Kinship KG–QA (Hinton)

A lightweight knowledge-graph + question answering (KGQA) resource adapted from the UCI Kinship dataset created by Geoff Hinton.

This dataset provides a small family-tree knowledge graph paired with templated multi-hop QA tasks designed for MultiHop KGQA.

Project: THESEUS
Paper: Theseus in the Graph
Original dataset: UCI Kinship

Key Features

• Small knowledge graph with two family trees
• Human-readable entities and relations
• Templated 1--3 hop QA pairs
• Explicit start entity for navigation-based models
• Gold reasoning paths provided
• Paraphrased questions for robustness evaluation

⚠️ This dataset is distinct from the Australian Kinship dataset (Denham et al.).

---

Contents

1) Knowledge Graph (KG) Triples

KG triples are stored under the kg/ directory.

kg/
├─ orig/
│ ├─ train.txt
│ ├─ dev.txt
│ ├─ test.txt
│ └─ triplets.txt
└─ gender/
├─ train.txt
├─ dev.txt
├─ test.txt
└─ triplets.txt

All triple files follow the format:

head relation tail

• Files are tab-separated.
• Entities and relations are stored using human-readable labels.

File descriptions

File	Description
train.txt	Training triples
dev.txt	Validation triples
test.txt	Test triples
triplets.txt	Full graph used for navigation

Because the Kinship graph is very small, dev.txt and test.txt are identical.

---

1-a) kg/orig/ — Original Kinship Graph

This folder contains the original Kinship triples derived from the dataset.

• These triples represent the core family relationships.
• This graph should be used for navigation-based QA experiments.

---

1-b) kg/gender/ — Extended Graph with Gender Information

This folder contains the same triples as kg/orig/ with additional relations describing gender.

Purpose:

• Provide additional semantic signals when pre-training knowledge graph embeddings (KGE).

Important restrictions:

• The gender relation does not appear in the QA questions.
• Using this graph for navigation may allow models to exploit shortcut paths that produce invalid or nonsensical reasoning chains.

Therefore:

• ✅ Suitable for embedding pre-training
• ❌ Should not be used for reasoning or navigation

---

2) Question Answering (QA) Files

QA datasets are stored under the qa/ directory.

qa/
├─ kinship_qa_2hop.csv
├─ kinship_qa_3hop.csv
└─ kinship_qa_nhop.csv

File	Description
kinship_qa_2hop.csv	Contains only 2-hop questions
kinship_qa_3hop.csv	Contains only 3-hop questions
kinship_qa_nhop.csv	Mixed dataset containing 1–3 hop questions

---

Important Note on 1-Hop Questions

1-hop questions are used only for training in the mixed-hop dataset.

This design choice is motivated by two factors:

• The overall dataset is very small
• Multi-hop reasoning (2–3 hops) requires learning the underlying 1-hop relations

Using 1-hop questions during training helps models learn the base relational structure needed for multi-hop reasoning.

---

3) Original Source Data

Data formats

source/
└─ kinship.data

This file contains the original Kinship dataset from which the KG triples were derived. The source file is redistributed from the UCI Machine Learning Repository under the Creative Commons Attribution 4.0 International (CC BY 4.0) license.

---

Triple example

Victoria	father	Christopher      # (head_id, relation_id, tail_id)

CSV schema

Column	Description
Question-Number	Integer question ID
Question	Natural-language question
Question-Paraphrased	List of paraphrased questions
Source	Start entity (human-readable)
Source-Entity	Encoded start entity (kept for compatibility)
Answer	Gold answer (human-readable)
Answer-Entity	Encoded answer entity
Hops	Reasoning hop count (1 / 2 / 3)
Paths	Gold reasoning path as an ordered list of triples (h, r, t) from source to answer
Paths-Label	Same path using human-readable labels
SplitLabel	Dataset split (train/dev/test)

Compatibility Note

Since this dataset already uses human-readable labels, the following columns are identical:

• Source == Source-Entity
• Answer == Answer-Entity
• Paths == Paths-Label

These columns are preserved for compatibility with other MultiHopKGQA datasets that require entity ID mappings.

Path encoding example

[[Victoria, father, Christopher], [Christopher, wife, Penelope]]

---

Usage notes

• Directionality. Triples are provided as directed; many experiments treat the KG as undirected (mirror edges at load time if needed).
• Embedding sizes. This is a toy graph; small KGE dimensions (e.g., d=12) are sufficient.

---

Quick example

Question: “Who is the wife of the father of Victoria?”
Question-Paraphrased: “What is the name of Victoria's father's wife?”
Start: Victoria
Gold path: (Victoria, father, Christopher) → (Christopher, wife, Penelope)
Answer: Penelope (2-hop)

---

Citation

If you use this resource, please cite the original UCI Kinship entry and our paper introducing the QA setup.

@misc{kinship_55,
  author       = {Hinton, Geoff},
  title        = {{Kinship}},
  year         = {1986},
  howpublished = {UCI Machine Learning Repository},
  note         = {{DOI}: https://doi.org/10.24432/C5WS4D}
}

@article{hernandez2026theseus,
  author = {Hernandez, Eduin E. and Garcia, Luis F. and Askar, Nurassyl and Diaz, Sergio A. and Rini, Stefano},
  title = {Theseus in the Graph: Towards Traceable Multi-Hop Graph Navigation},
  journal = {arXiv preprint arXiv:2609.14528},
  year    = {2026}
}

Dataset Statistics

• Knowledge Graph

  • Entities: 24
  • Relations: 12
  • Triples: 112

• Question Answering

Hop	Total	Train	Dev	Test
1-hop	112	112	0	0
2-hop	248	173	37	38
3-hop	420	294	63	63

Total QA pairs: 780

Note: 1-hop questions are used exclusively for training due to the small size of the dataset.

License

The MultiHop KGQA contributions to this dataset, including the knowledge-graph representation, QA pairs, reasoning paths, dataset splits, and paraphrased questions, are licensed under the Creative Commons Attribution 4.0 International (CC BY 4.0) license.

The original UCI Kinship dataset, including source/kinship.data, is distributed by the UCI Machine Learning Repository under the CC BY 4.0 license.

Please cite both the original Kinship dataset and our work when using this resource. See the Citation section above for the corresponding references.