基于感染结果的传播网络推断方法

为揭示传播网络中节点之间的父子影响关系,现有工作大多需要知道节点的感染时间,而该信息往往只有通过对传播过程进行实时监控才能获得.研究如何基于传播结果来学习获得传播网络中节点之间的父子影响关系.传播结果只包含每个传播过程中节点的最终感染状态,而节点的最终感染状态在实际中往往比节点的感染时间更容易获得.提出了一种基于条件熵的方法来推断网络中每个节点的潜在候选父节点.此外,能够通过从基于条件熵的推断结果中发现并修剪那些实际不太可能存在的父子影响关系来优化最终的影响关系推断结果.在人工网络和真实网络上的大量实验,验证了该方法的有效性和运行效率.

Diffusion Network Inference Based on Infection Results

To reveal parent-child influence relationships between nodes in a diffusion network, most prior work requires knowledge of node infection time, which is possible only by carefully monitoring the diffusion process. This work investigates how to solve this problem by learning from diffusion results, which contain only the final infection statuses of nodes in each diffusion process and are often more easily accessible in practice. A conditional entropy-based method is presented to infer potential candidate parent nodes for each node in the network. Furthermore, the inference results are able to be refined by identifying and pruning the inferred influence relations that are unlikely to exist in reality. Experimental results on both synthetic and real-world networks verify the effectiveness and efficiency of our approach.