Description
Public opinion on social media will change rapidly with user interaction, forming a highly dynamic communication environment, which makes it difficult for traditional prediction methods to effectively model it. In order to meet this challenge, this study proposes a temporal graph neural network network framework TGN-SAGE, which innovatively integrates graph construction, temporal graph network and sampling-based GraphSAGE aggregator to accurately capture the core characteristics of public opinion dissemination. In the model design, we abstract users as graph nodes and model the interaction between users into edges with timestamps. In order to verify the performance of the model, this study adopts a step-by-step experimental design. First, it completes the verification of component validity on the Wikipedia data set, and then builds a user hierarchical diagram on the Xiaohongshu platform for targeted testing. Finally, cross-platform evaluation is carried out on the Douyin platform to verify the universality. The experimental results show that compared with GAT, the use of GraphSAGE as an aggregator can significantly improve computing efficiency while maintaining strong prediction accuracy, and the user-centred graph structure can more clearly present the path of opinion dissemination. The introduction of the time coding module has further improved the prediction performance. At the specific data level, the AUC value of the model on Xiaohongshu reaches 0.8014, and the AUC value on Douyin is as high as 0.8692, which fully proves its cross-platform universality. It is worth noting that this study found that TGN-SAGE has a high sensitivity to sudden and irregular interaction patterns. This sensitivity mainly comes from the neighbourhood sampling mechanism during the peak of sudden activity. Although the model can still maintain high prediction accuracy at this time, this phenomenon also highlights the key challenges that need to be overcome in the field of dynamic graph learning in the future. In general, TGN-SAGE provides an efficient and practical solution for real-time public opinion prediction on different social media platforms, with good theoretical value and application prospects.| Period | 1 Jan 2025 → 14 Dec 2025 |
|---|---|
| Degree of Recognition | International |