REGULARIZED QUADRATIC DISCRIMINANT ANALYSIS BASED ELLIOT SYMMETRIC RECURSIVE DEEP STRUCTURE NEURAL LEARNING CLASSIFIER FOR SENTIMENT CLASSIFICATION

Abstract

Sentiment analysis is the process of categorizing the opinions in the text to determine whether the attitude towards a particular product is positive, or negative. Due to a large number of reviews, sentiment analysis faces a several challenges to extract useful information from reviews. In order to improve the sentiment classification accuracy, Regularized quadratic discriminant Analysis based on Elliot Symmetric Recursive Deep Structure Neural Learning classifier (RQDA-ESRDSNLC) technique is introduced for identifying the positive and negative classification with higher accuracy. The RQDA-ESRDSNLC technique comprises four layers, namely one input layer, two hidden layers, and one output layer. In the RQDA-ESRDSNLC technique, the number of customer reviews is taken as an input and sent to the input layer. After that, the tokenization is performed in the hidden layer 1. In hidden layer 2, the stopwords are eliminated using Regularized quadratic discriminant Analysis for improving the classification accuracy in the sentiment analysis process. In the output layer, the Elliot Symmetric activation function is used to classify the customer reviews into positive and negative reviews for providing recommendations to the user for a particular item. This in turn helps to improve the classification accuracy and time complexity. Experimental evaluation is carried out on factors such as classification accuracy, error rate, time complexity, and space complexity with respect to a number of customer reviews. The results and discussion demonstrate that the proposed RQDA-ESRDSNLC technique increases the accuracy and minimizes the error as well as time complexity than the existing techniques.