Inductive Relation Prediction by Subgraph Reasoning

Komal Teru, Etienne Denis, Will Hamilton
Proceedings of the 37th International Conference on Machine Learning, PMLR 119:9448-9457, 2020.

Abstract

The dominant paradigm for relation prediction in knowledge graphs involves learning and operating on latent representations (i.e., embeddings) of entities and relations. However, these embedding-based methods do not explicitly capture the compositional logical rules underlying the knowledge graph, and they are limited to the transductive setting, where the full set of entities must be known during training. Here, we propose a graph neural network based relation prediction framework, GraIL, that reasons over local subgraph structures and has a strong inductive bias to learn entity-independent relational semantics. Unlike embedding-based models, GraIL is naturally inductive and can generalize to unseen entities and graphs after training. We provide theoretical proof and strong empirical evidence that GraIL can rep-resent a useful subset of first-order logic and show that GraIL outperforms existing rule-induction baselines in the inductive setting. We also demonstrate significant gains obtained by ensembling GraIL with various knowledge graph embedding methods in the transductive setting, highlighting the complementary inductive bias of our method.

Cite this Paper


BibTeX
@InProceedings{pmlr-v119-teru20a, title = {Inductive Relation Prediction by Subgraph Reasoning}, author = {Teru, Komal and Denis, Etienne and Hamilton, Will}, booktitle = {Proceedings of the 37th International Conference on Machine Learning}, pages = {9448--9457}, year = {2020}, editor = {III, Hal Daumé and Singh, Aarti}, volume = {119}, series = {Proceedings of Machine Learning Research}, month = {13--18 Jul}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v119/teru20a/teru20a.pdf}, url = {http://proceedings.mlr.press/v119/teru20a.html}, abstract = {The dominant paradigm for relation prediction in knowledge graphs involves learning and operating on latent representations (i.e., embeddings) of entities and relations. However, these embedding-based methods do not explicitly capture the compositional logical rules underlying the knowledge graph, and they are limited to the transductive setting, where the full set of entities must be known during training. Here, we propose a graph neural network based relation prediction framework, GraIL, that reasons over local subgraph structures and has a strong inductive bias to learn entity-independent relational semantics. Unlike embedding-based models, GraIL is naturally inductive and can generalize to unseen entities and graphs after training. We provide theoretical proof and strong empirical evidence that GraIL can rep-resent a useful subset of first-order logic and show that GraIL outperforms existing rule-induction baselines in the inductive setting. We also demonstrate significant gains obtained by ensembling GraIL with various knowledge graph embedding methods in the transductive setting, highlighting the complementary inductive bias of our method.} }
Endnote
%0 Conference Paper %T Inductive Relation Prediction by Subgraph Reasoning %A Komal Teru %A Etienne Denis %A Will Hamilton %B Proceedings of the 37th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2020 %E Hal Daumé III %E Aarti Singh %F pmlr-v119-teru20a %I PMLR %P 9448--9457 %U http://proceedings.mlr.press/v119/teru20a.html %V 119 %X The dominant paradigm for relation prediction in knowledge graphs involves learning and operating on latent representations (i.e., embeddings) of entities and relations. However, these embedding-based methods do not explicitly capture the compositional logical rules underlying the knowledge graph, and they are limited to the transductive setting, where the full set of entities must be known during training. Here, we propose a graph neural network based relation prediction framework, GraIL, that reasons over local subgraph structures and has a strong inductive bias to learn entity-independent relational semantics. Unlike embedding-based models, GraIL is naturally inductive and can generalize to unseen entities and graphs after training. We provide theoretical proof and strong empirical evidence that GraIL can rep-resent a useful subset of first-order logic and show that GraIL outperforms existing rule-induction baselines in the inductive setting. We also demonstrate significant gains obtained by ensembling GraIL with various knowledge graph embedding methods in the transductive setting, highlighting the complementary inductive bias of our method.
APA
Teru, K., Denis, E. & Hamilton, W.. (2020). Inductive Relation Prediction by Subgraph Reasoning. Proceedings of the 37th International Conference on Machine Learning, in Proceedings of Machine Learning Research 119:9448-9457 Available from http://proceedings.mlr.press/v119/teru20a.html.

Related Material