This study develops a hybrid data mining model to analyze user relationships, content trends, and influencer–audience dynamics on TikTok, addressingchallenges of scalability, adaptability, and multi-dimensional relationship analysis in its dynamic, multi-modal ecosystem. The model integrates optimized Apriori association rule mining, k-Means clustering with Principal Component Analysis (PCA), Dynamic Graph Neural Networks (DGNN), and Natural Language Processing (NLP) techniques, including Named Entity Recognition (NER) and Word2Vec. Implemented in Python using Scikit-learn,NetworkX,and PyTorch Geometric, the system processes synthetic TikTok datasets, adhering to ethical constraints. Employing Constructive Research and Design Science Research methodologies, the model achieves a Silhouette Score of 0.1886 for clustering, a DaviesBouldin Index of 1.4676, and sentiment analysis accuracy of 0.9231, outperforming baseline systems. Results demonstrate effective identification of interaction patterns (e.g., 'like' with 0.92 support), user segmentation, real-time adaptability (graph expansion from 799 to 870 edges), and semantic insights such as 'comedy', 'nike' entities. This integrative approach enhances social media analytics, offering scalable solutions for digital marketing and content creation.