IMPROVING SENTIMENT ANALYSIS IN PRODUCT REVIEWS THROUGH PREPROCESSING AND CLASSIFICATION TECHNIQUES

Abstract

In the digital marketplace, product reviews are vital in forming consumer decisions and brand reputation. Because there is so much user-generated information on the internet, it is now crucial for businesses looking to understand customer sentiment better and enhance their offerings to analyze and categorize these evaluations effectively. This study proposes a Support Vector Machine (SVM) based model for classifying product reviews evaluated on the Amazon Product Reviews dataset from Kaggle. To improve classification accuracy, the model uses an extensive preprocessing pipeline that includes text normalization, tokenization, padding, word embeddings, and aspect extraction techniques. These approaches are used to discover important elements of the reviews. These elements are mixed with the review text to create enriched feature vectors, which provide the SVM classifier with better input. According to experimental results, this model outperforms conventional techniques regarding accuracy, precision, and recall, greatly improving classification performance. This hybrid strategy, which combines SVM and aspect extraction, presents a viable way to analyze sentiment in product reviews.