The exponential growth of research literature across domains has brought along with it difficulties in successfully analyzing big corpora to identify upcoming themes, underlying relations, and evolving trends. This work proposes a new framework that merges Large Language Models (LLMs) with cutting-edge topic modeling and graph analysis to address these challenges. Semantic embeddings are created and optimized with Uniform Manifold Approximation and Projection (UMAP) and Hierarchical Density-Based Spatial Clustering of Applications with Noise (HDBSCAN) for improved topic relevance. After topic assignment, keyword extraction is performed using KeyBERT and further optimized using Maximal Marginal Relevance (MMR). The LLM aggregates representative documents in order to enhance the interpretability of topics, after which they are converted into graph models to reveal centrality features and influence within research themes. This dual strategy offers both scalability and interpretability and outperforms previous strategies in the identification of sophisticated patterns and offering actionable insights. Future research could focus on dynamic topic tracking to uncover temporal trends, highlighting emerging and declining research areas.