A deep learning framework for sarcastic sentiment classification in opinion polls

Previous studies have demonstrated that sarcasm detection in sentiment analysis can be time-consuming and laborious task. Through a theoretical and empirical experiment, one can examine the significant influence and effect of sarcastic sentiment in an opinion poll. This paper aims to classify a given sentiment into either sarcastic or non-sarcastic class. A framework is introduced that comprises of three operators, namely Cluster + expert judgement, Train&validate and Classify_new that facilitates sentiment classification. We applied the X-means clustering algorithm with expert judgment to ensure the efficient sarcastic and non-sarcastic labeling of the studied semi-supervised twitter dataset. Based on literature, we found that the LSTM is relatively simple to setup compared to a state-of-the-art deep learning technique, namely BLSTM. Empirical results from the study indicates that LSTM performs better than the benchmark as far as precision and recall are concerned due to the complexities and higher processing cost of the state-of-the-art technique. Although LSTM is relatively simple and cost effective, it is recommended as a baseline for setting up classification model in sarcastic sentiment analysis.