Deep-Learning-Based Drug Recommendation and ADR Detection Healthcare Model on Social Media

The rapid rise of healthcare social media websites captures a significant amount of healthcare information, leading to mining medical content for pharmacovigilance, medication repositioning, and the healthcare industry. Paramedical advertisers gather health information from social media to propose suitable therapy. Social media makes patients aware of the revolution in clinical care and medical services. The researcher created a centralized social-media-based pharmacovigilance virtual podium to extract relevant medical services. And simultaneously built a virtual community of homogeneous e-patients for informational and emotional support. This article presents a new research-based review of the clinical recommendation system and possible medication repository and adverse drug reaction (ADR) research gaps. This article also explores ADR categorization with SVM, naive Bayes, linear regression, random forest, and deep neural networks. Subsequently, this article presents a preprocessing and clinical feature extraction approach described for drug recommendation and ADR classification. This methodology analyzed preprocessed clinical tweets for language diversity and clinical characteristics. This system also does multilayer content and collaborative filtering for patient-to-disease and disease-to-drug similarity indexing. It reveals the scope and collaborative filtering’s strengths and weaknesses. The collaborative association for filtering comparable clinical entities improves ADR detection and medicine recommendation. DNN comes out with the best result for ADR detection and classification and naive Bayes and SVM gain the comparatively biased result. In contrast, SVM achieves the highest improvement after preprocessing and with the amalgamation of clinical vector space. However, with the amalgamation of clinical vector space, the performance of ADR classification increased by approximately 12% and gained 86% accurate classification with the SIDER dataset.