Answers are grounded in this paper and cite their sources.
Computational study demonstrates improved social media popularity forecasting using multi-level multimodal fusion, suggesting more accurate tracking of engagement peaks.
As the forms of social media data become increasingly diverse, the comprehensive utilization of multimodal information can more accurately predict the popularity of content. This paper first designs a temporal popularity prediction framework for the multimodal social media time series popularity dataset. This framework uses a loss function that considers the peak of popularity for model training. Experiments show that this framework can effectively predict the temporal sequence of social media content popularity, and the prediction accuracy has significantly improved. Secondly, based on the multimodal feature extraction framework, a sliding window average method based on temporal correlation is proposed to enhance feature stability. By establishing a social media information popularity regression model, the effectiveness of the multimodal feature fusion model and the proposed sliding window average strategy in information popularity prediction is verified on the SMPD dataset. Finally, through a multi-level deep fusion method to integrate multimodal information, the fusion process is divided into three stages to achieve full interaction, and the fused features are combined with popularity prediction to further improve the prediction performance.
No takes yet. Share an insight, caveat, or question.
Zhongzi Wang (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: