Abstract: Traditional community detection methods can only process small-scale low-dimensional social network data because of its complexity. Aiming at this problem,this paper proposes a deep learning method for community structure based on compressive sensing. With the great advantages of reducing the feature dimension of the social network through random measurement matrix, it uses Deep Belief Network ( DBN) to learn unsupervised from the low-dimensional samples. The model is fine-tuned by supervised learning from a small scale sample sets with class labels. Experimental results show that random measurement method has a good effect of dimensionality reduction for sparse features,and DBN performs well in processing large data. It is shown to be advantageous over other community detection methods on largescale high-dimensional actual social network data.

Key words: compressive sensing, deep learning, community detection, Deep Belief Network ( DBN ), community structure, modularity

摘要： 传统社团结构发现算法复杂度高,且只适合处理小规模低维度的社会网络数据,而无法处理大规模高维度 实际网络数据。为此,提出一种基于压缩感知的社团结构深度学习方法。通过随机测量矩阵对社会网络数据进行特征降维,并使用深度信度网(DBN)对降维后的特征样本集进行无监督学习,利用带类标的小样本集进行有监督调优。仿真结果表明,随机测量方法对高维稀疏特征具有较好的降维效果,DBN 对大规模数据集具有较好的处理性能,该方法适合对大规模高维度实际社会网络数据进行高效处理。

关键词: 压缩感知, 深度学习, 社团发现, 深度信念网, 社团结构, 模块度

CLC Number: 

• TP393. 08