Abstract: Traditional classification algorithms in heterogeneous networks map the original network into multiple homogeneous networks,and neglect the correlation between nodes of different types.This paper represents the relationships between heterogeneous nodes as latent variants,and proposes a labeling model and corresponding classification algorithm in heterogeneous networks.This paper describes the problem of node labeling in homogeneous networks,analyzes the drawbacks of the algorithms that map one heterogeneous network into multiple homogeneous networks,represents the nodes in heterogeneous networks as vectors,proposes a labeling model based on vectors,applies stochastic gradient descent method to solve the proposed model,and analyzes the complexity of the algorithm.Experimental results show that,the proposed node classification model in heterogeneous networks is more accurate than both mapping homogeneous model and unsupervised latent space model.

Key words: social network, label, classification algorithm, community discovery, learning algorithm

摘要： 传统异构网络节点标注算法将网络映射为多个同构网络,忽视了不同类型节点之间的相关性,降低了分类结果的准确性。为此,将异构网络节点之间的关系表示为潜在变量,提出一种异构网络环境下的节点标签模型。描述同构网络的节点标注问题,分析传统同构网络标签模型扩展算法的局限性,将异构网络中的节点用潜在的多维向量表示,基于该潜在向量给出异构网络节点标签模型,应用随机梯度下降法进行模型求解,并分析其复杂性。实验结果表明,该模型的预测准确性优于同构映射模型和非监督潜在空间模型。

关键词: 社会网络, 标签, 分类算法, 社团挖掘, 学习算法

CLC Number: 

• TP391