Integrating Sentiment Analysis and Machine Learning for Patient-Centric Drug Recommendation Systems

Abstract

The proliferation of online platforms has enabled patients to share their experiences with medications, thereby offering valuable real-world insights into drug efficacy, side effects, and overall satisfaction. Traditional drug recommendation systems predominantly rely on clinical data and fail to incorporate the nuanced perspectives captured in patient reviews. This study addresses these limitations by developing a sentiment-driven drug recommendation system that integrates real-world feedback through machine learning. Using the UCI ML Drug Review Dataset, the system employs BERT for sentiment analysis and Word2Vec for feature extraction to process unstructured textual data. A hybrid recommendation engine combines collaborative and content-based filtering to deliver personalised suggestions. Evaluation metrics, including Precision@K, Recall@K, and Mean Reciprocal Rank (MRR), demonstrate the superior performance of the proposed system compared to existing solutions, such as Epocrates, IBM Watson Health, and GoodRx. The results highlight its ability to rank drugs more effectively, with Precision@5 reaching 86.4% and MRR achieving 0.812. The integration of sentiment analysis allows the system to recommend drugs based on both clinical efficacy and patient-reported satisfaction, bridging the gap between structured medical data and real-world experience. Although the model shows promising results, challenges such as data imbalance, noisy text data, scalability, and interpretability remain. Addressing these issues through explainable AI and transfer learning can further enhance system robustness and usability. This study underscores the potential of integrating advanced NLP techniques into healthcare applications, paving the way for more patient-centric drug recommendation systems