Detection of Sarcasm Using Bi-Directional RNN Based Deep Learning Model in Sentiment Analysis

Abstract

Detecting sarcasm in text can be challenging, but it's an opportunity to exercise the ability to consider various perspectives and aspects. Considering multiple viewpoints while analysing sentences can lead to comprehensive results and constructive discussions. Sentiment analysis algorithms are constantly improving and becoming more accurate in identifying sarcasm, even in implicit expressions where the sarcasm is not explicitly stated. It's exciting to have the opportunity to detect sarcasm conveyed through subtle cues, as it adds another layer of complexity to the task. The field of sarcasm detection has made great strides with the use of supervised classification models that can accurately distinguish between sarcastic and non-sarcastic sentences through labelled data. This approach shows promise for identifying explicitly incongruous statements, and there is an increasing need for techniques that can successfully detect sarcasm in sentences with implicit sentiment incongruity. This study employs a deep learning model to tackle the challenge of distinguishing implicit sarcasm. The model being used is a recurrent neural network (RNN), which has the ability to retain numerical representations of previously processed information. The proposed model is designed to effectively identify sarcasm by capturing context-based incongruity within the presented text, which is a crucial factor. The proposed model, Proposed LSTM (P-LSTM), effectively captures the dependencies among words in both suffixes and prefixes by scanning the given sentences bi-directionally. The classification is organised into two levels: emotional and semantic. The suggested approach offers a way to accurately evaluate and categories tweets based on their sarcastic state. Excitingly, the proposed method is being evaluated through experiments on two automatically annotated datasets and two manually annotated datasets to determine its effectiveness. The suggested model is compared to numerous state-of-the-art methods, highlighting its efficacy. The findings demonstrate that the suggested model is highly effective for sarcasm detection compared to other methods currently in use. The suggested model has great potential in recognising implicit sarcasm by utilising deep learning and context-based incongruity analysis, which could lead to significant improvements over existing methods.