focused on using specific patterns or novel features. Then, as twitter emerged as a viable

5 Figure 1: Key Milestones in Sarcasm Detection Research: Illustration source of data, hashtag-based supervision became popular. Recently, using context beyond the text to be classified has become popular. This can be done in several ways, as we describe later. In general, approaches to sarcasm detection can be classified into: rule-based and statistical approaches. We look at these approaches in the next subsections. 5.1 Rule-based Approaches Rule-based approaches aim to identify sarcasm through specific evidences. As can be seen, these rules use peculiar indicators of sarcasm. Veale and Hao (2010) focus on identifying whether a given simile (of the form * as a * ) is intended to be sarcastic. They use Google search in order to determine how likely a simile is. They present a 9-step approach where at each step/rule, a simile is validated using the number of search results. A strength of this approach is that they present an error analysis corresponding to multiple rules. Maynard and Greenwood (2014) propose that hashtag sentiment is a key indicator of sarcasm. Hashtags are often used by authors to highlight the sarcasm, and hence, if the sentiment expressed by a hashtag does not agree with rest of the tweet, the tweet is predicted as sarcastic. They use a hashtag tokenizer to split hashtags made of concatenated words. Bharti et al. (2015) present two rule-based classifiers. The first uses a parse based lexicon generation algorithm that creates parse trees of sentences and identifies situation phrases that bear sentiment. If a negative phrase occurs in a positive sentence, it is predicted as sarcastic. The second algorithm aims to capture hyperboles by using interjection and intensifiers occur together. Riloff et al. (2013) present rule-based classifiers that look for a positive verb and a negative situation phrase in a sentence. They experiment with different configurations such as restricting the order of the verb and situation phrase. 5.2 Statistical Approaches Several statistical approaches have been explored for sarcasm detection. These approaches vary in terms of features and classifiers. We look at the two in forthcoming subsections Features Used In this subsection, we look at the set of features that have been reported for sarcasm detection. Table 3 summarizes some salient features for statistical approaches. In this subsection, we focus on features related to the text to be classified. Contextual features (i.e., features that use information beyond the text to be classified) are described in a latter subsection. Most approaches

6 use bag-of-words as features. However, in addition to these, there are peculiar features introduced in different works. We look at them hereafter. Tsur et al. (2010) use pattern-based features extracted from a sarcasm-labeled corpus. These patternbased features are included as numeric features that take three possible values: exact match, partial overlap and no match. González-Ibánez et al. (2011) use sentiment lexicon-based features. In addition, pragmatic features like emoticons and user mentions are also used. Reyes et al. (2012) use features related to ambiguity, unexpectedness, emotional scenario, etc. Ambiguity features cover structural, morpho-syntactic, semantic ambiguity, while unexpectedness features measure semantic relatedness. Riloff et al. (2013) use a set of patterns, specifically positive verbs and negative situation phrases, as features for a classifier. Liebrecht et al. (2013) introduce bigrams and trigrams as features. Reyes et al. (2013) explore skip-gram and character n-gram-based features. Maynard and Greenwood (2014) include seven sets of features. Some peculiar features include maximum/minimum/gap of intensity of adjectives and adverbs, max/min/average number of synonyms and synsets for words in the target text, etc. Apart from a subset of these, Barbieri et al. (2014a) use frequency and rarity of words as indicators. Buschmeier et al. (2014) experiment with a suite of features to incorporate ellipsis, hyperbole and imbalance. Joshi et al. (2015) use features corresponding to the linguistic theory of incongruity. The features are classified into two sets: implicit and explicit incongruitybased features. Ptácek et al. (2014) use word-shape and pointedness features given in the form of 24 classes. Rajadesingan et al. (2015) use extensions of words, number of flips, readability features in addition to others. Hernández-Farías et al. (2015) present