Modeling Relational Data with Graph Convolutional Networks

Michael Schlichtkrull, Thomas N. Kipf, Peter Bloem, Rianne van den Berg, Ivan Titov, Max Welling

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

Knowledge graphs enable a wide variety of applications, including question answering and information retrieval. Despite the great effort invested in their creation and maintenance, even the largest (e.g., Yago, DBPedia or Wikidata) remain incomplete. We introduce Relational Graph Convolutional Networks (R-GCNs) and apply them to two standard knowledge base completion tasks: Link prediction (recovery of missing facts, i.e. subject-predicate-object triples) and entity classification (recovery of missing entity attributes). R-GCNs are related to a recent class of neural networks operating on graphs, and are developed specifically to handle the highly multi-relational data characteristic of realistic knowledge bases. We demonstrate the effectiveness of R-GCNs as a stand-alone model for entity classification. We further show that factorization models for link prediction such as DistMult can be significantly improved through the use of an R-GCN encoder model to accumulate evidence over multiple inference steps in the graph, demonstrating a large improvement of 29.8% on FB15k-237 over a decoder-only baseline.

LanguageEnglish
Title of host publicationThe Semantic Web - 15th International Conference, ESWC 2018, Proceedings
PublisherSpringer/Verlag
Pages593-607
Number of pages15
ISBN (Print)9783319934167
DOIs
StatePublished - 2018
Event15th International Conference on Extended Semantic Web Conference, ESWC 2018 - Heraklion, Greece
Duration: 3 Jun 20187 Jun 2018

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume10843 LNCS
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference15th International Conference on Extended Semantic Web Conference, ESWC 2018
CountryGreece
CityHeraklion
Period3/06/187/06/18

Fingerprint

Graph in graph theory
Modeling
Recovery
Information retrieval
Factorization
Knowledge Base
Neural networks
Question Answering
Prediction
Accumulate
Encoder
Predicate
Information Retrieval
Completion
Baseline
Maintenance
Attribute
Model
Neural Networks
Demonstrate

Cite this

Schlichtkrull, M., Kipf, T. N., Bloem, P., van den Berg, R., Titov, I., & Welling, M. (2018). Modeling Relational Data with Graph Convolutional Networks. In The Semantic Web - 15th International Conference, ESWC 2018, Proceedings (pp. 593-607). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 10843 LNCS). Springer/Verlag. DOI: 10.1007/978-3-319-93417-4_38
Schlichtkrull, Michael ; Kipf, Thomas N. ; Bloem, Peter ; van den Berg, Rianne ; Titov, Ivan ; Welling, Max. / Modeling Relational Data with Graph Convolutional Networks. The Semantic Web - 15th International Conference, ESWC 2018, Proceedings. Springer/Verlag, 2018. pp. 593-607 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).
@inproceedings{d5f424635cc0477bade9d24490fc3e1c,
title = "Modeling Relational Data with Graph Convolutional Networks",
abstract = "Knowledge graphs enable a wide variety of applications, including question answering and information retrieval. Despite the great effort invested in their creation and maintenance, even the largest (e.g., Yago, DBPedia or Wikidata) remain incomplete. We introduce Relational Graph Convolutional Networks (R-GCNs) and apply them to two standard knowledge base completion tasks: Link prediction (recovery of missing facts, i.e. subject-predicate-object triples) and entity classification (recovery of missing entity attributes). R-GCNs are related to a recent class of neural networks operating on graphs, and are developed specifically to handle the highly multi-relational data characteristic of realistic knowledge bases. We demonstrate the effectiveness of R-GCNs as a stand-alone model for entity classification. We further show that factorization models for link prediction such as DistMult can be significantly improved through the use of an R-GCN encoder model to accumulate evidence over multiple inference steps in the graph, demonstrating a large improvement of 29.8{\%} on FB15k-237 over a decoder-only baseline.",
author = "Michael Schlichtkrull and Kipf, {Thomas N.} and Peter Bloem and {van den Berg}, Rianne and Ivan Titov and Max Welling",
year = "2018",
doi = "10.1007/978-3-319-93417-4_38",
language = "English",
isbn = "9783319934167",
series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",
publisher = "Springer/Verlag",
pages = "593--607",
booktitle = "The Semantic Web - 15th International Conference, ESWC 2018, Proceedings",

