Deep Network Embedding Method Based on Community Optimization

EasyChair Preprint 3331

Abstract

With the rapid development of modern network communication and social media technology, the networked big data is difficult to apply due to the lack of efficient and available node representation. Network representation learning is widely concerned by transforming high-dimensional sparse network data into low-dimensional, compact and easy to use expressions. However, the existing methods get the low-dimensional feature vector of nodes, and then use it as the input of other applications (classification, clustering, prediction, visualization, etc.) for further analysis, which is lack of specific application in designing model. In this paper, a deep auto-encoder clustering model, CADNE, was proposed to represent the low dimensional features of nodes based on community structure optimization. This method can learn the low-dimensional representation of nodes and the indicator vector of their communities at the same time, so that the low-dimensional representation of nodes can not only maintain the neighborhood characteristics of the original network structure, but also maintain the clustering characteristics of nodes. Experiments on multiple data sets show that the CADNE method has better ability of low dimensional representation of nodes.

Keyphrases: Large scale complex networks, Network Embedding, community structure, deep learning

@booklet{EasyChair:3331,
  author    = {Yafang Li and Ye Liang and Weiwei Feng and Baokai Zu and Yujian Kang},
  title     = {Deep Network Embedding Method Based on Community Optimization},
  howpublished = {EasyChair Preprint 3331},
  year      = {EasyChair, 2020}}