Umeå University's logo

Mapping network flows provides insight into the organization of networks, but even though many real networks are bipartite, no method for mapping flows takes advantage of the bipartite structure. What do we miss by discarding this information and how can we use it to understand the structure of bipartite networks better? The map equation models network flows with a random walk and exploits the information-theoretic duality between compression and finding regularities to detect communities in networks. However, it does not use the fact that random walks in bipartite networks alternate between node types, information worth 1 bit. To make some or all of this information available to the map equation, we developed a coding scheme that remembers node types at different rates. We explored the community landscape of bipartite real-world networks from no node-type information to full node-type information and found that using node types at a higher rate generally leads to deeper community hierarchies and a higher resolution. The corresponding compression of network flows exceeds the amount of extra information provided. Consequently, taking advantage of the bipartite structure increases the resolution and reveals more network regularities.

Detecting significant community structure in networks with incomplete observations is challenging because the evidence for specific solutions fades away with missing data. For example, recent research shows that flow-based community detection methods can highlight spurious communities in sparse undirected and unweighted networks with missing links. Current Bayesian approaches developed to overcome this problem do not work for incomplete observations in weighted and directed networks that describe network flows. To overcome this gap, we extend the idea behind the Bayesian estimate of the map equation for unweighted and undirected networks to enable more robust community detection in weighted and directed networks. We derive an empirical Bayes estimate of the transitions rates that can incorporate metadata information and show how an efficient implementation in the community-detection method Infomap provides more reliable communities even with a significant fraction of data missing.

To measure node importance, network scientists employ centrality scores that typically take a microscopic or macroscopic perspective, relying on node features or global network structure. However, traditional centrality measures such as degree centrality, betweenness centrality, or PageRank neglect the community structure found in real-world networks. To study node importance based on network flows from a mesoscopic perspective, we analytically derive a community-aware information-theoretic centrality score based on network flow and the coding principles behind the map equation: map equation centrality. Map equation centrality measures how much further we can compress the network's modular description by not coding for random walker transitions to the respective node, using an adapted coding scheme and determining node importance from a network flow-based point of view. The information-theoretic centrality measure can be determined from a node's local network context alone because changes to the coding scheme only affect other nodes in the same module. Map equation centrality is agnostic to the chosen network flow model and allows researchers to select the model that best reflects the dynamics of the process under study. Applied to synthetic networks, we highlight how our approach enables a more fine-grained differentiation between nodes than node-local or network-global measures. Predicting influential nodes for two different dynamical processes on real-world networks with traditional and other community-aware centrality measures, we find that activating nodes based on map equation centrality scores tends to create the largest cascades in a linear threshold model.

Node similarity scores constitute a foundation for machine learning in graphs. Besides clustering, node classification, and anomaly detection, they are a basis for link prediction with critical applications in biological systems, information networks, and recommender systems. Recent works on link prediction use vector space embeddings to calculate node similarities. While these methods can provide good performance in undirected networks, they have several disadvantages: limited interpretability, problem-specific hyperparameter tuning, manual model fitting through dimensionality reduction, and poor performance of symmetric similarities in directed link prediction. To address these issues, we propose MapSim, a novel information-theoretic approach to assess node similarities based on modular compression of network flows. Different from vector space embeddings, MapSim represents nodes in a discrete, non-metric space of communities and yields asymmetric similarities suitable to predict directed and undirected links in an unsupervised fashion. The resulting similarities can be explained based on a network's hierarchical modular organisation, facilitating interpretability. MapSim naturally accounts for Occam's razor, leading to parsimonious representations of clusters at multiple scales. Addressing unsupervised link prediction, we compare MapSim to popular embedding-based algorithms across 47 data sets of networks from a few hundred to hundreds of thousands of nodes and millions of links. Our analysis shows that MapSim's average performance across all networks is more than 7% higher than its closest competitor, outperforming all embedding methods in 14 of the 47 networks, and a more than 33% better worst-case performance. Our method demonstrates the potential of compression-based approaches in graph representation learning, with promising applications in other graph learning tasks.