Comprehending E-commerce product reviews: a sentiment analysis approach

Recently, with the advancement of Internet technology, users have increasingly opted for online shopping as a convenient and preferred mode of buying and using products. To enhance user satisfaction, Sentiment Analysis (SA) is performed on many user reviews on e-commerce platforms. However, accurately predicting the sentiment polarities of user reviews is still challenging due to variations in sequence length, textual order, and complex logic. In this paper, by employing sentiment analysis through TextBlob, the study categorized analyzed tweets and Amazon reviews positive and negative sentiments. The SA process has four main steps: (i) Gathering data (DC), (ii) cleaning and preparing the data, (iii) extracting features (FE) or assigning weights to terms (TW), selecting relevant features (FS), and iv) classifying the polarity or sentiments (SC) of the data. Initially, the Web Scrapping Tool (WST) was used to extract customer reviews from E-commerce websites. The data taken amounted to 9872 tweets, then pre-processed and went through the TF-IDF process. 80% of the data are used for training, and 20% are used for testing, which is then classified using Random Forest, which classifies customer reviews' sentiment as positive and negative. Based on the test results, the values were calculated and obtained, and they were as follows: 89.93% accuracy, 10.06% error, 92.05% precision, 89.18% recall, and 90.59% F1-score. Finally, the data results, with 68.7% positive reviews, suggest that Amazon's dataset benefits from prebuilt security against abusive language.