HIGH-PERFORMANCE SEMANTIC SIMILARITY ANALYSIS FOR MEDICAL RESEARCH DOCUMENTS USING TRANSFORMER MODELS (BIOBERT/CLINICALBERT) WITH WMD/WMS

The fast growth of medical literature poses great difficulties for finding really important and relevant research publications. While keyword-based search techniques ignore latent semantic linkages, traditional citation-based ranking systems-such as impact factors and h-index scores-often fail to adequately represent the complex influence of research. Leveraging BioBERT and ClinicalBERT models with word mover's distance or word mover's similarity, this thesis develops an advanced citation influence and semantic analysis framework that integrates parallel-influenced citation analysis with semantic similarity measures, so addressing these constraints. The methodology increases research connection by discovering semantically relevant papers that lack direct citations, bridging hidden knowledge gaps. Moreover, the suggested approach goes beyond citation counts to increase semantic similarity and relevance among several research publications detection by using deep learning-based text embeddings, thereby stressing clinically significant papers. By using this, researchers can travel beyond obsolete citation measures and identify research linked with real-time medical developments. Analyzing millions of papers with high-dimensional embeddings, however, imposes a great computational cost. To handle this, a High-Performance Computing (HPC) framework is created to parallelize similarity computations, clustering, and summarization operations. This method speeds up extensive literature review, therefore enabling real-time study discovery. This work provides a scalable, efficient, semantically enriched analysis system overall that enables researchers to find pertinent studies, rank influential publications, and more precisely and insightfully negotiate the always expanding terrain of medical literature.