一种基于用户距离改进的线性影响力传播模型

摘要

The prediction of information diffusion based on historical behavior of users in on-line social network is one of the hot spots of current research.However,the traditional propagation models can only explain the diffusion regular pattern of information in social networks,and it cannot predict the dissemination of information.Since uninfected users will be affected by infected users,Jaewan Yang and Jwe Leskovec proposed a linear influence model (LIM),but LIM model only considers the time factor in the process of information dissemination,and it ignores the spatial information,namely the relationship of users.Therefore,firstly,a measure of users' distance in social network was introduced in this paper.Combining the distance measure with LIM model,we proposed an improved LIM model based on distance regularization,namely d-LIM model.Through experiments on real data sets,the result shows that d-LIM model can achieve better prediction accuracy than the compared methods.%根据在线社交网络中用户的历史行为进行信息传播的预测是当前研究的热点之一,然而传统的传播模型仅解释了信息在社交网络中的传播规律,不具备信息传播预测能力.Jaewan Yang和Jwe Leskovec根据未激活的用户会受到激活用户的影响,提出了线性影响力模型LIM(Linear Influence Model),但是LIM模型在信息传播的过程中只考虑了时间因素,忽略了信息在传播过程中的空间因素,即用户间的相互关系.首先引入社交网络中用户间距离的度量,并结合距离的度量对LIM模型进行了改进,提出了基于距离正则化的LIM模型,即d-LIM模型.真实数据集上的对比实验表明,d-LIM模型能获得更准确的预测结果.