基于网络表示学习与随机游走的链路预测算法

发布时间：2018-05-17 02:39

本文选题：链路预测 + 相似性　；参考：《计算机应用》2017年08期

【摘要】：现有的基于随机游走链路预测指标在无权网络上的转移过程存在较强随机性,没有考虑在网络结构上不同邻居节点间的相似性对转移概率的作用。针对此问题,提出一种基于网络表示学习与随机游走的链路预测算法。首先,通过基于深度学习的网络表示学习算法——Deep Walk学习网络节点的潜在结构特征,将网络中的各节点表征到低维向量空间;然后,在重启随机游走(RWR)和局部随机游走(LRW)算法的随机游走过程中融合各邻居节点在向量空间上的相似性,重新定义出邻居节点间的转移概率;最后,在5个真实数据集上进行大量实验验证。实验结果表明:相比8种具有代表性的基于网络结构的链路预测基准算法,所提算法链路预测结果的AUC值均有提升,最高达3.34%。
[Abstract]:There is a strong randomness in the transfer process based on the random walk link prediction index in the unauthorized network, and the effect of the similarity between the different neighbor nodes on the transfer probability is not considered in the network structure. To solve this problem, a link prediction algorithm based on network representation learning and random walk is proposed. First, the nodes in the deep Walk learning network are represented into the low-dimensional vector space through the deep learning-based network representation learning algorithm, which is the potential structural feature of the deep Walk learning network node. In the random walk process of restarting random walk (RWR) and local random walk (LRW) algorithm, the similarity of neighbor nodes in vector space is fused, and the transfer probability between neighbor nodes is redefined. A large number of experiments were carried out on 5 real data sets. The experimental results show that the AUC value of the proposed algorithm is improved to 3.34, compared with the 8 typical network based link prediction benchmark algorithms.
【作者单位】：华南师范大学计算机学院;广东省高性能计算重点实验室;仲恺农业工程学院信息科学与技术学院;
【基金】：广东省自然科学基金自由申请项目(2016A030313441) 广东省科技计划项目(2015B010129009,2016A030303058,2016A090922008,2015A020209178) 广东省高性能计算重点实验室开放课题项目(T191527) 广州市科技计划项目(201604016035)~~
【分类号】：O157.5

【参考文献】