Abstract:Current graph clustering methods emphasize individual node and edge con nections, while ignoring higher-order organization at the level of motif. Re cently, higher-order graph clustering approaches have been designed by motif based hypergraphs. However, these approaches often suffer from hypergraph fragmentation issue seriously, which degrades the clustering performance greatly. Moreover, real-world graphs usually contain diverse motifs, with nodes participating in multiple motifs. A key challenge is how to achieve precise clustering results by integrating information from multiple motifs at the node level. In this paper, we propose a multi-order graph clustering model (MOGC) to integrate multiple higher-order structures and edge connections at node level. MOGC employs an adaptive weight learning mechanism to au tomatically adjust the contributions of different motifs for each node. This not only tackles hypergraph fragmentation issue but enhances clustering accuracy. MOGC is efficiently solved by an alternating minimization algo rithm. Experiments on seven real-world datasets illustrate the effectiveness of MOGC.

What problem does this paper attempt to address?

### What problems does this paper attempt to solve? This paper aims to solve several key problems in graph clustering, especially: 1. **Integration of high - order structures**: Current graph clustering methods mainly focus on the low - order connections of nodes and edges, while ignoring the higher - order organizational structures (such as motifs) in the network. The author points out that motifs are the basic building blocks of the network and can provide new insights beyond edge - based node clustering. 2. **Hypergraph fragmentation problem**: High - order graph clustering methods based on motifs usually assume that the motif - based hypergraph is a connected graph. However, in actual networks, such hypergraphs often become fragmented, resulting in the original connected graph being divided into multiple connected components and isolated nodes. These isolated nodes cannot get the support of the original network during the clustering process, so that their class labels are random, which seriously affects the clustering performance. 3. **Integration of multi - motif information**: Large - scale networks in the real world usually contain multiple motifs, and each node may participate in multiple motifs. Different types of motifs have different importance to nodes. How to integrate the information from multiple motifs at the node level to achieve accurate clustering results is a key challenge. To solve these problems, the author proposes a multi - order graph clustering model (Multi - order Graph Clustering, MOGC). This model automatically adjusts the contribution of different motifs to each node through an adaptive weight learning mechanism, thereby not only solving the hypergraph fragmentation problem, but also enhancing the clustering accuracy. MOGC is efficiently solved by an alternating minimization algorithm, and its effectiveness is verified by experiments on seven real - world datasets. ### Formula summary - **Fused adjacency matrix**: \[ A_f=\frac{1}{2} \sum_{j = 1}^{m}(W^{M^p_{q_j}}A^{M^p_{q_j}}\text{diag}(\Lambda_{:,j})+\text{diag}(\Lambda_{:,j})A^{M^p_{q_j}}W^{M^p_{q_j}}) \] where \(W^{M^p_{q_j}}\) represents the isolated node matrix based on motif \(M^p_{q_j}\), \(A^{M^p_{q_j}}\) is the motif adjacency matrix, and \(\Lambda\) is the motif weight matrix. - **Optimization objective function**: \[ \min_{\Lambda, U}\text{tr}(U^T L_f U)+\alpha\|\Lambda\|_F^2 \] Subject to the constraints: \[ U^T D_f U = I,\quad\Lambda1_m = 1_n,\quad\Lambda\geq0 \] - **Alternating optimization process**: - Fix \(\Lambda\) and solve for \(U\): \[ \min_{U}\text{tr}(U^T L_f U)\quad\text{s.t.}\quad U^T D_f U = I \] - Fix \(U\) and solve for \(\Lambda\): \[ \min_{\Lambda}\text{tr}(U^T L_f U)+\alpha\|\Lambda\|_F^2\quad\text{s.t.}\quad U^T D_f U = I,\quad\Lambda1_m = 1_n,\quad\Lambda\geq0 \] Through this method, MOGC effectively solves the hypergraph fragmentation problem in high - order graph clustering and improves the clustering accuracy through the adaptive weight learning mechanism.

Multi-order Graph Clustering with Adaptive Node-level Weight Learning

Self-Adaptive Clustering of Dynamic Multi-Graph Learning

Multi-scale Graph Clustering Network

Multi-view and Multi-order Graph Clustering via Constrained l1,2 -norm

Cluster-aware Multiplex InfoMax for Unsupervised Graph Representation Learning

CoHomo: A Cluster-Attribute Correlation Aware Graph Clustering Framework

Incorporating Higher-order Structural Information for Graph Clustering

Motif-induced Graph Normalization

Adaptive Local Modularity Learning for Efficient Multilayer Graph Clustering

Representation learning of graphs using graph convolutional multilayer networks based on motifs

Revisiting Modularity Maximization for Graph Clustering: A Contrastive Learning Perspective

Motif Graph Neural Network

PSMC: Provable and Scalable Algorithms for Motif Conductance Based Graph Clustering

Adaptive Harmony Learning and Optimization for Attributed Graph Clustering

Cross-linking of actin filaments by heavy meromyosin.

Dual-Optimized Adaptive Graph Reconstruction for Multi-View Graph Clustering

Graph Clustering with High-Order Contrastive Learning

Balanced Multi-Relational Graph Clustering

Attributed Multi-Order Graph Convolutional Network for heterogeneous graphs

Multilayer Clustered Graph Learning

Homophily-enhanced Structure Learning for Graph Clustering