Telugu Conversational Dataset for Sarcasm Detection

Sentiment analysis, which aims to identify the positive or negative tone of a given text, has seen a surge in interest over the past two decades, making it one of the most studied areas of study in the fields of Natural Language Processing and Information Extraction. Due to the ambiguous nature of sarcasm, however, sarcasm detection is an essential part of sentiment analysis. The task becomes exceedingly challenging when applied to a language with a more intricate morphology and a lack of available resources, such as Telugu. Collecting appropriate and well-annotated corpora is the main challenge in this area of study. In this work, we developed a Telugu dataset of 10,000 conversations, of which 5,000 are sarcastic, and the remaining 5,000 are not. Sarcastic conversations have been collected from various television comedy shows like Jabardasth, Extra Jabardasth, Pataas, etc., and various internet resources. At the same time, non-sarcastic conversations are collected from serials, movies, and celebrity interviews. Three experts, who are teachers and practitioners, have annotated the collected dataset.