双模网络下基于节点流行度的潜在空间模型

A Popularity-Scaled Latent Space Model for Bipartite Network

本文提出了双模网络下基于节点流行度的潜在空间模型,不仅能够显式地表达节点间产生连接的概率,而且可以推导出双模网络的连接的传递性、节点度的异质性等特征,这些特征可以通过数值化定量的方式描述网络生成过程中的常见规律。在此基础之上,本文进一步提出了加权概率指标,用以衡量双模网络的节点间未来产生连接的可能性。最后,本文分别在模拟数据、公开数据集和某在线点评网站的商户一消费者网络数据上验证了模型假设符合实际数据的分布,并使用加权概率指标与其他多种双模网络链路预测的方法进行比较分析。实验结果表明,本文提出的方法不仅可以量化分析网络生成过程中的特征,而且在实验数据上的链路预测能力整体优于其他双模链路预测方法。

In order to study the probabilistic characteristics of bipartite network,we propose a popularityscaled latent space model for bipartite network. It considers various network features,such as heterogeneity and transitivity. Furthermore, these features can be expressed and explained explicitly under the model assumptions. Based on the model,a weighted probability index is proposed to measure the possibility of future connection between nodes in the bipartite network. To demonstrate the finite sample performance of the index,abundant simulation studies are conducted. In addition,both a public dataset and a real dataset are analyzed.Experimental results show that the proposed method can not only quantify the characteristics of the network generation process,but also outperform other link prediction methods for bipartite network.