Unparalleled sarcasm: a framework of parallel deep LSTMs with cross activation functions towards detection and generation of sarcastic statements

Abstract

Sarcasm is a modest kind of mockingly expressing one’s own thoughts. With the advent of social networking communication, new routes of sociability have proliferated. It may also be stated that the four chariots of being socially hilarious nowadays are humour, irony, sarcasm, and wit. Sarcasm is a clever means of encapsulating any intrinsic truth, message, or even satire in a humorous way. In this paper, we manually extract the features of a benchmark pop culture sarcasm corpus encompassing sarcastic conversations and monologues in order to build padding sequences from the vector representations’ matrices. We also suggest a hybrid of four Parallel Long Short Term Networks, each with its own activation classifier. Consecutively it achieves 98.31% accuracy among the test cases on open-source English literature. Our approach transcends several previous state-of-the-art works and results in sophisticated sarcastic statement generation. We also culture the probable prospects for producing even better refined automated sarcasm generation.