.png){kind=icon}
.png){kind=icon}
Return
A faster deep graph clustering network based on dynamic graph weight update mechanism
.png){kind=icon}
PRE
AI
DOI：10.1007/s10586-024-04549-6.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Deep graph clustering has attracted considerable attention for its potential in handling complex graph-structured data. However, existing approaches often prioritize improving embedded representations, leading to computationally intensive models, particularly with large-scale graphs. Meanwhile, neglecting changes in node similarity and corresponding adjustments in the original structure during iterative optimization can hinder the effectiveness of graph embedding methods. To address these challenges, this paper develops a Faster Deep graph Clustering network based on a dynamic Graph Weight update mechanism (FDCGW). In particular, we substantially reduce feature dimensions through a linear transformation that maintains the original node similarity. Also, FDCGW uses the Embedding Graph Auto-Encoder (EGAE) model to improve representation learning, which includes one encoder and dual decoders. Additionally, the proposed graph weight update mechanism uses dynamic similarity information and learned representations to increase embedding detection capability by reducing low-similarity edges. Extensive experiments and theoretical analysis are performed to prove the superiority of FDCGW on real-world datasets. In addition to the superior performance of the proposed scheme, the results show a speedup of orders of magnitude compared to the existing graph clustering methods.
Keywords:
Deep graph clustering
Embedding graph auto-encoder
Graph weight update
Representation learning
Journal
C
IF:
4.1
Papers:
5.0K
Citations:
7.5K