NHP: Neural Hypergraph Link Prediction

Abstract

Link prediction in simple graphs is a fundamental problem in which new links between vertices are predicted based on the observed structure of the graph. However, in many real-world applications, there is need to model relationships among vertices which go beyond pairwise associations. For example, in a chemical reaction, relationship among the reactants and products is inherently higherorder. Additionally, there is need to represent the direction from reactants to products. Hypergraphs provide a natural way to represent such complex higher-order relationships. Graph Convolutional Networks (GCN) have recently emerged as a powerful deep learning-based approach for link prediction over simple graphs. However, their suitability for link prediction in hypergraphs is underexplored-we fill this gap in this paper and propose Neural Hyperlink Predictor (NHP). NHP adapts GCNs for link prediction in hypergraphs. We propose two variants of NHP-NHP-U and NHP-D-for link prediction over undirected and directed hypergraphs, respectively. To the best of our knowledge, NHP-D is the first ever method for link prediction over directed hypergraphs. An important feature of NHP is that it can also be used for hyperlinks in which dissimilar vertices interact (e.g. acids reacting with bases). Another attractive feature of NHP is that it can be used to predict unseen hyperlinks at test time (inductive hyperlink prediction). Through extensive experiments on multiple real-world datasets, we show NHP's effectiveness. CCS CONCEPTS • Computing methodologies → Neural networks; Unsupervised learning.

Unsupervised Attributed Multiplex Network Embedding

“…Most recently, Ma et al (2019) proposed a graph convolutional network (GCN) based method called mGCN, which is not only unsupervised, but also naturally incorporates the node attributes by using GCNs. However, since it is based on GCNs that capture the local graph structure (Yadav et al 2019), it fails to fully model the global properties of a graph (Zhuang and Ma 2018;Wang, Cui, and Zhu 2016;Veličković et al 2019).…”

“…However, as node labeling is often expensive and time-consuming, it would be the best if a method can show competitive performance even without any label. Third, most of these methods fail to model the global properties of a graph, because they are based on random walk-based skip-gram model or graph convolutional network (GCN) (Kipf and Welling 2016), both of which are known to be effective for capturing the local graph structure (Yadav et al 2019). More precisely, nodes that are "close" (i.e., within the same context window or neighborhoods) in the graph are trained to have similar representations, whereas nodes that are far apart do not have similar representations, even though they are structurally similar (Ribeiro, Saverese, and Figueiredo 2017).…”

Neural Predicting Higher-order Patterns in Temporal Networks

“…Recently, higher-order patterns in networks have attracted much research interest because of their tremendous success in many real-world applications (Milo et al, 2002;Alon, 2007) including discovering data insights (Benson et al, 2016;Paranjape et al, 2017;Benson et al, 2018;Lambiotte et al, 2019;Do et al, 2020) and building scalable computing algorithms (Yin et al, 2017;Paranjape et al, 2017;Fu et al, 2020;Veldt et al, 2020). Previous works on higher-order structure prediction can be generally grouped into two categories, predicting multiple edges/subgraphs in graphs (Lahiri and Berger-Wolf, 2007;Meng et al, 2018;Nassar et al, 2020;Cotta et al, 2020) and predicting hyperedges in hypergraphs Benson et al, 2018;Yadati et al, 2020;Alsentzer et al, 2020). Subgraphs, e.g., cliques of nodes (Benson et al, 2016), could be used to describe higher-order patterns.…”

“…Neural networks (NN) have more potential to encode complex structural information and recently achieved great success in various graph applications (Hamilton et al, 2017;Meng et al, 2018;Jiang et al, 2019;Sankar et al, 2020;Trivedi et al, 2019;Xu et al, 2020;Rossi et al, 2020a). However, previous works focused on either hyperedge prediction in static networks (Rossi et al, 2020b;Yadati et al, 2020;Cotta et al, 2020;Alsentzer et al, 2020) or simple edge prediction in temporal networks but not for temporal higher-order patterns prediction. Moreover, none of the previous works (neither heuristic nor NN-based ones) are able to predict the entire spectrum of higher-order patterns (e.g., Fig.…”

Principled Hyperedge Prediction with Structural Spectral Features and Neural Networks

“…However, these methods showed unsatisfactory performance due to the limited expressive power of structural heuristics. Recently, graph neural network (GNN) has been introduced as a powerful method for hyperedge prediction, and showed much improved performance [3,12,25,31,39]. Essentially, all these methods could be considered as aggregation functions that integrate the information of individual nodes for Preprint.…”