SENTIMENT ANALYSIS USING DEEP LEARNING AND MACHINE LEARNING

Abstract

Sentiment analysis a classification procedure that uses machine learning algorithms to analyse the sentiment of text-driven datasets., e.g., a message which can be positive (+) or negative (-) about a specific area. The main purpose of this work is to check whether this technique is also feasible for application on customer review on amazon. A dataset is used to compare, train, and test various machine learning methods. (N = 1,00,800) having product reviews from Amazon.com which were chosen at random from a Kaggle dataset comprising 4 million reviews. Seven distinct algorithms' performance was compared.: Random Forest Classifier (RFC), XGBC Classifier (XGBC), LGBM Classifier (LGBM), Multinomial Naïve Bayes (MNB), Gradient Boosting Classifier (GBC), Decision Tree Classifier (DTC) and Bidirectional Long short-term memory network (Bi-LSTM). On finding the result from the experiment performed on the amazon dataset we got a conclusion that Bi-LSTM outperforms all the other model with the highest performance (Accuracy = 0.987, AUC = 0.895). Seven distinct algorithms' performance was compared. A comprehensive evaluation was conducted using the remaining 25200 reviews from the Amazon Kaggle dataset and a newly scraped dataset of product reviews from various categories on Amazon.com. The application of Bi-LSTM networks yielded highly accurate sentiment classification, particularly excelling in test reviews on the Amazon dataset (Accuracy = 0.832). In summary, Bi-LSTM networks demonstrate exceptional performance in categorizing customer sentiment in product reviews, with consistent results across different categories. Further investigation is necessary to determine the accuracy of classification when additional classes, such as a neutral class, are introduced.