A Novel Approach to Classify Sentiments on Different Datasets Using Hybrid Approaches of Sentiment Analysis

Abstract

Objectives:The objective of this study is to introduce an innovative hybrid approach that incorporates CNN and Bi-LSTM models to provide a solution to the sentiment analysis problem. The HCNN-BiLSTM Model is the acronym that we present for this methodology. Methods: Pre-processing, feature extraction, and sentiment classification are the three steps in this procedure. In the pre-processing stage, unneeded data gathered from the source text reviews is filtered out utilizing NLP systems. The prior studies presented an integrated strategy referred to as RBDT, which generates particular feature sets depending on the examination, for effectively extracting features. Next, sentiments are predicted using the proposed cutting-edge HCNN-BiLSTM model and grouped various sentimental phrases into five main groups: interest, sadness, anger, happiness, and disinterest. Findings: The findings showed that in terms of F-measure, accuracy, word count, and computational time, this suggested the HCNN-BiLSTM Model operates better than conventional deep learning (CNN) and machine learning techniques (SVM). Novelty: This proposed approach uses advanced methods on five review datasets, which include the Amazon dataset, Spotify app reviews, FIFA World Cup reviews, COVID-19 Vaccination reviews, and ChatGPT reviews, to produce competitive outcomes.

A Hybrid Approach to Analyse the Public Sentiment on Covid-19 Tweets

“…The authors used the BERT model and got 82.90% accuracy (20) . The proposed experiment showed 96.40% accuracy (21) . The accuracy of the GA-DNN model is 85% (22) .…”

“…When texts reflect negative sentiment, the model predicts the negative class properly; when texts reflect good emotion, it predicts the "positive" class correctly. https://www.indjst.org/ (25) Transfer-based BERT 70% Yuchen Chai et al (20) BERT 82.90% S Ashika Parvin et al (21) HCNN-BiLSTM 96% Omar Al-Harbi et al (22) GA-DNN 85% Gyananjaya Tripathy et al (23) AEGA 78.60% Laura Imanuela Mustamu et al (24) RBF and GA-SVM 69.52% Our Proposed Model LSTM-With GA 96.40%…”