Sarcasm Target Identification: Dataset and An Introductory Approach

Sarcasm Target Identification: Dataset and An Introductory Approach Aditya Joshi2,3,4, Pranav Goel1, Pushpak Bhattacharyya2, Mark J Carman3 1IIT (BHU) Varanasi, India, 2IIT Bombay, India, 3Monash University, Australia 4IITB-Monash Research Academy, India adityaj@cse.iitb.ac.in, pranav.goel.cse14@iitbhu.ac.in, pb@cse.iitb.ac.in, mark.carman@monash.edu Abstract Past work in computational sarcasm deals primarily with sarcasm detection. In this paper, we introduce a novel, related problem: sarcasm target identification (i.e., extracting the target of ridicule in a sarcastic sentence). As a benchmark, we introduce a new dataset for the task. This dataset is manually annotated for the sarcasm target in book snippets and tweets based on our formulation of the task. We then introduce an automatic approach for sarcasm target identification. It is based on a combination of two types of extractors: one based on rules, and another consisting of a statistical classifier. Our introductory approach establishes the viability of sarcasm target identification, and will serve as a baseline for future work. Keywords: Sentiment Analysis, Computational Sarcasm, Aspect Extraction 1. Introduction Sarcasm is a form of verbal irony that is intended to express contempt or ridicule (Source: The Free Dictionary). While several approaches have been reported for sarcasm detec- tion (Rajadesingan et al., 2015; Joshi et al., 2015; Tsur et al., 2010; Gonz´alez-Ib´anez et al., 2011), no past work, to the best of our knowledge, has attempted to identify a cru- cial component of sarcasm: the target of ridicule (Campbell and Katz, 2012). This is important because the sentiment of the sarcastic text needs to be attributed to this target of ridicule. Towards this motivation, we introduce ‘sarcasm target identification’: the task of extracting the target of ridicule (i.e., sarcasm target) of a sarcastic text. The input is a sarcastic text while the output is either (a) a subset of words in the sentence that point to the sarcasm target, or (b) a fall-back label ‘Outside’1. In this paper, we present a manually labeled dataset consisting of text from two do- mains: tweets and book snippets. We also report an auto- matic approach that takes as input a sarcastic text and returns its sarcasm target. Our hybrid approach combines a rule-based extractor (that implements a set of rules) and a statistical extractor (that uses a word-level classifier for every word in the sentence, to predict if the word will con- stitute the sarcasm target). Since this is the first work in sarcasm target detection, no past work exists to be used as a baseline. Hence, we de- vise two baselines to validate the strength of our work. The first is a simple, intuitive baseline to show if our approach (which is computationally more intensive than this simple baseline) holds value. In absence of past work, using sim- ple and obvious techniques to solve a problem have been considered as baselines in sentiment analysis (Tan et al., 2011; Pang and Lee, 2005). As the second baseline, we use a technique reported for sentiment/opinion target iden- tification since sentiment target identification appears to be related to sarcasm target identification, on the surface. Our …

Content truncated to 500 words