Abstract:Dynamic network data analysis requires joint modelling individual snapshots and time dynamics. This paper proposes a new two-way heterogeneity model towards this goal. The new model equips each node of the network with two heterogeneity parameters, one to characterize the propensity of forming ties with other nodes and the other to differentiate the tendency of retaining existing ties over time. Though the negative log-likelihood function is non-convex, it is locally convex in a neighbourhood of the true value of the parameter vector. By using a novel method of moments estimator as the initial value, the consistent local maximum likelihood estimator (MLE) can be obtained by a gradient descent algorithm. To establish the upper bound for the estimation error of the MLE, we derive a new uniform deviation bound, which is of independent interest. The usefulness of the model and the associated theory are further supported by extensive simulation and the analysis of some real network data sets.

What problem does this paper attempt to address?

### Problems the paper attempts to solve This paper aims to solve two main problems in dynamic network data analysis: 1. **Jointly modeling individual snapshots and temporal dynamics**: - Dynamic network data analysis needs to consider both the states of the network at different time points (i.e., multiple snapshots) and the trends of these states over time. Most of the existing research mainly focuses on static networks, that is, only data at one time point is observed. However, with the increase of multi - time - point network data, how to effectively model and analyze these dynamic changes has become an important research topic. 2. **Capturing the static and dynamic heterogeneity of nodes**: - Network nodes in real life often have different connection tendencies, and these tendencies may change over time. Specifically, some nodes (such as "hub nodes" in social networks) tend to form a large number of connections, while other nodes may have only a small number of connections. In addition, some nodes may be more active in seeking new connections. Therefore, how to capture these static and dynamic heterogeneities simultaneously in one model is a challenge. To address these problems, the paper proposes a new Two - Way Heterogeneity Model (TWHM). This model equips each node with two heterogeneity parameters: one is used to characterize the node's tendency to establish connections with other nodes, and the other is used to distinguish the node's ability to maintain existing connections. Through this method, the model can more comprehensively describe the complex characteristics of dynamic networks. ### Main contributions of the model 1. **Reparameterization**: - By re - parameterizing the general autoregressive network model (AR(1) model), TWHM can simultaneously handle the heterogeneity and dynamic fluctuations of node degrees. This new method can be regarded as an extension of the static β - model in a dynamic framework, and the number of parameters is reduced from \(p(p - 1)\) to \(2p\). 2. **Local maximum likelihood estimation**: - Although the negative log - likelihood function is non - convex, it is locally convex in the neighborhood of the true parameter values. By using a novel moment estimation method as the initial value, a consistent local maximum likelihood estimate (MLE) can be obtained through the gradient descent algorithm. 3. **Upper bound of estimation error**: - In order to establish the upper bound of the MLE estimation error, the paper derives a new uniform bias bound, which is of great significance independent of the number of network parameters. These results are not only applicable to TWHM, but can also be extended to other models with a large number of parameters. 4. **Generality of theoretical results**: - The paper provides general results applicable to functions of the form \(L(\theta)=\frac{1}{p}\sum_{1\leq i\neq j\leq p}l_{i,j}(\theta_i,\theta_j)Y_{i,j}\). These results explore the sparse structure of \(L(\theta)\) and provide a new bound, significantly expanding the application range of empirical processes in models with a fixed number of parameters. ### Summary This paper successfully solves the key problems in dynamic network data analysis by proposing the TWHM model, especially in capturing the static and dynamic heterogeneity of nodes. Through a series of strict theoretical analyses and experimental verifications, the paper demonstrates the effectiveness and wide applicability of this model.

A two-way heterogeneity model for dynamic networks

A Generative Hypergraph Model for Double Heterogeneity

A degree-corrected Cox model for dynamic networks

Modelling heterogeneity in Latent Space Models for Multidimensional Networks

A New Dynamic Community Model For Social Networks

Differential privacy statistical inference for a directed graph network model with covariates

A Markov-switching Hidden Heterogeneous Network Autoregressive Model for Multivariate Time Series Data with Multimodality

Neural Latent Space Model for Dynamic Networks and Temporal Knowledge Graphs

Deterministic, quenched and annealed parameter estimation for heterogeneous network models

Simultaneous estimation and group identification for network vector autoregressive model with heterogeneous nodes

How does homophily shape the topology of a dynamic network?

Bipartite Network Influence Analysis of a Two-Mode Network

Consensus In A Heterogeneous Influence Network

Variable Selection for High-dimensional Nodal Attributes in Social Networks with Degree Heterogeneity

Impact of Heterogeneity on Network Embedding

Nonparametric Multi-group Membership Model for Dynamic Networks

Network Heterogeneity In An Undirected Network

Varying-coefficient models for dynamic networks

Degree Heterogeneity in Higher-Order Networks: Inference in the Hypergraph $\boldsymbolβ$-Model

Two-mode Network Autoregressive Model for Large-Scale Networks

Mixed Membership Stochastic Blockmodels for Heterogeneous Networks