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
AN - SCOPUS:85048485418
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
A2 - Gangemi, Aldo
A2 - Navigli, Roberto
A2 - Vidal, Maria-Esther
A2 - Hitzler, Pascal
A2 - Troncy, Raphaël
A2 - Hollink, Laura
A2 - Tordai, Anna
A2 - Alam, Mehwish
PB - Springer/Verlag
T2 - 15th International Conference on Extended Semantic Web Conference, ESWC 2018
Y2 - 3 June 2018 through 7 June 2018
ER -