Abstract This meta-analysis synthesizes findings from 30 studies examining sentiment analysis techniques applied to mobile phone reviews on e-commerce platforms. The analysis reveals that Amazon is the predominant data source, with datasets ranging from 158 to 400,000 reviews. Machine learning approaches span classical algorithms (Random Forest, SVM, Naive Bayes) to deep learning models (LSTM, CNN) and transformer-based architectures (BERT, RoBERTa). Performance metrics vary widely, with reported accuracies ranging from 51.33% to 97.48%. Common preprocessing techniques include tokenization, stopword removal, and POS tagging, while feature extraction methods encompass Bag-of-Words, TF-IDF, Word2Vec, and contextual embeddings. Consumer sentiment analysis consistently identifies key themes including battery life, camera quality, display performance, user interface, and device performance issues. Random Forest and BERT-based models emerge as top performers; though comprehensive comparative evaluation remains limited across the literature. JEL Classification: C88, D83, L86 (C88 – Methodology of data collection and data estimation; D83 – Search; Learning; Information and Knowledge; Communication; L86 – Information and Internet services; Computer software)
J et al. (Mon,) studied this question.