features that measure semantic relatedness between words using Wordnet::similarity. Liu et al. (2014) introduce POS sequences and semantic imbalance as features. Since they also experiment with Chinese datasets, they use language-typical features like use of homophony, use of honorifics, etc. 5.3 Classifiers A variety of classifiers have been experimented for sarcasm detection. González-Ibánez et al. (2011) use SVM with SMO and logistic regression. Chi-squared test is used to identify discriminating features. Reyes and Rosso (2012) use Naive Bayes and SVM. They also show Jaccard similarity between labels and the features. Riloff et al. (2013) compare rule-based techniques with a SVM-based classifier. Liebrecht et al. (2013) use balanced winnow algorithm in order to determine high-ranking features. Reyes et al. (2013) use Naive Bayes and decision trees for multiple pairs of labels among irony, humor, politics and education. Bamman and Smith (2015) use binary logistic regression. Wang et al. (2015) use SVM-HMM in order to incorporate sequence nature of output labels in a conversation. Liu et al. (2014) compare several classification approaches including bagging, boosting, etc. and show results on five datasets. 5.4 Key Findings This section describes the observations that past sarcasm detection approaches make. These observations will be useful pointers for future work. González- Ibánez et al. (2011) show that unigram-based features outperform the use of a subset of words as derived from a sentiment lexicon. They compare the accuracy of the sarcasm classifier with the human ability to detect sarcasm. While the best classifier achieves 57.41%, the human performance for sarcasm identification is 62.59%. Reyes and Rosso (2012) observe that sentiment-based features are their top discriminating features. The logistic classifier in Rakov and Rosenberg (2013) results in an accuracy of 81.5%. Joshi et al. (2015) present an analysis of errors like incongruity due to numbers and granularity of annotation. Rajadesingan et al. (2015) show that historical features along with flip-based features are the most discriminating features. These are also the features that Khattri et al. (2015) use. Bharti et al. (2015) report a F-score of around 84%. 6 Trends in Sarcasm Detection Approaches Two research trends can be observed in approaches for sarcasm detection. They are (a) discovery of sarcastic patterns, and use of these patterns as features, and (b) use of contextual information i.e., information beyond the target text for sarcasm detection. We describe the two trends in detail in the forthcoming subsections. 6.1 Pattern discovery Discovering sarcastic patterns was an early trend in sarcasm detection. Several approaches dealt with extracting patterns that are indicative of sarcasm, or carry implied sentiment. These patterns may then be used as features for a statistical classifier, or as rules in a rule-based classifier. Tsur et al. (2010) extract sarcastic patterns from a seed set of labeled sentences. They first select words that either occur more than an upper threshold or less than a lower threshold. Among these words, identify a large set of candidate patterns. The patterns which occur discriminatively in either classes are then selected. Ptácek et al. (2014) also use a similar approach for Czech and English tweets. Riloff et al. (2013) hypothesize that sarcasm occurs due to a contrast between positive verbs and negative situation phrases. To discover a lexicon of these verbs and phrases, they propose an iterative algorithm. Starting with a seed set of positive verbs, they identify discriminative situation phrases that occur with these verbs in sarcastic tweets. These phrases are then used

7 (Tsur et al., 2010) (González-Ibánez et al., 2011) (Reyes et al., 2012) (Reyes and Rosso, 2012) (Riloff et al., 2013) (Liebrecht et al., 2013) (Reyes et al., 2013) (Barbieri et al., 2014b) (Buschmeier et al., 2014) (Barbieri et al., 2014a) (Joshi et al., 2015) (Rajadesingan et al., 2015) (Hernández-Farías et al., 2015) (Liu et al., 2014) (Ptácek et al., 2014) Salient Features Sarcastic patterns, Punctuations User mentions, emoticons, unigrams, sentiment-lexicon-based features Ambiguity-based, semantic relatedness N-grams, POS N-grams Sarcastic patterns (Positive verbs, negative phrases) N-grams, emotion marks, intensifiers Skip-grams, Polarity skip-grams Synonyms, Ambiguity, Written-spoken gap Interjection, ellipsis, hyperbole, imbalance-based Freq. of rarest words, max/min/avg # synsets, max/min/avg # synonyms