The microscale organization of directed hypergraphs

Many real-world complex systems are characterized by non-pairwise – higher-order – interactions among system’s units, and can be effectively modeled as hypergraphs. Directed hypergraphs distinguish between source and target sets within each hyperedge, and allow to account for the directional flow of information between nodes. Here, we provide a framework to characterize the structural organization of directed higher-order networks at their microscale. First, we extract the fingerprint of a directed hypergraph, capturing the frequency of hyperedges with a certain source and target sizes, and use this information to compute differences in higher-order connectivity patterns among real-world systems. Then, we formulate reciprocity in hypergraphs, including exact, strong, and weak definitions, to measure to which extent hyperedges are reciprocated. Finally, we extend motif analysis to identify recurring interaction patterns and extract the building blocks of directed hypergraphs. We validate our framework on empirical datasets, including Bitcoin transactions, metabolic networks, and citation data, revealing structural principles behind the organization of real-world systems.

💡 Research Summary

**
The paper presents a comprehensive framework for analysing the microscale structure of directed hypergraphs, which are mathematical objects capable of representing higher‑order (non‑pairwise) interactions with an explicit source‑target directionality. After formally defining a simple directed hypergraph (H=(V,E,s,t)) where each hyperedge (e) consists of a non‑empty, disjoint source set (s(e)) and target set (t(e)), the authors identify four canonical interaction types: one‑to‑one, one‑to‑many, many‑to‑one and many‑to‑many.

Hyperedge signature vectors
For each empirical dataset the authors construct a “hyperedge signature vector” (v) that records the frequency of hyperedges for every combination of source size (|S|) and target size (|T|) (restricted to total cardinality ≤ 6 to keep the representation tractable). These vectors serve as fingerprints of the higher‑order connectivity pattern of a system. By computing weighted Kendall‑τ correlations between the vectors of different datasets and applying hierarchical agglomerative clustering, the authors reveal that datasets from the same domain (e.g., metabolic reactions and citation networks) cluster together, whereas domains with different interaction styles (e‑mail vs. Q&A forums) are anti‑correlated.

Source/target overlap analysis
The paper then quantifies how often nodes appear together in source or target sets. For each node the average pairwise overlap among its incident source sets and among its incident target sets is measured, and a null model preserving the degree‑distribution is used to compute a z‑score. A z‑score ≥ 2 indicates statistically significant excess overlap (the node repeatedly participates with the same co‑senders or co‑receivers), while ≤ ‑2 signals unusually low overlap (high diversity). Results show strong excess overlap in metabolic and citation networks (reflecting recurring enzyme‑product or co‑author groups), moderate excess in e‑mail target sets (broadcast‑like behaviour), and low overlap in Bitcoin transactions (novel counterparties).

Higher‑order reciprocity
A major contribution is the introduction of three levels of reciprocity for directed hypergraphs:

1. Exact reciprocity – a single hyperedge ((S,T)) has a counterpart ((T,S)).
2. Strong reciprocity – a collection of hyperedges collectively reverses the interaction, possibly involving additional external nodes.
3. Weak reciprocity – at least one node in the target set reciprocates with at least one node in the source set (pairwise reverse link).

For each definition the observed reciprocity (r) is compared to the expectation (\langle r\rangle_{NM}) under a configuration‑model null, and a normalized score (\rho_{NM} = (r-\langle r\rangle_{NM})/(1-\langle r\rangle_{NM})) is reported. Metabolic and citation networks exhibit high strong and weak reciprocity, whereas e‑mail and Bitcoin show low values, indicating that bidirectional higher‑order exchanges are domain‑specific.

Directed hypergraph motif analysis
Finally, the authors extend classic graph motif analysis to directed hypergraphs. They enumerate all small directed hypergraph substructures (up to four nodes, various source/target cardinalities) and count their occurrences in each dataset. Frequently observed motifs include:

• Feedback loops – many‑to‑many hyperedges that appear in reciprocal pairs, suggesting cyclic information flow.
• Broadcast‑receive patterns – one‑to‑many together with many‑to‑one, prominent in e‑mail and Bitcoin, reflecting a node that both disseminates to many and receives from many.
• Multi‑collaborative motifs – many‑to‑many patterns abundant in metabolic and citation data, reflecting simultaneous participation of multiple enzymes or authors.

Overall, the study delivers a set of quantitative tools—hyperedge signatures, overlap z‑scores, three‑tier reciprocity, and directed hypergraph motifs—that together enable a fine‑grained comparison of directed higher‑order systems. By applying these tools to diverse real‑world datasets, the authors uncover domain‑specific structural principles (e.g., redundancy in metabolic pathways, broadcast nature of e‑mail, novelty in Bitcoin transactions) and lay the groundwork for future investigations of dynamics (diffusion, synchronization, evolutionary processes) on directed hypergraphs, where traditional pairwise‑only analyses would miss crucial higher‑order effects.