Research on Large-scale User Association Feature Data Visualization Based on Nonlinear Dimensionality Reduction Methods

In large-scale user data analysis, nonlinear correlations in high-dimensional features pose challenges to data reduction and visualization. This paper constructs a dimensionality reduction framework by optimizing the t-SNE algorithm's neighborhood preservation and UMAP's manifold learning. The framework integrates adaptive bandwidth selection with hierarchical sampling for high-fidelity feature mapping, and improves visualization clarity through enhanced force-oriented layouts and multi-level scaling. Testing with million-level datasets shows the improved method achieves 92.3% feature fidelity, 3.6× computing efficiency, 59.2% reduction in edge intersections, and 0.892 cluster integrity, providing efficient solutions for large-scale user feature analysis.