brandburner/star-trek-tng-narrative-kg-season-1

Dataset Card for Star Trek: The Next Generation Season 1 Narrative Knowledge Graph

Dataset Description

Dataset Summary

A comprehensive narrative knowledge graph extracted from all 26 episodes of Star Trek: The Next Generation Season 1. This dataset represents characters, locations, objects, organizations, events, and their complex relationships as a labeled property graph (LPG), suitable for graph machine learning, visualization, and narrative analysis research.

Important: This dataset contains multiple configurations. Load specific parts:

from datasets import load_dataset

# Load nodes
nodes = load_dataset("brandburner/fabula-tng-season1", "nodes")

# Load edges with rich narrative properties
edges = load_dataset("brandburner/fabula-tng-season1", "edges")

# Load 3D positions
positions = load_dataset("brandburner/fabula-tng-season1", "positions")

Supported Tasks

• Graph Machine Learning: Node classification, link prediction, graph embeddings
• Narrative Analysis: Story structure analysis, character arc tracking, plot development
• Knowledge Graph Completion: Entity resolution, relationship inference
• Graph Visualization: 3D interactive narrative visualization with pre-computed positions
• Question Answering: Graph-based QA over narrative content
• Recommendation: Character and episode similarity analysis

Languages

• English (screenplay text and descriptions)

Dataset Structure

Data Instances

Example node (Captain Picard):

{
  "node_id": "agent_4fa456b0c9d2",
  "primary_label": "Agent",
  "name": "Jean-Luc Picard",
  "description": "Captain of the USS Enterprise, a thoughtful and diplomatic leader...",
  "episode": 1,
  "act": 1,
  "scene": "3",
  "properties_json": "{\"canonical_name\":\"Jean-Luc Picard\",\"character_type\":\"main\",...}"
}

Example edge (participation relationship with rich narrative properties):

{
  "source_node_id": "agent_4fa456b0c9d2",
  "target_node_id": "event_7b3c892f1a56",
  "relationship_type": "PARTICIPATED_AS",
  "properties_json": "{\"emotional_state_at_event\":\"From contemplative curiosity to controlled anger\",\"observed_status\":\"Captain Jean-Luc Picard commands the USS Enterprise with calm authority...\",\"goals_at_event\":[\"Ensure crew safety\",\"Challenge Q's ultimatum\"],\"beliefs_at_event\":[\"Humanity deserves a fair chance\"]}"
}

Data Fields

Nodes Dataset

• node_id: Unique identifier (UUID)
• primary_label: Entity type (Agent/Location/Object/Organization/Event/SceneBoundary/Episode)
• name: Canonical name or title
• description: Detailed description
• episode: Episode number (1-26)
• act: Act number within episode
• scene: Scene identifier
• properties_json: Full properties as JSON string

Edges Dataset

• source_node_id: Source entity UUID
• target_node_id: Target entity UUID
• relationship_type: Type of relationship
• properties_json: Rich relationship properties as JSON (73% of edges have properties)
  • PARTICIPATED_AS: emotional_state, goals, beliefs, observed_status, traits
  • INVOLVED_WITH: involvement description, status before/after
  • IN_EVENT: atmosphere, significance, environmental details
  • OWNS: ownership type, confidence, inference reason

Positions Dataset

• node_id: Entity UUID
• x, y, z: 3D coordinates (computed via node2vec→UMAP)
• size: Visual size based on degree centrality
• r, g, b: RGB color values by entity type
• community: Community cluster ID (Louvain algorithm)

Data Splits

Single split containing the complete graph for all 26 episodes.

Dataset Creation

Curation Rationale

Created to provide researchers with a rich, structured narrative dataset that captures the complex relationships and story elements in a major science fiction series. Enables research at the intersection of NLP, knowledge graphs, and computational narratology.

Source Data

Initial Data Collection

Original screenplay texts from Star Trek: The Next Generation Season 1 (1987-1988).

Annotation Process

Automated extraction using the Fabula V2 pipeline:

1. Parsing: Hybrid regex/LLM screenplay parsing
2. Decomposition: Hierarchical structure extraction (Episodes→Acts→Scenes)
3. Entity Synthesis: LLM-powered entity identification and description
4. Relationship Extraction: Semantic relationship establishment
5. Harmonization: Cross-episode entity resolution and deduplication
6. Enrichment: Property and description enhancement

Personal and Sensitive Information

The dataset contains only fictional character names and narrative elements from a television series. No personal information about real individuals is included.

Considerations for Using the Data

Social Impact of Dataset

This dataset promotes research in:

• Computational narrative understanding
• Graph-based representation learning
• Digital humanities and cultural studies
• Accessible visualization of complex narratives

Discussion of Biases

The dataset reflects the narrative choices and cultural perspectives present in the original 1987-1988 television series. Researchers should be aware of potential period-specific representations.

Other Known Limitations

• Limited to Season 1 (26 episodes)
• Entity extraction may have minor inconsistencies
• Some relationships may be inferred rather than explicit
• Temporal ordering within scenes is approximate

Additional Information

Dataset Curators

Created using the Fabula V2 narrative analysis pipeline.

Licensing Information

Dataset structure and extracted knowledge graph: CC BY-NC 4.0 Original content: © Paramount Pictures Corporation

Citation Information

@dataset{fabula_tng_s1_2025,
  title={Star Trek: The Next Generation Season 1 - Narrative Knowledge Graph},
  author={Fabula V2 Pipeline},
  year={2025},
  publisher={HuggingFace},
  url={https://huggingface.co/datasets/[username]/fabula-tng-season1}
}

Contributions

Thanks to the Fabula V2 development team and the open-source communities behind Neo4j, ChromaDB, and BAML.