Prerequisites: Some experience with R and familiarity with descriptive network concepts and statistical methods for network analysis in the R/statnet platform is expected. This workshop also assumes familiarity with ERGMs. Synopsis: This workshop will provide an introduction to modeling of network dynamics in continuous time using ERGM generating processes (EGPs). The exponential family random graph models (ERGMs) are a widely used framework for describing graph distributions, allowing flexible and parsimonious specification of both inhomogeneity (i.e., some ties are more likely than others) and dependence (i.e., some ties depend on others). EGPs complement ERGMs by providing ways of specifying continuous time dynamics whose long-run behavior recapitulates a specified ERGM distribution - thus allowing for dynamic network models that are consistent with specific cross-sectional behavior. In this session, we will begin with an overview of known classes of EGPs, with an eye to understanding the types of dynamic behavior embodied by each (and where they might be appropriate as empirical models). We will then discuss simulation and calibration of EGPs within the R/statnet platform, using the ergmgp package. We will show examples of the use of EGPs to generate dynamics consistent with cross-sectional network data combined with information on tie durations, including continuous time generalizations of the separable temporal ERGMs (STERGMs). Attendees are expected to have had some prior exposure to R and statnet, and completion of the statnet ERGM workshop session is strongly suggested as preparation for this session (as we will make extensive use of the ergm package). statnet is a collection of packages for the R statistical computing system that supports the representation, manipulation, visualization, modeling, simulation, and analysis of relational data. statnet packages are contributed by a team of volunteer developers, and are made freely available under the GNU Public License. These packages are written for the R statistical computing environment, and can be used with any computing platform that supports R (including Windows, Linux, and Mac). statnet packages can be used to handle a wide range of simulation and analysis tasks, including support for large networks, statistical network models, network dynamics, and missing data.

Hyperlink prediction, a natural extension of link prediction in graphs, focuses on inferring missing hyperlinks in hypergraphs, where a hyperlink can connect more than two nodes. This technique has diverse applications across systems such as bibliometric networks, chemical reaction networks, social communication networks, and protein-protein interaction networks (among others). In this workshop, we provide a systematic and comprehensive demonstration of hyperlink prediction using Python, primarily leveraging the PyTorch library. We will explore three structural similarity-based methods (Common Neighbors, Katz Index, and Resource Allocation), a probability-based method (Node2Vec, based on random walks), and a deep learning-based method (CHESHIRE: Chebyshev Spectral Hyperlink Predictor). To evaluate the performance of hyperlink prediction, we will use a range of metrics, including F1 scores, ROC AUC, accuracy, precision, recall, log loss, and Matthews correlation coefficient—metrics widely utilized in machine learning. Additionally, we will discuss how hyperlink prediction extends to temporal hypergraphs. To compare, benchmark, and evaluate the hyperlink prediction methods, we will use a selection of well-known or randomly generated medium-sized networks. This is a hands-on computational workshop, and participants should have some prior knowledge of Python. All computations will be performed in Jupyter Notebooks, which will be made available on GitHub before the workshop.

Advances in information technology have increased the availability of time-stamped relational data such as those produced by email exchanges or interaction through social media. Whereas the associated information flows could be aggregated into cross-sectional panels, the temporal ordering of the interactions frequently contains information that requires new ideas for the analysis of continuous-time interactions, subject to both endogenous and exogenous influences. The Relational Event Model (REM) has turned out to be a versatile framework that has allowed further methodological extensions to address a multitude of applied demands: how to deal with non-linear and time-varying effects, how to account for network heterogeneity, how to analyze relational hypergraphs, how to address goodness-of-fit. In this short course, we introduce the REM, define its core properties, and discuss why and how it has been considered useful in empirical research. Then we will focus on how new applications have pushed the development of relational event modelling forward. 1. Introduction to REMs If a process consists of a sequence of temporally ordered events involving a sender and a receiver, such as email communication or bank transactions, the REM can be used to identify drivers of this process. It is based on event history modelling, in particular the Cox model, which allows for convenient and efficient estimation. 2. Mixed effect additive REMs We show how to extend REM formulations with non-linear specifications of endogenous effects, as well as time-varying influence of covariates on the event rate. We explain how the incorporation of random effects can uncover latent heterogeneity associated with individual actors or groups of them. Furthermore, we will describe a general method to assess the goodness-of-fit of such models. 3. Modelling relational hyperevents We will discuss "polyadic" social interaction processes in which events can connect varying and potentially large numbers of nodes simultaneously. Examples of such polyadic events (or "hyperevents") include meeting events or social gatherings, multicast (i.e., "one-to-many") communication events such as emails in which one actor sends the same message to several receivers, co-offending, or scientific collaboration (e.g. co-authoring and citation networks). The workshop will feature a mix short explanatory sessions with hands-on computer practicals. Participants are encouraged to bring their own laptop with Rstudio pre-installed. The workshop is targeted at participants interested in statistical modelling of networks based on relational event data, with a specific focus on non-linear, time-varying and random effects, and polyadic interaction events. The software eventnet together with R-package mgcv will be explained and used in the context of this tutorial. Additional reading material will be made available to the participants beforehand.