Fine Grained Sentiment Analysis using Deep Learning Techniques

Abstract

In recent times, the popularity of social networking sites has skyrocketed. The input newlinefrom customers holds immense importance for businesses, and social media platforms newlinehave emerged as a promising tool for enhancing and optimizing business prospects. newlineSentiment analysis (SA) plays a fundamental role in interpreting the contextual meaning newlinewithin user sentiments, enabling both individuals and enterprises to gain insights into newlinehow their products and services are perceived by consumers. Predicting the sentiment newlinescore of customer feedback faces challenges such as variations in text structure, length newlineof content, and intricate reasoning. It is worth noting that previous methods primar- newlineily focused on binary or tri-categorization of reviews, classifying opinions as positive, newlineneutral, or negative. However, this approach overlooked the extremely positive and ex- newlinetremely negative feedback, potentially leading to a misinterpretation of consumer sen- timents regarding a particular product or service. Such oversight can ultimately harm a business s reputation or hinder its growth trajectory. newlineA Novel Heuristic approach known as the Bidirectional-Recurrent Neural Network newline(NHBi-RNN) along with coot optimization is proposed for the purpose of multiclass newlinesentiment classification. The proposed model adeptly assigns polarity labels to sen- newlinetences within consumer feedback, encompassing very positive, very negative, positive, newlinenegative, and neutral sentiments. newlineAn innovative hybrid approach, denoted as NH-ResNeXt-RNF, is introduced by combining the ResNeXt architecture with the Recurrent Neural Framework. The pro- posed framework focuses on discerning the sentiment polarity of words related to a spe-cific domain across the entire dataset and employs a hybrid optimization approach to reduce the need for extensive annotation efforts in the context of unsupervised learning. newlineFurthermore, the effectiveness of our suggested framework is assessed through standard performance metrics. The proposed hybrid model accuracy is 96.5% and 95.37% for newline