针对链接挖掘中网络的结构难以预测这个难点问题,提出了一个关于链接预测的新型半监督学习方法——基于快速共轭梯度方法和链接相似性传递增殖原理的链接预测算法,利用节点相似性等辅助信息去预测未知结构。该算法利用张量的形式去表示多维的复杂的多关系数据,利用克罗内克积与克罗内克和去计算张量之间的相似性,利用向量特技方法降低了算法的时间和空间复杂度。在社会网络和生物信息网络等环境下,通过实验验证了算法的有效性和健壮性。
It is very hard to forecast about structure of network in link mining. To slove the problem,this paper proposed a new semi-supervisor learning algorithmic based on an accelerated conjugate gradient method and link similarity delivery proliferation,by using auxiliary information such as node similarity to predict the unknown structure. Used the tensor to represent the multidimensional complexity multi-relation data,calculated the similarity of tensors by Kronecker product and Kronecker sum,reduced the complexity of the compute time and RAM. The effectiveness and robustness of the algorithmic was tested in social networks and biological networks.