<![CDATA[The exponential growth of customer review data on e-commerce platforms generates high volume, variety, and velocity of data, necessitating efficient Big Data Analytics approaches beyond the constraints of traditional relational databases. This study aims to analyze product review sentiments to identify customer satisfaction patterns and inconsistencies between star ratings and textual reviews, while evaluating the efficiency of NoSQL databases in managing such non-relational data structures. The method employed involves text mining and Natural Language Processing (NLP), applying data preprocessing, Term Frequency-Inverse Document Frequency (TF-IDF) feature extraction with bigram-with-punctuation modeling, and text classification using the TextBlob library. The unstructured review data is managed using a NoSQL MongoDB database cluster with a scale-out architecture. The results indicate that integrating symbol-only word filtration significantly improves keyword relevance, and n-gram modeling successfully captures negation contexts like "not good". Word cloud visualizations effectively extract dominant keywords in negative reviews (Score 1), highlighting product quality criticisms ("taste", "product") and origin concerns ("made in"). Furthermore, sentiment distribution inconsistencies were detected, wherein low-rating reviews scores 1 and 2 still exhibited 15.1% and 29.2% positive sentiment, respectively. The implications of this research provide strategic recommendations for e-commerce executive management to integrate real-time sentiment monitoring systems for data-driven decision-making, proactive product development, and customer service optimization.]]>
Copyrights © 2026
