Research on early content virality recognition integrating morphological features and signal-to-noise separation features of signals

Purpose To overcome the inability of existing approaches to detect weak, multimodal and emerging early signs of virality, this study proposes a novel recognition framework. By integrating morphological and signal-to-noise separation (SNS) features, the framework accurately identifies authentic viral signals and predicts their growth trajectories. Design/methodology/approach The proposed framework constructs content signal morphology by combining multimodal sentiment analysis based on BERT and ViT, CRITIC-based influence weighting and HDBSCAN clustering. Three SNS features, frequency orderliness, temporal dependency and semantic novelty, are extracted to characterize the separation between emerging viral signals and background noise. An LSTM model is then employed for dynamic classification of early-stage virality signals. Findings Experiments conducted on a large-scale Sina Weibo dataset demonstrate that the integrated framework significantly outperforms traditional approaches and single-feature baselines in terms of precision, recall and overall accuracy. The results confirm the complementary and synergistic effects of morphological and SNS features in distinguishing authentic early-stage virality signals from noisy content streams. Originality/value This study advances early content virality recognition by integrating signal morphology and SNS features within a unified analytical framework. It shifts the analytical focus from reactive viral event detection to proactive discovery of latent virality signals embedded in noisy, multimodal social media streams. The proposed framework offers a theoretically grounded and empirically validated solution for real-time content virality detection and early warning systems on social media platforms.