Every Node is Different: Dynamically Fusing Self-Supervised Tasks for Attributed Graph Clustering

This article introduces a new methodology for attributed graph clustering, utilizing a dynamic fusion of self-supervised learning tasks. The proposed method, DyFSS, assigns different weights to different SSL tasks for each node. This dynamic approach significantly improves performance, as demonstrated by experiments on five datasets. The study highlights the limitations of previous methods that use a single SSL task and proposes an innovative solution that takes into account the diversity of nodes in a graph.

Publication date: 15 Jan 2024
Project Page: https://github.com/q086/DyFSS
Paper: https://arxiv.org/pdf/2401.06595