Detecting sarcasm in Bengali comments using NLP

Abstract

Natural Language Processing (NLP) is a subset of Machine Learning which resides at the intersection of Linguistics and Computer Science. It deals with the capability of computers to learn and work with human languages. With the emergence of social media platforms, modern-day communication is being digitalized more than ever. To keep up with this rapid flow of development, the advancement of automated text processing and artificial language interpretation has become necessary. These concerns have given birth to a domain called Sentiment Analysis where blocks of text are processed to extract prominent sentiments that are prevalent within them. These sentiments can be happiness, sadness, anger, disgust, etc. Over the past few years, similar studies have garnered the attention of a vast number of computer scientists and linguists but as the study progresses and expands in the form of lan guages, concentrations, and contexts more and more challenges have started to show up. One of these challenges is the interpretation of figurative language. Figurative language refers to the structure of speech where the actual meaning defers from the literal meaning. The best example of this is Sarcasm which is a sort of figu rative language used with an intention of mockery or humor. Detecting sarcasm is considered to be one of the most challenging tasks in the domain of NLP due to the figurative structure and creative nature of sarcastic texts and the lack of relevant data on the internet. Determining sarcasm can often be difficult for even human beings as one has to have a strong understanding of the context to detect sarcasm. However, many studies have achieved respectable results by following the context unaware unimodal methods using classical Machine Learning, Deep and Hy brid Neural Networks. Motivated by such research, the objective of this paper is to take a step toward detecting sarcasm in the Bengali Language domain using Sup port Vector Machine (SVM), Cogniinsight(Word2Vec), and Bidirectional Encoder Representations from Transformers (BERT) on a novel dataset. To the best of my knowledge, this will be the first-ever initiative taken toward detecting sarcasm in Bengali Language using BERT.