The widespread use of TikTok has generated a vast number of user reviews, offering a rich dataset for sentiment analysis. This study aims to classify TikTok reviews from the Google Play Store into positive, negative, and neutral categories, a complex task due to the informal and unstructured text. The research seeks to develop a reliable sentiment analysis model using deep learning to understand user perceptions, aiding platform improvements and marketing strategies. We collected 10,000 reviews via web scraping, preprocessed through text cleaning, normalization, tokenization, filtering, and stemming. Sentiment labels were assigned automatically using a lexicon-based approach, showing predominantly positive reviews. Word2Vec transformed text into numerical vectors for feature extraction. The Bidirectional Long Short-Term Memory (Bi-LSTM) model, with Embedding, Bidirectional LSTM, Dropout, and Dense layers, achieved 80% accuracy and an F1-score of 0.78 using a 90:10 train-test split. While effective for positive and negative sentiments, neutral expressions were less accurately detected due to lower recall. Compared to traditional methods like Naive Bayes, Support Vector Machine, and K-Nearest Neighbors, Bi-LSTM offered superior accuracy and better handling of linguistic variability, making it valuable for analyzing social media feedback.