基于转移自洽和偏好连接的链路预测算法研究

Research on a Link Prediction Algorithm Based on Transferring Similarity and Preferential Attachment

在大规模复杂网络中,基于网络结构相似性的链路预测方法是目前综合考虑计算复杂度低和准确性较优的预测模式。但在稀疏、低聚簇的复杂网络中,仅依靠共同邻居和节点度信息进行链路预测难以取得较为理想的预测结果。文中主要研究归纳了复杂网络中基于结构相似性的链路预测方法,并在比较现有的相似性链路预测算法特性的基础上,提出了一种基于偏好连接相似性和转移自洽相似性的TSPA相似性指标的链路预测算法。该算法从新连边概率正比于节点度和节点间相似性可传递的角度出发,构造全新的相似性指标。将该算法与其他相似性算法在经典复杂网络数据集上进行比较,实验结果表明,基于该相似性指标的算法在经典复杂网络数据集中取得了较好的预测性能。

In large-scale complex networks,the link prediction algorithm based on network structure similarity is a better mode considering computation complexity and accuracy.But in sparse and oligomeric complex networks,link prediction can hardly get ideal result through common neighbors and node degree.In this paper,the link prediction methods based on structural similarity in complex networks are mainly studied and concluded.On the basis of comparing the characteristics of existing similar link prediction algorithms,a link prediction algorithm based on TSPA similarity index based on transferring similarity and preferential attachment is proposed.This algorithm constructs a new similarity index based on the new connection probability proportional to the degree of nodes and the similarity transfer between nodes.The algorithm is compared with other similarity algorithms on the data sets of classical complex networks.The experiment shows that the proposed algorithm based on the similarity index achieves better prediction performance in the data sets of classical complex networks.