Conference Agenda

We propose a method to detect communities in multi-relational networks, based on a graph neural network pipeline. The method allows to target areas where communities are consensual over the different modes of the network, which are processed as different networks in the pipeline. This is done by combining the outcomes of multiple simple Graph Neural Networks, applied on each of the graphs representing different forms of interactions between users of the social platform. The method is validated on a synthetic benchmark, as a first step for further improvements. In particular, the flexible architecture of the pipeline allows to swap its subparts and create variants of community detection.

Social network analysis research is typically done on real-world social network data. Together with a research paper, academics are confronted with the question of whether or not to publish the network data underlying their analyses. Here, a problem is that the structure of the network alone might disclose the identity of the nodes, i.e., the individual people present in the network. For example, if in a social network only one person has three friends of which two also know each other, then the ego network structure of this person reveals the person’s identity. Yet, to derive meaningful insights from this person’s social network, information about their ego network structure is inherently necessary. At first glance, the need for protecting people’s privacy is inherently at odds with ensuring reproducibility of social network analysis research. In this work, we discuss the network anonymization problem, which aims to achieve both objectives.

In the network anonymization problem, the goal is to perturb a given social network dataset in such a way that individuals can no longer be uniquely identified based on their surrounding network structure. We discuss how this problem relates to the more generic problem of statistical disclosure control, and cover a number of trade-offs that one needs to consider to address this problem in a social network scenario.

The first is that of the attacker scenario: what de-anonymizing information should we realistically assume is in the hands of an adversary, in case we are considering the structure of the social network as our main object of study? The second is the anonymity-utility-tradeoff: if we alter the network structure to increase anonymity, how can we still make sure that data utility is guaranteed? Utility can be measured in various ways, including preserving structural network properties and ensuring reliable performance on downstream tasks such as centrality analysis. The third is that of balancing computational complexity: depending on the chosen attacker scenario and corresponding measure of node anonymity, efficiently perturbing the network while retaining sufficient data utility can be a challenging computational problem.

We provide an overview of common network anonymity measures and network anonymization algorithms from the literature, and demonstrate a case study on the population-scale social network of the Netherlands, in which we show how various measures and algorithms perform in terms of attaining privacy and utility of the anonymized social network data. We furthermore demonstrate accompanying software that can anonymize a given social network dataset.

The Resilience Network (ResNet, see resnet.me) is a free mobile app together with several free web applications (surveys, scheduling polls, quizzes, etc.) designed to improve individual and group resilience by fostering social relationships and personal development. While primarily focused on improving the "7 life categories" of each app user, ResNet also offers many features for SNA and (health) intervention research.

ResNet collects data through an integrated psychological profiling questionnaire including the Big Five personality traits, through longitudinal sentiment tracking of all 7 life categories, and through interacting with others in the app. The collected data allows researchers to study sentiment dynamics, psychological profiles, and social behaviors. Health interventions such as promoting physical activity, improving nutrition, or leveraging positive outcomes from social interactions are within the scope of ResNet. Other types of interventions can be explored.

In this talk, we will demonstrate ResNet’s basic capabilities, including its web-based survey tool and its psychological profiling features. Using ResNet in research studies will be beneficial for the further development of the app. We especially welcome suggestions and collaborations for the use of ResNet in SNA and public health research.

Synthetic networks are an important tool for simulation studies of social phenomenon such as disease spread and behaviour adoption. Algorithms to generate such networks must be able to incorporate important features of real world social networks such as positive degree assortativity, clustering coefficient, and skewed degree distributions.

We have implemented a network generator that uses social proximity to drive edge creation. It adapts the algorithm of Pasta and colleagues (2014) that was designed to reproduce empirical Facebook friendship networks. Our implementation simplifies and generalises their algorithm, taking it beyond those specific networks.

In our paper, we present this implementation and preliminary results about the effects of algorithm parameters on the structural properties of the generated networks. These structural properties include mean and variation of centrality measures, network distances, clustering, and degree assortativity.

Collecting social network data, particularly complete networks and cognitive social structures, can be challenging. One challenge is that network surveys can be repetitive and tedious for respondents. While interfaces have been developed to make this process more engaging, they are typically limited to larger screens (e.g., laptops or tablets), whereas smartphones are more widely accessible. Another challenge is that complete network data regularly rely on the collection of names before obtaining informed consent, raising ethical concerns.

We introduce SICCEN (Smartphone Interface for Collecting Complete and Ego Networks), which was designed to collect quantitative social network data, including cognitive social structures, in an intuitive, engaging, and ethically responsible manner using smartphones. A key design feature of SICCEN is that connections are visualized by proximity rather than lines, ensuring clarity and scalability on small screens, even when capturing complex cognitive social structures. Another key feature is that when network members participate simultaneously, they can enter their names at the start of the questionnaire, which are then dynamically integrated into other participants’ surveys in real time. This method eliminates the need to collect personal data in advance and automatically excludes non-participants’ names, addressing ethical concerns.

We will discuss the background of SICCEN, its benefits for social network research, how it was developed, its novel features, how we used it for a large-scale school study (N = 1491), and directions for its future development.