This study presents a comprehensive analysis of the UCI drug reviews dataset, focusing on generating condition-specific drug recommendations and utilizing machine learning and deep learning techniques for sentiment classification. A healthcare-specific semantic filtering methodology is introduced to address the challenges posed by medical terminology in sentiment analysis, including a rule- based approach to filter condition-related terms. The sentiment analysis employs the VADER sentiment analyzer, enhanced with custom medical lexicons. Additionally, a custom measure was developed to incorporate sentiment scores, review ratings, and social validation metrics, providing targeted drug recommendations based on specific conditions. We applied both machine learning and deep learning models, including Bidirectional Long Short-Term Memory (Bi-LSTM) networks, to predict sentiment across drug reviews. The Bi-LSTM model, enhanced by an Embedding layer, was designed to capture long-range word dependencies, achieving an accuracy of 81.89%. Furthermore, CNN-Bidirectional LSTM models were explored using pretrained Word2Vec and GloVe embeddings, which increased accuracy to 89.94%. The findings offer valuable insights into patient experiences with various drugs and demonstrate the effectiveness of deep learning models in capturing the nuanced context of patient reviews. Future work will focus on expanding this methodology to real-time data sources, further enhancing predictive accuracy and relevance.