Sarcasm plays a crucial role during communication on social media platforms, which is often expressed to mock or belittle a specific person. Detection of sarcasm has gained more attention in various fields, especially in Natural Language Processing (NLP). Several methods have been developed based on audio signals, but faced certain difficulties, including information loss, over-fitting issues, and complications in dealing with unstructured attributes in audio signals. In order to address these aforementioned challenges, this research proposes a search and track optimization-enabled distributed recurrent neural network (SrTaO-DRNN) model for effective sarcasm detection. Besides, the proposed model adapts a distributed learning mechanism, which drastically reduces time consumption and enables quick training for effective sarcasm detection. Moreover, the SrTaO algorithm improves the model's performance by adjusting the model's hyperparameters and contributes to minimize the error during the detection. In addition to that, the SrTaO-DRNN model evacuates unwanted noises of signals, thereby minimizing over-fitting challenges and model complexity during computation. Extensive experimental results demonstrate that the proposed SrTaO-DRNN model attained superior results by attaining an accuracy of 96.11% and precision of 94.53% using the MUStARD++ dataset, outperforming the other conventional methods.
Chaudhary et al. (Fri,) studied this question.