This visualization displays thousands of articles from two Australian strategic policy think tanks as points in a 2D space. Articles with semantically similar content appear closer together.
The Strategist is the blog of the Australian Strategic Policy Institute, covering defence, cyber security, and Indo-Pacific policy. The Interpreter is published by the Lowy Institute, focusing on international affairs and Australian foreign policy.
Each article's full text is converted to an embedding vector—a list of hundreds of numbers that represent the article's meaning. These vectors are produced by a neural network trained to place semantically similar text closer together in this high-dimensional space. Two articles about submarine procurement will have similar vectors; an article about cyber security will be far away.
Since we can't visualize hundreds of dimensions directly, we use UMAP (Uniform Manifold Approximation and Projection) to project the vectors down to 2D while preserving their relative distances as much as possible.
The n_neighbors parameter controls the balance between local and global structure: lower values emphasize tight local clusters, while higher values reveal broader thematic regions. Multiple projections are precomputed so you can switch between them.
The result: articles about similar topics cluster together. You can explore thematic regions, discover connections between articles via internal links, and filter by author, tag, source, or date.