Spectral Clustering-Based Segmentation Framework for TikTok Influencer Classification

Saputra, Rizky Ageng (Mathematics, University Ahmad Dahlan, Yogyakarta, 55191, Indonesia)
Purwadi, Joko (Mathematics, University Ahmad Dahlan, Yogyakarta, 55191, Indonesia)

Zero : Jurnal Sains, Matematika, dan Terapan
Vol 9, No 2 (2025): Zero: Jurnal Sains Matematika dan Terapan

This study presents a data-driven segmentation model for TikTok influencers using Spectral Clustering on 120 verified beauty influencers from FastMoss TikTok Analytics (2024-2025). Five engagement metrics views, likes, comments, shares, and followers were selected via variance thresholding, explaining 92.6% of behavioral variance. A similarity graph with a Radial Basis Function (RBF) kernel (σ = 0.5) and k = 3 clusters yielded a Silhouette Score of 0.9473, indicating highly cohesive and well-separated clusters. Compared to K-Means and Hierarchical Clustering, Spectral Clustering achieved 7.8% higher cohesion, capturing complex, nonlinear engagement patterns. Principal Component Analysis (PCA) confirmed clear distinctions among Micro-Mid, Macro, and Mega influencers. Results show that influencer impact depends more on interaction dynamics than follower count, offering a graph-based approach to optimize brand strategies effectively.