Modeling Relational Data with Graph Convolutional Networks

Schlichtkrull, Michael Sejr; Kipf, Thomas; Bloem, Peter; Berg, Rianne van den; Titov, Ivan; Welling, Max

doi:10.48550/arxiv.1703.06103

Cited by 69 publications

(125 citation statements)

References 0 publications

Supporting

Mentioning

122

Contrasting

Order By: Relevance

“…include in a particular query. Notably, our results differ from those of Schlichtkrull et al (2017), who found performance decreases with increasing node degree in a graph convolution-based model, indicating that our approach is promising to pursue in graphs with high mean node degree.…”

Section: Scaling Propertiescontrasting

confidence: 99%

“…The approach is inspired by computational modeling of biological neural architectures for knowledge representation (Crawford et al, 2015), and is related to KBC methods based on convolution of graph neighborhoods (Schlichtkrull et al, 2017;Dettmers et al, 2018;Nguyen et al, 2018), in which inference is performed over representations of aggregated entity neighborhoods. Recent work has extended this idea using Graph Attention Networks (GANs) (Nathani et al, 2018), which assign attention weights to entries in a graph neighborhood, these being later combined.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Scalable knowledge base completion with superposition memories

Lalisse¹,

Rosen²,

Smolensky³

2021

Preprint

View full text Add to dashboard Cite

We present Harmonic Memory Networks (HMem), a neural architecture for knowledge base completion that models entities as weighted sums of pairwise bindings between an entity's neighbors and corresponding relations. Since entities are modeled as aggregated neighborhoods, representations of unseen entities can be generated on the fly. We demonstrate this with two new datasets: WNGen and FBGen. Experiments show that the model is SOTA on benchmarks, and flexible enough to evolve without retraining as the knowledge graph grows.1 Code and datasets are available at github.com/ MatthiasRLalisse/HMemNetworks.

show abstract

Section: Scaling Propertiescontrasting

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Scalable knowledge base completion with superposition memories

Lalisse¹,

Rosen²,

Smolensky³

2021

Preprint

View full text Add to dashboard Cite

show abstract

“…Several approaches focus on learning embeddings that leverage structured graph representations like Word-Net [18] to capture the semantics represented explicitly in such resources. Among such approaches, TransE [19], HolE [20], RDF2Vec [21], Graph Convolutions [22] and ComE+ [23] (see [24] for a recent review) directly encode the representations contained in the knowledge graph, which is typically an already condensed and filtered version of real-world data. A related research direction, with approaches like NASARI [25], SW2V [26] and Vecsigrafo [12], proposes to co-train word and concept embeddings based not only on knowledge graphs, but also on text corpora through lexical specificity or word-sense disambiguation.…”

Section: Related Workmentioning

confidence: 99%

On the impact of knowledge-based linguistic annotations in the quality of scientific embeddings

García-Silva¹,

Denaux²,

Gómez-Pérez³

2021

Future Generation Computer Systems

View full text Add to dashboard Cite

In essence, embedding algorithms work by optimizing the distance between a word and its usual context in order to generate an embedding space that encodes the distributional representation of words. In addition to single words or word pieces, other features which result from the linguistic analysis of text, including lexical, grammatical and semantic information, can be used to improve the quality of embedding spaces. However, until now we did not have a precise understanding of the impact that such individual annotations and their possible combinations may have in the quality of the embeddings. In this paper, we conduct a comprehensive study on the use of explicit linguistic annotations to generate embeddings from a scientific corpus and quantify their impact in the resulting representations. Our results show how the effect of such annotations in the embeddings varies depending on the evaluation task. In general, we observe that learning embeddings using linguistic annotations contributes to achieve better evaluation results.

show abstract

“…delft is not the first to use graph neural networks [29,47,48, inter alia] for question answering. entity-gcn [13], dfgn [43], and hde [52] build the entity graph with entity co-reference and cooccurrence in documents and apply gnn to the graph to rank the top entity as the answer.…”

Section: Graph Network For Qamentioning

confidence: 99%

Complex Factoid Question Answering with a Free-Text Knowledge Graph

Zhao

Xiong

Qian

et al. 2020

Proceedings of the Web Conference 2020

View full text Add to dashboard Cite

We introduce delft, a factoid question answering system which combines the nuance and depth of knowledge graph question answering approaches with the broader coverage of free-text. delft builds a free-text knowledge graph from Wikipedia, with entities as nodes and sentences in which entities co-occur as edges. For each question, delft finds the subgraph linking question entity nodes to candidates using text sentences as edges, creating a dense and high coverage semantic graph. A novel graph neural network reasons over the free-text graph-combining evidence on the nodes via information along edge sentences-to select a final answer. Experiments on three question answering datasets show delft can answer entity-rich questions better than machine reading based models, bert-based answer ranking and memory networks. delft's advantage comes from both the high coverage of its free-text knowledge graph-more than double that of dbpedia relations-and the novel graph neural network which reasons on the rich but noisy free-text evidence.

show abstract

Modeling Relational Data with Graph Convolutional Networks

Cited by 69 publications

References 0 publications

Scalable knowledge base completion with superposition memories

Scalable knowledge base completion with superposition memories

On the impact of knowledge-based linguistic annotations in the quality of scientific embeddings

Complex Factoid Question Answering with a Free-Text Knowledge Graph

Contact Info

Product

Resources

About