A free interactive tool called The Global Research Space now visualizes 11 million academic papers using SPECTER2 embeddings and UMAP projection, with temporal slicing and daily updates. This helps researchers quickly grasp macroscopic trends across vast scientific literature, making it easier to stay current with daily paper floods. The map uses voronoi partitioning around high-density peaks at multiple depth levels, supports keyword and semantic queries, and includes an analytics layer for ranking institutions, authors, and topics.
Background
SPECTER2 is a transformer model from Allen AI that generates high-quality embeddings for scientific documents, capturing semantic relationships. UMAP (Uniform Manifold Approximation and Projection) is a dimensionality reduction technique that preserves local and global structure, commonly used for visualizing high-dimensional data. OpenAlex is a free, open-source catalog of scholarly works with over 250 million entries, serving as the data source for this project.