<![CDATA[The phenomenon of micro-virality on TikTok illustrates how content can rapidly spread on a small scale before reaching broader virality. Understanding its driving factors is essential for supporting digital marketing strategies, managing content creators, and analyzing social media trends. This study aims to detect and predict the potential of micro-virality in TikTok videos by integrating a social listening approach with machine learning techniques. The dataset consists of approximately 4,000 TikTok posts enriched with 20 features across five categories, including user metadata (author popularity, follower ratio), temporal features (posting time and day), network features (hashtags and mentions), content features (text length and keywords), and contextual elements (trending music and video duration). To ensure objective labeling, a quantile-based threshold was applied, categorizing videos in the top 25% of view counts (≥ 26,300,000 views) as viral, resulting in a class distribution of 24.74% viral and 75.26% non-viral. To address this imbalance, the SMOTENC technique was used to oversample the minority class and enhance data representativeness. Three machine learning algorithms Random Forest, Extreme Gradient Boosting (XGBoost), and Artificial Neural Network (ANN) were implemented. Experimental results show that Random Forest improved from 88% to 92%, XGBoost maintained strong performance at 95%, and ANN increased significantly from 92% to 93% after SMOTENC application. These findings indicate that SMOTENC effectively improves model generalization and reduces bias toward majority classes, supporting more reliable early-stage virality prediction. Overall, the study enriches social media analytics research and provides practical insights for optimizing TikTok content strategies and early trend detection.]]>
Copyrights © 2026
