Sarcasm Detection in Social Media Based on Imbalanced Classification

Abstract

Abstract. Sarcasm is a pervasive linguistic phenomenon in online documents that express subjective and deeply-felt opinions. Detection of sarcasm is of great importance and beneficial to many NLP applications, such as sentiment analysis, opinion mining and advertising. Current studies consider automatic sarcasm detection as a simple text classification problem. They do not use explicit features to detect sarcasm and ignore the imbalance between sarcastic and non-sarcastic samples in real applications. In this paper, we first explore the characteristics of both English and Chinese sarcastic sentences and introduce a set of features specifically for detecting sarcasm in social media. Then, we propose a novel multi-strategy ensemble learning approach(MSELA) to handle the imbalance problem. We evaluate our proposed model on English and Chinese data sets. Experimental results show that our ensemble approach outperforms the state-of-the-art sarcasm detection approaches and popular imbalanced classification methods.

Automatic Sarcasm Detection

“…Manual: [Riloff et al 2013;Maynard and Greenwood 2014;Ptácek et al 2014;Abhijit Mishra and Bhattacharyya 2016;Abercrombie and Hovy 2016] Hashtag-based: González-Ibánez et al 2011;Reyes et al 2013;Barbieri et al 2014a;Ghosh et al 2015b;Bharti et al 2015;Liebrecht et al 2013;Bouazizi and Ohtsuki 2015a;Wang et al 2015;Barbieri et al 2014b;Bamman and Smith 2015;Fersini et al 2015;Khattri et al 2015;Rajadesingan et al 2015 [Lukin and Walker 2013;Reyes and Rosso 2014;Buschmeier et al 2014;Liu et al 2014;Filatova 2012] Other datasets [Tepperman et al 2006;Kreuz and Caucci 2007;Veale and Hao 2010;Rakov and Rosenberg 2013;Ghosh et al 2015a;Joshi et al 2016a;Abercrombie and Hovy 2016]…”

“…consider a large dataset of 66000 amazon reviews. Liu et al [2014] use a dataset from multiple sources such as Amazon, Twitter, Netease and Netcena. In these cases, the datasets are manually annotated because markers like hashtags are not available.…”

“…Hernández-Farías et al [2015] present features that measure semantic relatedness between words using Wordnet-based similarity. Liu et al [2014] introduce POS sequences and semantic imbalance as features.…”

“…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.…”

Towards the Understanding of Gaming Audiences by Modeling Twitch Emotes

“…Recent years have seen an increase in models for detecting #irony and #sar-casm. Many of these models adopted hand crafted features (amoung others (Reyes et al, 2013a;Barbieri and Saggion, 2014;Liu et al, 2014;Joshi et al, 2015)), and others employed pretrained word embeddings or deep learning systems such as CNN or LSTMs (Joshi et al, 2016;Ghosh and Veale, 2016;Poria et al, 2016;Amir et al, 2016).…”