基于隐含社团预测的社交网络约简方法

Network simplification algorithm based on hidden community prediction

针对现有网络约简方法未考虑隐含社团的问题,提出一种基于社团预测的网络约简算法,通过图嵌入预测网络中隐含的社团关系,提高约简网络的准确度。将网络中的节点表示为欧式空间中的向量,通过节点在网络中的位置关系学习向量表示,通过层次聚类对节点进行划分,预测隐含社团,对每一层次的聚类进行网络约简。在大规模社交网络数据集上的实验结果表明,采用该方法能够在百万级大规模网络中得到更高质量的精简网络,在大规模网络的分析、挖掘及可视化等方面有广泛用途。

To address the problem that existing network abstract methods do not consider the implicit communities,a network reduction algorithm based on community prediction was proposed.The nodes in the network was represented as vectors in the Euclidean space,the vector representation was learnt through the positional relationship of the nodes in the network,the nodes was divided by hierarchical clustering,the hidden communities were predicted,and each level was clustered to reduce network.Results of experiments on large-scale social network datasets show that social network reduction methods using graph-embedding algorithms achieve higher-quality in large-scale networks,indicating a wide range of uses in large-scale network analysis,mining,and visualization.