[2602.24069] Leveraging Non-linear Dimension Reduction and Random Walk Co-occurrence for Node Embedding

Computer Science > Machine Learning arXiv:2602.24069 (cs) [Submitted on 27 Feb 2026] Title:Leveraging Non-linear Dimension Reduction and Random Walk Co-occurrence for Node Embedding Authors:Ryan DeWolfe View a PDF of the paper titled Leveraging Non-linear Dimension Reduction and Random Walk Co-occurrence for Node Embedding, by Ryan DeWolfe View PDF HTML (experimental) Abstract:Leveraging non-linear dimension reduction techniques, we remove the low dimension constraint from node embedding and propose COVE, an explainable high dimensional embedding that, when reduced to low dimension with UMAP, slightly increases performance on clustering and link prediction tasks. The embedding is inspired by neural embedding methods that use co-occurrence on a random walk as an indication of similarity, and is closely related to a diffusion process. Extending on recent community detection benchmarks, we find that a COVE UMAP HDBSCAN pipeline performs similarly to the popular Louvain algorithm. Comments: Subjects: Machine Learning (cs.LG); Social and Information Networks (cs.SI) Cite as: arXiv:2602.24069 [cs.LG]   (or arXiv:2602.24069v1 [cs.LG] for this version)   https://doi.org/10.48550/arXiv.2602.24069 Focus to learn more arXiv-issued DOI via DataCite (pending registration) Submission history From: Ryan DeWolfe [view email] [v1] Fri, 27 Feb 2026 14:58:49 UTC (1,255 KB) Full-text links: Access Paper: View a PDF of the paper titled Leveraging Non-linear Dimension Reduction and Random Walk ...