lex-au-graph

Cross-reference knowledge graph over Australian Commonwealth Acts, built from the lex-au AKN 3.0 XML corpus — the retrieval layer of the AU Legislative Intelligence Stack.

This dataset publishes three files, rebuilt automatically whenever lex-au publishes a corpus update:

• graph.json — the full cross-reference graph (nodes: Acts, Sections, defined terms; edges: contains, defines, ref, mentions), as NetworkX node_link_data JSON. Not tabular — load it directly, not via the dataset viewer:

  import json, networkx as nx
  from huggingface_hub import hf_hub_download
  path = hf_hub_download("cchew/lex-au-graph", "graph.json", repo_type="dataset")
  graph = nx.node_link_graph(json.load(open(path)))
  

• centrality.json — PageRank centrality score per node (a flat {node_id: score} map, ~74k entries). Also not tabular for the same reason.
• complexity.json — per-Act structural complexity metrics (citation density, defined-term density, indeterminate-concept frequency, conditional-statement frequency). One row per Act, viewable in the preview above.

Quoted amendment text. Provisions an amending Act inserts into another Act (lex-au <quotedStructure>) appear in graph.json as Section nodes with provision_type="quoted". They carry no ref edges and no defined terms, are not citation targets, and are excluded from complexity.json and from centrality.json sums, so for an amending Act those figures describe its own provisions only.

Scoped terms and ids. Defined-term nodes carry scope: act, or the eId of the section, Part, Division or Schedule a local definition applies to. When two instruments share an AKN-style work URI, the Act keeps it and the other gets <uri>~<FRL title id>; the only case is SIS Regulations 1994, /akn/au/act/1994/57~F1996B00580.

See github.com/cchew/lex-au-graph for source code, the MCP server, and version history.

Licence

CC BY 4.0. Source legislation is Crown copyright - Commonwealth of Australia; reproduction permitted for non-commercial and research purposes under the PSI Framework. The code that builds this dataset is MIT-licensed.