Deep Learning-Based Sentiment Analysis of Social Media and E-Commerce Reviews

With the rapid growth of online platforms and ecommerce, it has become imperative to understand people's opinions to improve the user experience and customer loyalty. These platforms, such as TikTok and NYKAA, produce vast amounts of user reviews, which give us a sense of the general sentiment toward a particular taste. Nevertheless, sentiment analysis is a very difficult task, with its own issues such as idiomatic expression, sarcasm, tonality, etc. This study uses a mixture of machine learning (KNN, logistic regression) models and deep learning (LSTM, LSTM with attention, BERT) models to classify user sentiments as positive, negative, or neutral. Systematic preprocessing (text cleaning, normalization, tokenization) and feature extraction (TF-IDF, GloVe) optimize data for analysis. Results show BERT achieved the highest accuracy (84% on TikTok, 81% on NYKAA), while LSTM with Attention also performed well with complex texts. In contrast, traditional ML models like KNN and logistic regression were not accurate and lagged in performance. The study highlights the importance of context-aware models for effective sentiment classification and offers useful guidance regarding the improvement of end-user satisfaction in both social media and e-commerce contexts. These results will assist the future development of innovative techniques of sentiment analysis, which, in practice, may implement improvement of user interactions and involvement.