Social network news dissemination and detection model based on multi-scale information

At a critical stage of social network information evolution, the rapid spread of false news, emotional content, and rumours poses significant challenges to information governance.To address this issue, a multi-scale social network news dissemination and detection model is proposed.The model integrates multi-scale feature extraction, cascaded convolutional networks, and cross-modal information modelling to enhance feature representation and propagation pattern capture.Experimental results show that introducing macro-micro dual-scale modelling and gated fusion improves the F1 score to 0.895 and reduces the mean absolute percentage error to 8.9%, representing gains of 7.8 and 3.4 percentage points over single-scale baselines (p < 0.01).Across diverse communication scenarios, the model consistently outperforms comparison methods, achieving macro-F1 scores of 86%-91% and micro-F1 scores of 88%-92%.With an average detection delay of approximately 12 ms, the model balances real-time performance and robustness, demonstrating effectiveness and stability for multi-scenario news detection.