Sentiment Classification in Consumer Reviews Using Multinomial Naïve Bayes in E-Commerce Platforms
Abstract
Consumer reviews are a vital part of the buying decisions customers make in e-commerce websites; thus, sentiment classification plays a key role in ascertaining what customers feel regarding their experience. This paper utilized sentiment classification of user-authored product reviews via the Multinomial Naïve Bayes Classifier (MNBC). This probabilistic classifier is optimal for all sorts of text data classification tasks. Most existing sentiment classification methods have limited capabilities to handle imbalanced data, exhibit low adaptability, and incur high computational expense, resulting in reduced accuracy in practical applications. Conventional machine learning classification methods often struggle to analyze large-scale text data effectively and lack the capability to classify brief or unclear reviews accurately. To overcome such limitations, a sentiment classification system utilizing MNBC is proposed to determine user sentiment based on e-commerce product reviews. This approach allows sentiment aspects to be integrated into the recommendation reasoning, enhancing recommendations through Term Frequency–Inverse Document Frequency (TF-IDF). The approach employs effective preprocessing and feature extraction mechanisms to effectively extract and classify high-dimensional text content without losing computational simplicity. Experimental results show that MNBC achieves high accuracy and performance in sentiment classification, resulting in favorable sentiment aspects, including a sentiment polarity score of 0.86%, label accuracy of 94.2%, sentiment score of 0.89%, and sentiment consistency of 95.1%. The results demonstrate the scalability and accuracy of MNBC in real-time or instant placements, particularly in the context of e-commerce applications.