---
pretty_name: EIS Knowledge Graph Triples
license: other
language:
- en
tags:
- knowledge-graph
- rdf
- environmental-impact-statements
- nepa
configs:
- config_name: triples
  default: true
  data_files: triples.parquet
- config_name: documents
  data_files: documents.parquet
- config_name: entities
  data_files: entities.parquet
---

# EIS Knowledge Graph Triples

**DRAFT**

This project is representative of a "working in the open" ethos. It should not be used for training any
machine-learning algorithm. What it should be used for is prototyping a vision of how future students will interact
with collections.

RDF triples extracted by Claude from 1,493 OCR'd U.S. Environmental Impact Statements (mostly 1970s-era scans),
plus cross-document entity resolution and EIS-to-EIS links.

Facts were extracted automatically and should be treated as leads to verify against the evidence quotes and the
source documents.

## Files

| File | Rows | Contents |
|---|---:|---|
| `triples.parquet` | 883,344 | One row per quad (subject, predicate, object, named graph) |
| `documents.parquet` | 1,493 | One row per EIS: barcode, label, source URI |
| `entities.parquet` | 118,096 | One row per entity: URI, type, label, alternate labels |
| `graph.trig` | — | The full RDF dataset in TriG format (named graphs) |

## Named graphs

- `id:graph/doc/<barcode>` (547,874 quads) — facts asserted by one EIS. The graph name is the provenance.
- `id:graph/entities` (331,527) — `rdf:type`, `rdfs:label` and `skos:altLabel` for every entity.
- `id:graph/doclinks` (3,823) — EIS-to-EIS relations, and `owl:sameAs` links for references that resolve to an EIS in the set.
- `id:graph/ontology` (120) — class and property declarations.

## `triples` columns

| Column | Description |
|---|---|
| `graph` | Named graph |
| `barcode` | Source EIS barcode (null outside per-document graphs) |
| `subject`, `subject_label` | Subject URI and its label |
| `predicate` | Predicate URI |
| `object`, `object_label` | Object URI or literal value, and its label when it is an entity |
| `object_is_literal` | True when `object` is a literal value |
| `evidence` | Verbatim quotes from the source document supporting the relation |

URIs are shortened: `eis:` = `https://example.org/eis/` (vocabulary) and `id:` = `https://example.org/eis/id/`
(instances).

## Usage

```python
from datasets import load_dataset

triples = load_dataset("aerithnetzer/eis-graph-triples", "triples", split="train")
docs = load_dataset("aerithnetzer/eis-graph-triples", "documents", split="train")
```

```python
from pyoxigraph import Store, RdfFormat
store = Store()
store.bulk_load(path="graph.trig", format=RdfFormat.TRIG)
```