Enhancing Sentiment Analysis With Emotion And Sarcasm Detection: A Transformer-Based Approach

Abstract

Our research focuses on analyzing the reviews generated on ecommerce sites like Amazon, which is complex due to the diverse ways in which the customers express themselves. They may be informal, sarcastic, and contain underlying and hidden sentiments making it difficult to analyze the sentiments meticulously. Traditional methods have been successful in training sentiment analysis models which predict whether a given statement or review is positive, negative or neutral. They fail to record deeper and underlying emotions like frustration, joy, anger and also, cannot detect sarcasm, where a positively shaped statement is actually negative. Therefore, our research puts forward an advanced sentiment analysis system that incorporates both Natural Language Processing (NLP) and Machine Learning techniques to bridge these gaps. Consequently, integrating sentiment analysis, emotion and sarcasm detection gives a better understanding of the product and customer feedback. To enhance the readability, the system also consists of visualizations of the sentiments and emotions through interactive pie charts and word clouds. This hence helps businesses to take data-centric decisions, and also help customers get a concise analysis of the product. This research emphasizes on the usage of VADER and BERT for sentiment analysis. Additionally, a neural network is used for sarcasm detection and Amazon reviews are scraped using Beautiful Soup.