Unigrams, Implicit incongruity-based, Explicit incongruity-based Readability, flips, etc. Length, capitalization, semantic similarity POS sequences, Semantic imbalance. Chinese-specific features such as homophones, use of honorifics Word shape, Pointedness, etc. Table 3: Summary of Features Used For Statistical Classifiers to identify other verbs. The algorithm iteratively appends to the list of known verbs and phrases. Joshi et al. (2015) adapt this algorithm by eliminating subsumption, and show that it adds value. Lukin and Walker (2013) begin with a seed set of nastiness and sarcasm patterns, created using Amazon Mechanical Turk. They train a high precision sarcastic post classifier, followed by a high precision nonsarcastic post classifier. These two classifiers are then used to generate a large labeled dataset from a bootstrap set of patterns. 6.2 Role of context in sarcasm detection A recent advancement in sarcasm detection is the use of context. The term context must viewed as any information beyond the text to be predicted, and beyond common knowledge. As we will see, this context may be incorporated in a variety of ways. In the rest of this section, we refer to the textual unit to be classified as target text. Wallace et al. (2014) is an annotation study that first highlighted the need of context for sarcasm detection. The annotators mark reddit comments with sarcasm labels. During this annotation, annotators often request for additional context in the form of reddit comments. The authors also present a transition matrix that shows how many times authors change their labels after the context is displayed to them. Following this observation and the promise of context for sarcasm detection, several recent approaches have looked at ways of incorporating it. The contexts that have been reported are of three types: 1. Historical context refers to text by this author in the past. Khattri et al. (2015) follow the intuition A tweet is sarcastic either because it has words of contrasting sentiment in it, or because there is sentiment that contrasts with historical sentiment. Historical tweets by the same author are considered as the context. Named entity phrases in the target tweet are looked up in the timeline of the author in order to gather the true sentiment of the author. This historical sentiment is then used to predict whether the author is likely to be sarcastic, given the sentiment expressed towards the entity in the target tweet. Rajadesingan et al. (2015) incorporate context about author using the author s past tweets. This context is captured as features for a classifier. The features deal with various dimensions. They use features about author s familiarity with twitter (in terms of use of hashtags), familiarity with language (in terms of words and structures), and familiarity with sarcasm. Bamman and Smith (2015) consider historical context in features such as historical salient terms, historical topic, profile info, historical sentiment (how likely is he/she to be negative), etc. 2. Conversation context refers to text in the conversation of which the target text is a part. This incorporates the discourse structure of a conversation. Bamman and Smith (2015) capture conversational context using pairwise Brown features between the previous tweet and the target tweet. In addition, they also use audience features. These are author features of the tweet author who responded to the target tweet. Joshi et al. (2015) show that concatenation of the previous post in a discussion forum thread along with the target post leads to an improvement in precision. Wallace (2015) look at comments in the thread structure to obtain context for sarcasm detection. To do so, they use the subreddit name, and noun phrases from the thread to which the target post belongs. 3. Topical context: This context follows the intuition that some topics are likely to evoke sarcasm

