Recurrent Attention Walk for Semi-supervised Classification

Abstract

In this paper, we study the graph-based semi-supervised learning for classifying nodes in a ributed networks, where the nodes and edges possess content information. Recent approaches like graph convolution networks and a ention mechanisms have been proposed to ensemble the rst-order neighbors and incorporate the relevant neighbors. However, it is costly (especially in memory) to consider all neighbors without a prior di erentiation. We propose to explore the neighborhood in a reinforcement learning se ing and nd a walk path well-tuned for classifying the unlabelled target nodes. We let an agent (of node classi cation task) walk over the graph and decide where to direct to maximize classi cation accuracy. We de ne the graph walk as a partially observable Markov decision process (POMDP). e proposed method is exible for working in both transductive and inductive se ing. Extensive experiments on four datasets demonstrate that our proposed method outperforms several state-of-the-art methods. Several case studies also illustrate the meaningful movement trajectory made by the agent.

Class-Attentive Diffusion Network for Semi-Supervised Classification

“…However, the state distribution of PARWs is also determined solely by the graph structure. Recently, several methods (Lee, Rossi, and Kong 2018;Akujuobi et al 2019Akujuobi et al , 2020 adopted reinforcement learning that aims to learn a policy that attentively selects the next node in the RW process. However, unlike the proposed method, their attention does not explicitly utilize class similarity since they employed additional modules to learn the policy.…”

GraPASA: Parametric graph embedding via siamese architecture

“…while others incorporate node content information for learning the graph representation [49,50,20,1,11,31]. In general, recent graph representation methods utilize dimensionality reduction techniques to distill the high-dimensional information about nodes' neighborhood into the distributed representation vector.…”

“…The proposed method is named GraPASA for Graph PArametric representation with Siamese Architecture. It stands out from alternative models on the following aspects: (1) GraPASA is a parametric method and can be easily applied to both transductive and inductive network learning tasks. (2) GraPASA elegantly integrates network topology and node content information based on the contrastive loss.…”