OPEN-AMZPRE : Optimized Preprocessing with Ensemble Classification for Amazon Product Reviews Sentiment Prediction

Abstract

Customer feedback plays a vital role in helping consumers make informed purchasing decisions. Understanding customer opinions and preferences through sentiment analysis is crucial. However, existing sentiment analysis methods face challenges when dealing with noisy, unstructured text data, leading to limitations in accuracy, precision, recall, and F1-score. To address these limitations, this paper introduces OPEN-AMZPRE, an innovative solution for sentiment prediction in Amazon Product Reviews. Unlike traditional approaches that rely on standard techniques like tokenization and stopword removal, OPEN-AMZPRE utilizes a comprehensive preprocessing pipeline. This pipeline includes various steps such as text normalization, lowercasing, handling rare words, expanding contractions, removing HTML tags, tokenization, removing stopwords, replacing slang words, removing digits, stemming, lemmatization, punctuation and special character removal, white space removal, spell checking and correction, and removal of duplicate text. Additionally, the proposed algorithm employs an ensemble classification approach by combining optimized versions of K-Nearest Neighbors (KNN), Naive Bayes, J48 (C4.5 decision tree), and Random Forest classifiers. The hyperparameters of each classifier are tuned to achieve optimal accuracy and performance. By combining the outputs of these classifiers, the algorithm produces robust sentiment predictions. The methods of accuracy, precision, recall, and F1-score are utilized to improve sentiment prediction and provide valuable insights for both consumers and businesses in relation to Amazon Product Reviews.