}

Schlichtkrull, M, Kipf, TN, Bloem, P, van den Berg, R, Titov, I & Welling, M 2018, Modeling Relational Data with Graph Convolutional Networks. in The Semantic Web - 15th International Conference, ESWC 2018, Proceedings. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 10843 LNCS, Springer/Verlag, pp. 593-607, 15th International Conference on Extended Semantic Web Conference, ESWC 2018, Heraklion, Greece, 3/06/18. DOI: 10.1007/978-3-319-93417-4_38

Modeling Relational Data with Graph Convolutional Networks. / Schlichtkrull, Michael; Kipf, Thomas N.; Bloem, Peter; van den Berg, Rianne; Titov, Ivan; Welling, Max.

The Semantic Web - 15th International Conference, ESWC 2018, Proceedings. Springer/Verlag, 2018. p. 593-607 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 10843 LNCS).

Research output: Chapter in Book/Report/Conference proceedingConference contribution

TY - GEN

T1 - Modeling Relational Data with Graph Convolutional Networks

AU - Schlichtkrull,Michael

AU - Kipf,Thomas N.

AU - Bloem,Peter

AU - van den Berg,Rianne

AU - Titov,Ivan

AU - Welling,Max

PY - 2018

Y1 - 2018

N2 - Knowledge graphs enable a wide variety of applications, including question answering and information retrieval. Despite the great effort invested in their creation and maintenance, even the largest (e.g., Yago, DBPedia or Wikidata) remain incomplete. We introduce Relational Graph Convolutional Networks (R-GCNs) and apply them to two standard knowledge base completion tasks: Link prediction (recovery of missing facts, i.e. subject-predicate-object triples) and entity classification (recovery of missing entity attributes). R-GCNs are related to a recent class of neural networks operating on graphs, and are developed specifically to handle the highly multi-relational data characteristic of realistic knowledge bases. We demonstrate the effectiveness of R-GCNs as a stand-alone model for entity classification. We further show that factorization models for link prediction such as DistMult can be significantly improved through the use of an R-GCN encoder model to accumulate evidence over multiple inference steps in the graph, demonstrating a large improvement of 29.8% on FB15k-237 over a decoder-only baseline.

AB - Knowledge graphs enable a wide variety of applications, including question answering and information retrieval. Despite the great effort invested in their creation and maintenance, even the largest (e.g., Yago, DBPedia or Wikidata) remain incomplete. We introduce Relational Graph Convolutional Networks (R-GCNs) and apply them to two standard knowledge base completion tasks: Link prediction (recovery of missing facts, i.e. subject-predicate-object triples) and entity classification (recovery of missing entity attributes). R-GCNs are related to a recent class of neural networks operating on graphs, and are developed specifically to handle the highly multi-relational data characteristic of realistic knowledge bases. We demonstrate the effectiveness of R-GCNs as a stand-alone model for entity classification. We further show that factorization models for link prediction such as DistMult can be significantly improved through the use of an R-GCN encoder model to accumulate evidence over multiple inference steps in the graph, demonstrating a large improvement of 29.8% on FB15k-237 over a decoder-only baseline.

UR - http://www.scopus.com/inward/record.url?scp=85048485418&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85048485418&partnerID=8YFLogxK

U2 - 10.1007/978-3-319-93417-4_38

DO - 10.1007/978-3-319-93417-4_38

M3 - Conference contribution

SN - 9783319934167

T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

SP - 593

EP - 607

BT - The Semantic Web - 15th International Conference, ESWC 2018, Proceedings

PB - Springer/Verlag

ER -

Schlichtkrull M, Kipf TN, Bloem P, van den Berg R, Titov I, Welling M. Modeling Relational Data with Graph Convolutional Networks. In The Semantic Web - 15th International Conference, ESWC 2018, Proceedings. Springer/Verlag. 2018. p. 593-607. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). Available from, DOI: 10.1007/978-3-319-93417-4_38