| 摘要: |
| 为了应对网络谣言在传播初期因恶意信息危害大且观测数据常不完整所带来的挑战，提出了一种在数据不完整条件下进行消息传播扩散建模的方法。该方法旨在克服传统方法难以准确表征社交网络中信息扩散完整状态的局限，特别是在数据即时性导致级联信息不完整的情况下。结合网络结构特征与节点间的影响关系构建了用户特征输入向量，并运用图神经网络技术来聚合相邻节点的特征，从而捕捉复杂的节点关系与全局网络结构。随后，通过引入扩散传播动力学模型，该方法迭代地将节点的估计传播状态传递至其邻居，重构出更完整的信息传播级联过程。实验验证表明，在观测数据不完整的条件下，该方法仍能有效模拟信息扩散，实现高达88.79%的级联完整度，为相关领域研究提供了一种高效且可行的新思路。 |
| 关键词: 社交网络 消息源检测 消息扩散建模 信息级联 |
| DOI：10.20079/j.issn.1001-893x.240802001 |
|
| 基金项目:国家自然科学基金青年基金项目（62101095） |
|
| Modeling and Predicting Message Propagation Diffusion under Incomplete Observational Data |
| JIA Ying,HUANG Zhicheng,LI Jiabin,LU Mengping,JIAO Kaiyu,HU Hangyu |
| (1.Southwest China Institute of Electronic Technology,Chengdu 610036,China;2.School of Information and Communication Engineering,University of Electronic Science and Technology of China,Chengdu 611731,China) |
| Abstract: |
| To address the challenges posed by malicious information and incomplete observation data during the early stages of rumor propagation on social networks,a novel approach for message diffusion modeling is proposed under incomplete data conditions.The method aims to overcome the limitations of traditional approaches,which struggle to accurately represent the full state of information diffusion in social networks,especially when cascade information is incomplete due to data timeliness.To achieve this,the network structural features and inter-node influence relationships are combined to construct user feature input vectors,and graph neural network(GNN) is used to aggregate features from neighboring nodes,thereby capturing complex node relationships and global network structure.Furthermore,by incorporating a diffusion propagation dynamics model,the method iteratively propagates the estimated states of nodes to their neighbors,reconstructing a more complete diffusion cascade.Experimental results demonstrate that,even under incomplete observation data,this method can effectively simulate information diffusion,achieving a cascade completeness of up to 88.79%,which provides an efficient and feasible new approach for research in related fields. |
| Key words: social network source detection diffusion modeling information cascade |
