2021
DOI: 10.48550/arxiv.2106.06935
|View full text |Cite
Preprint
|
Sign up to set email alerts
|

Neural Bellman-Ford Networks: A General Graph Neural Network Framework for Link Prediction

Abstract: Link prediction is a very fundamental task on graphs. Inspired by traditional path-based methods, in this paper we propose a general and flexible representation learning framework based on paths for link prediction. Specifically, we define the representation of a pair of nodes as the generalized sum of all path representations, with each path representation as the generalized product of the edge representations in the path. Motivated by the Bellman-Ford algorithm for solving the shortest path problem, we show … Show more

Help me understand this report

Search citation statements

Order By: Relevance

Paper Sections

Select...
1
1
1
1

Citation Types

0
8
0

Year Published

2022
2022
2022
2022

Publication Types

Select...
3

Relationship

1
2

Authors

Journals

citations
Cited by 3 publications
(8 citation statements)
references
References 66 publications
0
8
0
Order By: Relevance
“…Generalization. While we restricted our description in this work to the evaluation of link prediction on KGs, the discussed approaches are directly applicable to other settings which use rank-based evaluation, e.g., the entity pair ranking protocol [34], entity alignment [12,21,28,35], query embedding [3,4,14,20,23,24], unirelational link prediction [18,36], and relation detection [26].…”
Section: Discussionmentioning
confidence: 99%
“…Generalization. While we restricted our description in this work to the evaluation of link prediction on KGs, the discussed approaches are directly applicable to other settings which use rank-based evaluation, e.g., the entity pair ranking protocol [34], entity alignment [12,21,28,35], query embedding [3,4,14,20,23,24], unirelational link prediction [18,36], and relation detection [26].…”
Section: Discussionmentioning
confidence: 99%
“…(u,v) for nodes u and v. Here, we directly use the pair-wise geodesic for the two nodes, z (e) (u,v) = g (u,v) . As pointed out by previous works [59,54], the advantage of this setup is that, when the downstream task is to rank one link against N other links with one same node v, we only need to compute the distance vector from v once and find the geodesic nodes efficiently.…”
Section: Task Specific Geodesic Representationmentioning
confidence: 99%
“…Zhang et al [55] systematically show that the labeling trick bridges the gap between node-level and edge-level GNN expressiveness. Sharing the spirits of labeling trick, NBFNet [59] extends the generalized Bellman-Ford algorithm to the neural setting and can encode many traditional path-based methods. NBFNet has better scalability when there are multiple links sharing the same node.…”
Section: Related Workmentioning
confidence: 99%
See 1 more Smart Citation
“…Graph neural networks (GNNs) generalize traditional neural network architectures for data in the Euclidean domain to data in non-Euclidean domains [24,38,31]. As graphs are very general and flexible data structures and are ubiquitous in the real world, GNNs are now widely used in a variety of domains and applications such as social network analysis [18], recommender systems [41], graph reasoning [47], and drug discovery [35]. Indeed, many GNN architectures (e.g., GCN [24], GAT [38], MPNN [14]) have been proposed.…”
Section: Introductionmentioning
confidence: 99%