Sarcasm Detection in Telugu Language Text Using Deep Learning

Sarcasm detection is a growing field in Natural Languageg Processing(NLP). Sarcasm is identified by using positive or more positive words, often with a negative connotation, to insult or mock others. In sentiment analysis, detecting sarcasm in the text has become critical. We have reviewed numerous relevant research articles, but due to the Telugu language's limited resources, detecting sarcasm in Telugu texts remains challenging. As a result, the sentiment detection model struggles to accurately identify the exact sentiment of a sarcastic statement, necessitating the development of an automated sarcasm detection system. Many researchers have trained and tested various machine-learning classification algorithms to identify sarcasm, but these algorithms require a dataset as their input, which often contains noise. Various pre-processing techniques are used to remove noise from the dataset. We created a Telugu News Headline dataset on our own and stored in our local machine. Labeled the statements as sarcastic or non-sarcastic by the annotators, and then input them into our proposed model. We built the proposed model using One Hot Encoding(OHE), to transform the dataset into vectors, then fed to the Sarcasm Detection Model to determine the model accuracy. We trained and tested the Sarcasm detection model on positive or even more positive sentences. Resulted accuracy with One Hot Encoding 90.30%. We observed that One Hot Encoding(OHE) had better accuracy on the balanced Telugu news headline dataset.In the future, we can apply more verticle datasets using deep learning algorithms to detect sarcasm for better accuracy.