8 more commonly than others. Wang et al. (2015) use three types of context: history, conversation and topic. Topical context refers to how likely the topic of this tweet is, in terms of sarcastic occurrences in the past. 7 Issues in Sarcasm Detection This section describes three important issues related to current techniques in automatic sarcasm detection. The first set of issues deal with annotation: hashtag-based supervision, data imbalance and inter-annotator agreements. The second issue deals with a specific kind of features that have been used for classification: sentiment as a label. Finally, the third issue lies in the context of classification techniques where we look at how past works handle dataset skews. 7.1 Issues in Annotation In addition to specific issues described in the previous subsections, some issues in sarcasm annotation have been analyzed. Hashtag-based supervision: Although hashtagbased labeling can provide large-scale supervision, the quality of the dataset may be debatable. This is particularly true in case of use of #not to indicate insincere sentiment. (Liebrecht et al., 2013) show how #not can be used to express sarcasm - while the rest of the sentence is non-sarcastic. For example, I love cooking Italian food. #not. The speaker expresses sarcasm through #not. In most cases that use hashtag-based supervision, the hashtag is removed in the preprocessing step. This reduces the sentence above to I love cooking Italian food - which may not have a sarcastic interpretation at all. Data imbalance: Sarcasm is an infrequent phenomenon of sentiment expression. This skew also reflects in the datasets. Tsur et al. (2010) use a dataset with a small set of sentences are marked as sarcastic. 12.5% of tweets in the Italian dataset given by Barbieri et al. (2014a) are sarcastic. On the other hand, Rakov and Rosenberg (2013) present a balanced dataset of 15k tweets. Liu et al. (2014) focus on this imbalance in data, and present sarcasm detection that is robust to this imbalance. Inter-annotator agreement: Since sarcasm is a subjective phenomenon, the inter-annotator agreement values reported in past work are diverse. Tsur et al. (2010) indicate an agreement of Tepperman et al. (2006) observe an agreement of 52.73%. Fersini et al. (2015) report an agreement of Riloff et al. (2013) observe an agreement of Sarcasm before sentiment or sentiment before sarcasm? The motivation behind sarcasm detection is often pointed as sarcastic sentences misleading a sentiment classifier. However, several approaches use sentiment as an input to the sarcasm classifier. It must, however, be noted that these approaches require surface polarity the apparent polarity of a sentence. Bharti et al. (2015) is a rule-based approach that predicts a sentence as sarcastic if a negative phrase occurs in a positive sentence. As described earlier, Khattri et al. (2015) uses sentiment of a past tweet by the author to predict sarcasm. In a statistical classifier, surface polarity may be used directly as a feature use polarity of the tweet as a feature (Reyes et al., 2012; Joshi et al., 2015; Rajadesingan et al., 2015; Bamman and Smith, 2015). Reyes et al. (2013) capture polarity value in terms of two emotion dimensions: activation and pleasantness. Buschmeier et al. (2014) incorporate sentiment imbalance as a feature. Sentiment imbalance is a situation where star rating of a review disagrees with the surface polarity. Bouazizi and Ohtsuki (2015) cascade sarcasm detection and sentiment detection and observes an improvement of about 4% in accuracy when sentiment detection is aware of sarcastic nature. 7.3 Dealing with Dataset Skews Data skew in sarcasm-labeled datasets is a critical challenge to sarcasm classifiers. Liebrecht et al. (2013) state that detecting sarcasm is like a needle in a haystack. Some approaches focus on mitigating the effects of this skew. In Liu et al. (2014), a multi-strategy ensemble learning approach is used that uses ensembles and majority voting. Similarly, in order to deal with sparse features and skew of data, Wallace (2015) introduce a LSS-regularization strategy. Thus, they use a sparsifying L1 regularizer over contextual features and L2-norm for bag of word features. Liebrecht et al. (2013) report AUC for balanced as well as skewed datasets, to demonstrate the benefit of their classifier. 8 Conclusion & Future Directions In this paper, we presented a literature survey of current approaches for automatic sarcasm detection. We observe three trends in sarcasm detection research: semisupervised pattern extraction to identify implicit sentiment, use of hashtag-based supervision, and use of context beyond target text. We tabulated datasets and approaches that have been reported. We also highlight two key issues in sarcasm detection: the relationship between sentiment and sarcasm, and data skew in case of sarcasm-labeled datasets. Sarcasm detection research has flourished significantly in the past few years, necessitating a look-back at the overall picture that these individual works have led to. Based on our survey of these works, we propose following possible directions for future:

