基于增量学习的动态社交网络链接预测

Link Prediction of Dynamic Social Networks Based on Incremental Learning

在现实世界中,社交网络的结构并不是一成不变的,而是随着时间的推移不断地发生变化。链接预测可以发现社交网络中隐藏的和未来可能产生的关系链接,这些关系链接在很多实际应用场景中得到了广泛的应用。现有的链接预测方法主要针对只具有单个时间节点的静态网络,较少关注连续时间节点的动态社交网络的链接预测。并且,直接把现有的链接预测方法用来对动态社交网络的每个时间节点的整个网络进行链接预测效率较低,不能满足在大数据背景下进行高效的链接预测。针对该问题,本文将资源分配算法改进之后,通过增量学习的思想将其引入到动态社交网络链接预测当中,提出了一种新的链接预测算法。该算法不仅仅考虑了公共邻居节点的资源分配,也考虑了待预测节点本身的资源分配。更加符合动态社交网络随着时间变化,待预测节点本身的邻居也可能会发生变化的特点。在数据集上进行的仿真实验得出的结果证明该算法提高了链接预测的时间效率,并且,预测准确率也略有提高。

In the real world,the structure of social networks is not static,but constantly changing with the passage of time.Link prediction can discover the hidden and possible relationship links.The relationship links which have been widely used in many practical application scenarios.The existing link prediction methods are mainly proposed for static networks with only a single time node,but have less concerned with link prediction of dynamic social networks with continuous multiple time nodes.Moreover,it is not efficient when the existing link prediction methods are used to process the entire network of each time node of dynamic social network directly in the context of big data.Aiming at this problem,this paper improves the resource allocation algorithm and introduces it into dynamic social networks links prediction by incremental learning.The proposed method considers not only the resource allocation of the common neighbor nodes,but also the resource allocation of the nodes to be predicted,which is more in line with the characteristics that dynamic social networks change over time and the neighbors of the nodes to be predicted may also change.The simulation results on the datasets show that the proposed method improves the time efficiency of link prediction significantly,and the prediction accuracy is also improved slightly.