HIF

HIF (Hypergraph Interchange Format) is a JSON-based standard for storing higher-order networks, that is, hypergraphs in which a single edge may connect any number of nodes rather than exactly two. It was published in 2025 in the journal Network Science by a cross-institution group (Martín Coll, Cliff Joslyn, Nicholas Landry, Quintino Francesco Lotito, Audun Myers, Joshua Pickard, Brenda Praggastis, Przemysław Szufel, and others) drawn from the maintainer teams of the major hypergraph libraries. It is governed openly by the HIF-org organization on GitHub, which hosts the canonical JSON Schema plus tutorials and validators.

The data model is built around an incidence list rather than an adjacency structure. A HIF document has up to five top-level fields: the required incidences, plus optional network-type, metadata, nodes, and edges. Each record in incidences is a node-edge pair such as {"edge": 1, "node": 3}; all records sharing an edge id together define one hyperedge, so the classic difficulty of serializing variable-arity edges is sidestepped entirely. Incidences may also carry a weight, a direction of "head" or "tail", and free-form attrs. The separate nodes and edges arrays let isolated nodes and empty hyperedges exist, and attach per-element weights and attributes; metadata holds network-level fields like author or dataset name.

The network-type enum selects "undirected" (the default), "directed" (using the incidence direction field), or "asc" for an abstract simplicial complex. Attributes are supported at three levels (node, edge, and, crucially, incidence), which enables edge-dependent node weights used in higher-order random walks. HIF is deliberately serialization-simple and human-readable; the authors note that Parquet or CSV would be more compact but sacrifice readability and rich attributes.

HIF's reason for existing is interoperability: it is designed as the exchange layer between hypergraph tools that otherwise use incompatible internal formats, and it has native read/write support in XGI (via read_hif/write_hif and the XGI-DATA collection), HyperNetX, HypergraphX, SimpleHypergraphs.jl, and the Hypergraph Analysis Toolbox. It complements rather than replaces pairwise-graph formats like GraphML, GEXF, and Pajek, which cannot express hyperedges directly. Limitations are honestly documented: the schema validates structure but not semantics (it cannot enforce the downward-closure invariant of a simplicial complex), and multiplex, temporal, signed, and ordered hypergraphs are only implicitly representable through attrs pending explicit schema extensions.