brandburner/doctorwho-s04-narrative-kg

---
license: cc-by-sa-4.0
task_categories:
- graph-ml
- text-generation
tags:
- narrative
- knowledge-graph
- screenplay
- fabula
- neo4j
- graph-gravity
size_categories:
- 1K<n<10K
---

# Doctor Who - Narrative Knowledge Graph

A rich narrative knowledge graph extracted from *Doctor Who* screenplays using the
[Fabula](https://fabula.productions) pipeline. Contains characters,
locations, objects, organizations, events, themes, and conflict arcs with full
participation semantics and Graph Gravity importance tiers.

## Dataset Overview

| Metric | Value |
|--------|-------|
| Source database | `doctorwho.s04` |
| Type | Season database |
| Episodes | 43 |
| Total nodes | 6,388 |
| Total edges | 22,039 |
| Schema version | 1.1.0 |
| Exported | 2026-04-08 |

### Entity Breakdown

| Type | Count |
|------|-------|
| Act | 117 |
| Agent | 308 |
| ConflictArc | 180 |
| Episode | 43 |
| Event | 1,147 |
| Location | 373 |
| Object | 1,191 |
| Organization | 133 |
| PlotBeat | 1,960 |
| SceneBoundary | 740 |
| Theme | 196 |

### Graph Gravity Tiers

| Tier | Count | Description |
|------|-------|-------------|
| anchor | 33 | Main characters / key locations |
| planet | 307 | Recurring entities |
| asteroid | 1,665 | Minor / one-off entities |

### Relationship Types

`AFFILIATED_WITH`, `BELONGS_TO_EPISODE`, `CALLBACK`, `CAUSAL`, `CHARACTER_CONTINUITY`, `CONTAINS_ACT`, `CONTAINS_BEAT`, `CONTAINS_SCENE`, `ESCALATION`, `EXEMPLIFIES_THEME`, `FORESHADOWING`, `INVOLVED_IN_ARC`, `INVOLVED_WITH`, `IN_EVENT`, `NARRATIVELY_FOLLOWS`, `OCCURS_IN`, `PARTICIPATED_AS`, `PART_OF`, `PART_OF_ACT`, `PART_OF_ARC` ... and 4 more

## Related Datasets

This is a **single-season** dataset containing entities and events as extracted from Season 4 screenplays.

- **Megagraph** (all seasons unified): [brandburner/doctorwho-mega-narrative-kg](https://huggingface.co/datasets/brandburner/doctorwho-mega-narrative-kg)

> **Note:** The megagraph is *not* a simple union of season datasets. Cross-season entities are reconciled through a Global Entity Registry (GER), receiving new canonical UUIDs and distilled descriptions. Graph Gravity tiers are recalculated across all episodes. Use individual season datasets for single-season analysis; use the megagraph for cross-season analysis.

## Files

| File | Description |
|------|-------------|
| `nodes.parquet` | All graph nodes with properties |
| `edges.parquet` | All relationships with properties |
| `positions.parquet` | 3D layout coordinates for visualization |
| `meta.json` | Dataset metadata and entity counts |

## Schema

### Nodes (`nodes.parquet`)

| Column | Type | Description |
|--------|------|-------------|
| `node_id` | string | Unique node identifier (UUID) |
| `primary_label` | string | Node type (Agent, Location, Event, etc.) |
| `name` | string | Display name |
| `description` | string | Foundational description |
| `tier` | string (nullable) | Graph Gravity tier: anchor / planet / asteroid |
| `episode_count` | int (nullable) | Number of distinct episodes entity appears in |
| `first_episode_seq` | int (nullable) | First appearance episode |
| `last_episode_seq` | int (nullable) | Last appearance episode |
| `properties_json` | string | Full node properties as JSON |

### Edges (`edges.parquet`)

| Column | Type | Description |
|--------|------|-------------|
| `source_node_id` | string | Source node UUID |
| `target_node_id` | string | Target node UUID |
| `relationship_type` | string | Relationship type (e.g., PARTICIPATED_AS) |
| `properties_json` | string | Edge properties as JSON |

### Positions (`positions.parquet`)

| Column | Type | Description |
|--------|------|-------------|
| `node_id` | string | Node UUID |
| `x`, `y`, `z` | float | 3D coordinates |
| `size` | float | Node size (Graph Gravity weighted) |
| `r`, `g`, `b` | int | RGB color by entity type |
| `community` | int | Louvain community index |
| `tier` | string (nullable) | Graph Gravity tier |

## Usage

```python
from datasets import load_dataset
import pandas as pd

# Load from HuggingFace
ds = load_dataset("brandburner/doctorwho-s04-narrative-kg")

# Or load parquet directly
nodes = pd.read_parquet("nodes.parquet")
edges = pd.read_parquet("edges.parquet")

# Filter to anchor characters
anchors = nodes[(nodes['primary_label'] == 'Agent') & (nodes['tier'] == 'anchor')]

# Build a NetworkX graph
import networkx as nx
G = nx.DiGraph()
for _, n in nodes.iterrows():
G.add_node(n['node_id'], label=n['primary_label'], name=n['name'])
for _, e in edges.iterrows():
G.add_edge(e['source_node_id'], e['target_node_id'], type=e['relationship_type'])
```

## Citation

```bibtex
@misc{fabula_doctorwho_s04,
title = {Doctor Who Narrative Knowledge Graph},
author = {Fabula Pipeline},
year = {2026},
publisher = {HuggingFace},
howpublished = {\url{https://huggingface.co/datasets/brandburner/doctorwho-s04-narrative-kg}}
}
```

## License

CC BY-SA 4.0