README.md

metadata

license: mit
task_categories:
  - graph-ml
language:
  - en
tags:
  - Graph
  - GNN
  - KG
pretty_name: AutoScenarioQA
size_categories:
  - n<1K

AutoScenarioQA Dataset

t_reference_questions.csv is a question-answer dataset for reasoning over the automotive-market scenario graph t_scenario_graph. It contains 680 question records across three difficulty levels that target information-extraction and reasoning tasks on the scenario graph.

What the QA Is Used For

The QA records model questions that members of a strategic foresight team may ask while developing, reviewing, or challenging automotive-industry scenarios. They can be used to evaluate a graph-query system, knowledge-graph assistant, or graph-reasoning model against expected answers and supporting evidence.

The questions cover both scenario content and graph structure. Scenario-focused questions address possible policy, market, technology, energy, and supply-chain developments and their consequences. Structure-focused questions address direct relationships between events, node labels and properties, chronological paths, branches, cycles, isolated nodes, predecessor and successor nodes, and filtered node sets.

This makes the dataset useful for testing whether graph-based QA systems distinguish plausible narrative answers from answers grounded in graph topology and node attributes.

CSV Schema

Column	Description
user_group	Intended user role, such as Scenario Team Member.
reference_question	Natural-language question about the scenario graph.
category	Question category: Attribute-Based Filtering, Graph Navigation, Node Properties, Node and Label Identification, Path and Sequence Analysis, Relationships and Connections, Scenario-Specific Questions, Temporal Reasoning, or Topological Queries.
level	Difficulty level: 1, 2, or 3.
reference_answer	Expected answer to the question.
reference_evidence	Supporting graph evidence, such as node IDs, edge direction, or property names.
all_node_ids	Relevant node IDs where provided.

Graph Schema and Arrows.app Example

The scenario graph used by the questions is provided in t_scenario_graph.json. The accompanying Python scripts can generate the AutoScenarioQA dataset from the supplied graph, making the dataset adaptable to other scenario-graph realizations.

The graph contains 79 nodes and 109 relationships.

Schema Elements

The graph has three top-level elements:

Element	Purpose
nodes	Scenario containers, regional splits, events, and effects.
relationships	Directed links between node IDs.
style	Default graph display settings.

Node Schema

Node objects use the following fields:

Field	Meaning
id	Unique n<number> identifier used by relationships.
position	Graph canvas coordinates: x and y.
caption	Human-readable name shown in the graph.
labels	One or more semantic categories for the node.
properties	Scenario attributes and vehicle sales volume where applicable.
style	Optional display overrides, including node-color, border-color, and caption-color.

The node labels present in the published graph are:

• Scenario
• Region-Split
• Certain Event
• Uncertain Event
• Less Certain Event
• Relevant Event
• Less Relevant Event
• Effect

Scenario labels a scenario branch, and Region-Split assigns it to a regional branch. Event nodes can combine one certainty label (Certain Event, Uncertain Event, or Less Certain Event) with one relevance label (Relevant Event or Less Relevant Event). Effect identifies an outcome node.

Node Properties

The published graph uses the following property keys where applicable:

Property	Meaning
START	Start date or start period.
END	End date or end period.
t	Temporal value or period.
BEV-SHARE	Battery-electric-vehicle share.
PremiumSegmentWithChinesePlayers	Premium-segment vehicle sales volume involving Chinese competitors.
TraditionalPremiumSegment	Traditional premium-segment vehicle sales volume.
TotalIndustry	Total-industry vehicle sales volume.

Relationship Schema

Each relationship uses id, fromId, toId, type, properties, and style.

fromId is the source node and toId is the target node. The relationship types present in the published graph are:

Type	Meaning
isIn	Places a scenario in a regional split.
hasEvent	Connects a scenario to an event in that scenario.
isPriorTo	States that the source event occurs before the target event.
hasEffect	Connects an event to an effect node.