Message Passing Query Embedding

Daza, Daniel; Cochez, Michael

doi:10.48550/arxiv.2002.02406

Cited by 1 publication

(3 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Complex Query Answering. In the complex (multi-hop) query answering setup with logical operators, existing models employ different approaches, e.g., geometric [15,23,38], probabilistic [24,9], neural-symbolic [26,8,5], neural [21,4], and GNN [11,3]. Still, all the approaches are created and evaluated exclusively in the transductive mode where the set of entities does not change at inference time.…”

Section: Related Workmentioning

confidence: 99%

“…Having a uniform featurization mechanism for both seen and unseen entities, it is now possible to apply any previously-transductive complex query answering model with learnable entity embeddings and logical operators [23,11,24,8]. Moreover, it was recently shown [5] that a combination of simple link prediction pre-training and a non-parametric logical executor allows to effectively answer complex FOL queries in the inference-only regime without training on any complex query sample.…”

Section: Nodepiece-qe: Inductive Node Representationmentioning

confidence: 99%

“…The conventional KG completion (link prediction) task can be viewed as a complex query with a single triplet pattern without logical operators, e.g., Citizen(USA, V ), which we also denote as a projection query. To tackle complex queries on incomplete knowledge graphs, query embedding methods are proposed to execute logic operations in the latent space, including variants that employ geometric [15,23,38], probabilistic [24,9], neural-symbolic [26,8,5], neural [21,4], and GNN [11,3] approaches for learning entity, relation, and query representations.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Inductive Logical Query Answering in Knowledge Graphs

Galkin¹,

Zhu²,

Ren³

et al. 2022

Preprint

View full text Add to dashboard Cite

Formulating and answering logical queries is a standard communication interface for knowledge graphs (KGs). Alleviating the notorious incompleteness of realworld KGs, neural methods achieved impressive results in link prediction and complex query answering tasks by learning representations of entities, relations, and queries. Still, most existing query answering methods rely on transductive entity embeddings and cannot generalize to KGs containing new entities without retraining the entity embeddings. In this work, we study the inductive query answering task where inference is performed on a graph containing new entities with queries over both seen and unseen entities. To this end, we devise two mechanisms leveraging inductive node and relational structure representations powered by graph neural networks (GNNs). Experimentally, we show that inductive models are able to perform logical reasoning at inference time over unseen nodes generalizing to graphs up to 500% larger than training ones. Exploring the efficiency-effectiveness trade-off, we find the inductive relational structure representation method generally achieves higher performance, while the inductive node representation method is able to answer complex queries in the inference-only regime without any training on queries and scales to graphs of millions of nodes. Code is available at https://github.com/DeepGraphLearning/InductiveQE.

show abstract

Section: Related Workmentioning

confidence: 99%

Section: Nodepiece-qe: Inductive Node Representationmentioning

confidence: 99%