Abstract:Analyzing and characterizing the differences between networks is a fundamental and challenging problem in network science. Most previous network comparison methods that rely on topological properties have been restricted to measuring differences between two undirected networks. However, many networks, such as biological networks, social networks, and transportation networks, exhibit inherent directionality and higher-order attributes that should not be ignored when comparing networks. Therefore, we propose a motif-based directed network comparison method that captures local, global, and higher-order differences between two directed networks. Specifically, we first construct a motif distribution vector for each node, which captures the information of a node's involvement in different directed motifs. Then, the dissimilarity between two directed networks is defined on the basis of a matrix, which is composed of the motif distribution vector of every node and the Jensen–Shannon divergence. The performance of our method is evaluated via the comparison of six real directed networks with their null models, as well as their perturbed networks based on edge perturbation. Our method is superior to the state-of-the-art baselines and is robust with different parameter settings.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is: How to effectively compare the differences between directed networks. Specifically, most traditional network comparison methods are limited to the topological properties of undirected networks, while many real - world networks (such as biological networks, social networks, and transportation networks) have inherent directionality and high - order properties, and these characteristics should not be ignored when comparing networks. Therefore, the author proposes a motif - based method for comparing directed networks, aiming to capture the local, global, and high - order differences between two directed networks. ### Specific problem description 1. **Limitations of existing methods**: - Most existing network comparison methods rely on topological properties, but these methods are usually only applicable to undirected networks. - For directed networks, especially those containing directionality and high - order interactions, existing methods cannot fully capture their complexity. 2. **Research objectives**: - Propose a new motif - based method for comparing directed networks (Dm) that can simultaneously consider differences in local, global, and high - order structures. - Verify the effectiveness and robustness of this method through experiments, especially its performance under different parameter settings. 3. **Core problems**: - How to use the motif distribution vector to represent the participation of nodes in different directed motifs? - How to define the difference measure between two directed networks, including local and global differences? - How to evaluate the performance of this method in real - world networks and compare it with existing baseline methods? ### Solution overview The method proposed by the author mainly includes the following steps: 1. **Construct the motif distribution matrix**: - For each node, calculate its distribution in different directed motifs to form a node motif distribution vector. - Combine the motif distribution vectors of all nodes into a matrix. 2. **Define the network difference measure**: - Use the Jensen - Shannon divergence (JS divergence) to quantify the difference between two directed networks. - The difference measure \(D_m(G_1, G_2)\) consists of two parts: - Part 1: Global difference based on the average motif distribution. - Part 2: Local difference based on the directed network node dispersion (DNND). - Introduce the parameter \(\phi\) to adjust the weights of the global and local differences. 3. **Experimental verification**: - Conduct experiments on six real - world directed networks, including biological networks, transportation networks, and social networks. - Compare the differences between the original network and its null models. - Evaluate the stability of the network under edge perturbation. ### Formula summary - **Definition of DNND**: \[ \text{DNND}(G)=\frac{\zeta(T_1, T_2,\ldots, T_N)}{\ln(N + 1)} \] where: \[ \zeta(T_1, T_2,\ldots, T_N)=\frac{1}{N}\sum_{i,j}t_i(j)\ln\left(\frac{t_i(j)}{\mu_j}\right) \] The formula for calculating the average value \(\mu_j\) is: \[ \mu_j=\frac{\sum_{i = 1}^Nt_i(j)}{N} \] - **Network difference measure \(D_m\)**: \[ D_m(G_1, G_2)=\phi\cdot\frac{\zeta(\mu_{G_1},\mu_{G_2})}{\ln 2}+(1 - \phi)\cdot\left|q\cdot\text{DNND}(G_1)-q\cdot\text{DNND}(G_2)\right| \] where: \[ \zeta(\mu_{G_1},\mu_{G_2})=\frac{1}{2}\sum_{j}

Directed Network Comparison Using Motifs

Directed network comparison using motifs

Research on Network Similarity Comparison Method Based on Higher-Order Information

Cross-disciplinary detection and analysis of network motifs

Motif-h: a novel functional backbone extraction for directed networks

Evaluation and Comparison of Diffusion Models with Motif Features

Quantification of network structural dissimilarities based on network embedding

Thermodynamic motif analysis for directed stock market networks

Network Classification in Temporal Networks Using Motifs

Motif adjacency matrix and spectral clustering of directed weighted networks

Motifs in Big Networks: Methods and Applications.

Detecting local network motifs

motif2vec: Motif Aware Node Representation Learning for Heterogeneous Networks

Motif discovery in networks: A survey

Motif analysis in directed ordered networks and applications to food webs

Network Comparison Via Encoding, Decoding, and Causality

City Motifs as Revealed by Similarity Between Hierarchical Features

Metrics for network comparison using egonet feature distribution

Analyzing Online Transaction Networks with Network Motifs

OFFER: A Motif Dimensional Framework for Network Representation Learning

A bi-directional approach to comparing the modular structure of networks