Emergent Knowledge Graphs from High-Order Semantic Spaces
Authors
Abstract
A reproducible, from-first-principles method for discovering emergent knowledge-graph relations in high-order semantic vector spaces. Combines manifold learning, spectral analysis, and residual factorization to extract non-linear conceptual structures from embedded text corpora — yielding interpretable graphs, cosine-based similarity measures, and labelled semantic relations suitable for automated ontology construction. Implementation lives in the nonlinear-semantic-graphs repository.
Permalink: https://patrickaudley.com/publications/emergent-knowledge-graphs.html