9 1. Implicit sentiment detection & sarcasm: Based on past work, it is well-established that sarcasm is closely linked to sentiment incongruity (Liebrecht et al., 2013). Several related works exist for detection of implicit sentiment in sentences, as in the case of The phone gets heated quickly v/s The induction cooktop gets heated quickly. This will help sarcasm detection, following the line of semisupervised pattern discovery. 2. Incongruity in numbers: Joshi et al. (2015) point out how numerical values convey sentiment and hence, is related to sarcasm. Consider the example of Took 6 hours to reach work today. #yay. This sentence is sarcastic, as opposed to Took 10 minutes to reach work today. #yay. 3. Coverage of different forms of sarcasm: In Section 2, we described four species of sarcasm: propositional, lexical, like-prefixed and illocutionary sarcasm. We observe that current approaches are limited in handling the last two forms of sarcasm: like-prefixed and illocutionary. Future work may focus on these forms of sarcasm. 4. Culture-specific aspects of sarcasm detection: As shown in Liu et al. (2014), sarcasm is closely related to language/culture-specific traits. Future approaches to sarcasm detection in new languages will benefit from understanding such traits, and incorporating them into their classification frameworks. References [Bamman and Smith2015] David Bamman and Noah A Smith Contextualized sarcasm detection on twitter. In Ninth International AAAI Conference on Web and Social Media. [Barbieri et al.2014a] Francesco Barbieri, Francesco Ronzano, and Horacio Saggion. 2014a. Italian irony detection in twitter: a first approach. In The First Italian Conference on Computational Linguistics CLiC-it 2014 & the Fourth International Workshop EVALITA, pages [Barbieri et al.2014b] Francesco Barbieri, Horacio Saggion, and Francesco Ronzano. 2014b. Modelling sarcasm in twitter, a novel approach. ACL 2014, page 50. [Bharti et al.2015] Santosh Kumar Bharti, Korra Sathya Babu, and Sanjay Kumar Jena Parsing-based sarcasm sentiment recognition in twitter data. In Proceedings of the 2015 IEEE/ACM International Conference on Advances in Social Networks Analysis and Mining 2015, pages ACM. [Bouazizi and Ohtsuki2015] Mondher Bouazizi and Tomoaki Ohtsuki Opinion mining in twitter how to make use of sarcasm to enhance sentiment analysis. In Proceedings of the 2015 IEEE/ACM International Conference on Advances in Social Networks Analysis and Mining 2015, pages ACM. [Buschmeier et al.2014] Konstantin Buschmeier, Philipp Cimiano, and Roman Klinger An impact analysis of features in a classification approach to irony detection in product reviews. ACL 2014, page 42. [Camp2012] Elisabeth Camp Sarcasm, pretense, and the semantics/pragmatics distinction*. Noûs, 46(4): [Campbell and Katz2012] John D Campbell and Albert N Katz Are there necessary conditions for inducing a sense of sarcastic irony? Discourse Processes, 49(6): [Davidov et al.2010] Dmitry Davidov, Oren Tsur, and Ari Rappoport Semi-supervised recognition of sarcastic sentences in twitter and amazon. In Proceedings of the Fourteenth Conference on Computational Natural Language Learning, pages Association for Computational Linguistics. [Eisterhold et al.2006] Jodi Eisterhold, Salvatore Attardo, and Diana Boxer Reactions to irony in discourse: Evidence for the least disruption principle. Journal of Pragmatics, 38(8): [Fersini et al.2015] Elisabetta Fersini, Federico Alberto Pozzi, and Enza Messina Detecting irony and sarcasm in microblogs: The role of expressive signals and ensemble classifiers. In Data Science and Advanced Analytics (DSAA), IEEE International Conference on, pages 1 8. IEEE. [Filatova2012] Elena Filatova Irony and sarcasm: Corpus generation and analysis using crowdsourcing. In LREC, pages [Ghosh et al.2015a] Aniruddha Ghosh, Guofu Li, Tony Veale, Paolo Rosso, Ekaterina Shutova, Antonio Reyes, and John Barnden. 2015a. Semeval-2015 task 11: Sentiment analysis of figurative language in twitter. In Int. Workshop on Semantic Evaluation (SemEval-2015). [Ghosh et al.2015b] Debanjan Ghosh, Weiwei Guo, and Smaranda Muresan. 2015b. Sarcastic or not: Word embeddings to predict the literal or sarcastic meaning of words. In EMNLP. [Giora1995] Rachel Giora On irony and negation. Discourse processes, 19(2): [González-Ibánez et al.2011] Roberto González- Ibánez, Smaranda Muresan, and Nina Wacholder Identifying sarcasm in twitter: a closer look. In Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics: Human Language Technologies: short papers-volume 2, pages Association for Computational Linguistics.

