Scholarlink

<p>Information diffusion prediction aims to infer potential propagation paths within social networks based on user relationships or interest preferences. Current research centers on modeling user preference dynamics, mainly by discretizing continuous processes into sequential temporal snapshots to capture evolving states. However, such methods infer preferences solely based on localized, discrete observations, failing to adequately account for the intrinsic process of continuous, smooth evolution inherent in preferences themselves. Consequently, they exhibit limitations when modeling preference dynamics. To this end, this paper proposes a <b>D</b>ynamic <b>S</b>tructure-aware <b>I</b>nformation <b>d</b>iffusion (DSId) modeling framework aimed at achieving high-precision prediction of the information propagation process. This model first integrates social relationship features with user content static interaction features to construct a heterogeneous graph, from which structural features are extracted via a graph neural network (GNN). Furthermore, diffusion paths are modeled as a diffusion hypergraph, and their evolutionary dynamics over continuous time are characterized using a coupled ordinary differential equations (ODEs) system. By coupling structural representation and dynamic processes within a unified framework, this approach effectively captures latent correlations among users, enabling continuous time dynamic inference of diffusion paths. Experiments on real-world datasets show that the model outperforms existing baselines, achieving higher accuracy and robustness. This research provides novel theoretical and methodological support for analyzing information propagation and maximizing temporal impact.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

DSId: Predicting information propagation in hybrid dynamic neural ODEs and static features within hypercurved spaces

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月6日

查看全文
  • Zhengmi Li,
  • Tao Wu,
  • Yuqing Sun,
  • Ling Liu,
  • Zhijian Zhang,
  • Lin Jiang

摘要

Information diffusion prediction aims to infer potential propagation paths within social networks based on user relationships or interest preferences. Current research centers on modeling user preference dynamics, mainly by discretizing continuous processes into sequential temporal snapshots to capture evolving states. However, such methods infer preferences solely based on localized, discrete observations, failing to adequately account for the intrinsic process of continuous, smooth evolution inherent in preferences themselves. Consequently, they exhibit limitations when modeling preference dynamics. To this end, this paper proposes a Dynamic Structure-aware Information diffusion (DSId) modeling framework aimed at achieving high-precision prediction of the information propagation process. This model first integrates social relationship features with user content static interaction features to construct a heterogeneous graph, from which structural features are extracted via a graph neural network (GNN). Furthermore, diffusion paths are modeled as a diffusion hypergraph, and their evolutionary dynamics over continuous time are characterized using a coupled ordinary differential equations (ODEs) system. By coupling structural representation and dynamic processes within a unified framework, this approach effectively captures latent correlations among users, enabling continuous time dynamic inference of diffusion paths. Experiments on real-world datasets show that the model outperforms existing baselines, achieving higher accuracy and robustness. This research provides novel theoretical and methodological support for analyzing information propagation and maximizing temporal impact.