10 [Hernández-Farías et al.2015] Irazú Hernández-Farías, José-Miguel Benedí, and Paolo Rosso Applying basic features from sentiment analysis for automatic irony detection. In Pattern Recognition and Image Analysis, pages Springer. [Ivanko and Pexman2003] Stacey L Ivanko and Penny M Pexman Context incongruity and irony processing. Discourse Processes, 35(3): [Joshi et al.2015] Aditya Joshi, Vinita Sharma, and Pushpak Bhattacharyya Harnessing context incongruity for sarcasm detection. In Proceedings of the 53rd Annual Meeting of the Association for Computational Linguistics and the 7th International Joint Conference on Natural Language Processing, volume 2, pages [Khattri et al.2015] Anupam Khattri, Aditya Joshi, Pushpak Bhattacharyya, and Mark James Carman Your sentiment precedes you: Using an authors historical tweets to predict sarcasm. In 6TH WORKSHOP ON COMPUTATIONAL APPROACHES TO SUBJECTIVITY, SENTIMENT AND SOCIAL MEDIA ANALYSIS WASSA 2015, page 25. [Kreuz and Caucci2007] Roger J Kreuz and Gina M Caucci Lexical influences on the perception of sarcasm. In Proceedings of the Workshop on computational approaches to Figurative Language, pages 1 4. Association for Computational Linguistics. [Liebrecht et al.2013] CC Liebrecht, FA Kunneman, and APJ van den Bosch The perfect solution for detecting sarcasm in tweets# not. [Liu et al.2014] Peng Liu, Wei Chen, Gaoyan Ou, Tengjiao Wang, Dongqing Yang, and Kai Lei Sarcasm detection in social media based on imbalanced classification. In Web-Age Information Management, pages Springer. [Liu2010] Bing Liu Sentiment analysis and subjectivity. Handbook of natural language processing, 2: [Lukin and Walker2013] Stephanie Lukin and Marilyn Walker Really? well. apparently bootstrapping improves the performance of sarcasm and nastiness classifiers for online dialogue. In Proceedings of the Workshop on Language Analysis in Social Media, pages [Maynard and Greenwood2014] Diana Maynard and Mark A Greenwood Who cares about sarcastic tweets? investigating the impact of sarcasm on sentiment analysis. In Proceedings of LREC. [Muresan et al.2016] Smaranda Muresan, Roberto Gonzalez-Ibanez, Debanjan Ghosh, and Nina Wacholder Identification of nonliteral language in social media: A case study on sarcasm. Journal of the Association for Information Science and Technology. [Ptácek et al.2014] Tomáš Ptácek, Ivan Habernal, and Jun Hong Sarcasm detection on czech and english twitter. In Proceedings COLING COLING. [Rajadesingan et al.2015] Ashwin Rajadesingan, Reza Zafarani, and Huan Liu Sarcasm detection on twitter: A behavioral modeling approach. In Proceedings of the Eighth ACM International Conference on Web Search and Data Mining, pages ACM. [Rakov and Rosenberg2013] Rachel Rakov and Andrew Rosenberg sure, i did the right thing : a system for sarcasm detection in speech. In INTER- SPEECH, pages [Reyes and Rosso2012] Antonio Reyes and Paolo Rosso Making objective decisions from subjective data: Detecting irony in customer reviews. Decision Support Systems, 53(4): [Reyes and Rosso2014] Antonio Reyes and Paolo Rosso On the difficulty of automatically detecting irony: beyond a simple case of negation. Knowledge and Information Systems, 40(3): [Reyes et al.2012] Antonio Reyes, Paolo Rosso, and Davide Buscaldi From humor recognition to irony detection: The figurative language of social media. Data & Knowledge Engineering, 74:1 12. [Reyes et al.2013] Antonio Reyes, Paolo Rosso, and Tony Veale A multidimensional approach for detecting irony in twitter. Language Resources and Evaluation, 47(1): [Riloff et al.2013] Ellen Riloff, Ashequl Qadir, Prafulla Surve, Lalindra De Silva, Nathan Gilbert, and Ruihong Huang Sarcasm as contrast between a positive sentiment and negative situation. In EMNLP, pages [Tepperman et al.2006] Joseph Tepperman, David R Traum, and Shrikanth Narayanan yeah right : sarcasm recognition for spoken dialogue systems. In INTERSPEECH. Citeseer. [Tsur et al.2010] Oren Tsur, Dmitry Davidov, and Ari Rappoport Icwsm-a great catchy name: Semi-supervised recognition of sarcastic sentences in online product reviews. In ICWSM. [Veale and Hao2010] Tony Veale and Yanfen Hao Detecting ironic intent in creative comparisons. In ECAI, volume 215, pages [Wallace et al.2014] Byron C Wallace, Laura Kertz Do Kook Choe, and Eugene Charniak Humans require context to infer ironic intent (so computers probably do, too). In Proceedings of the Annual Meeting of the Association for Computational Linguistics (ACL), pages [Wallace2013] Byron C Wallace Computational irony: A survey and new perspectives. Artificial Intelligence Review, 43(4):

11 [Wallace2015] Byron C Wallace Sparse, contextually informed models for irony detection: Exploiting user communities,entities and sentiment. In ACL. [Wang et al.2015] Zelin Wang, Zhijian Wu, Ruimin Wang, and Yafeng Ren Twitter sarcasm detection exploiting a context-based model. In Web Information Systems Engineering WISE 2015, pages Springer. [Wilson2006] Deirdre Wilson The pragmatics of verbal irony: Echo or pretence? Lingua, 116(10):