Graph Neighborhood Attentive Pooling

Abstract

Network representation learning (NRL) is a powerful technique for learning lowdimensional vector representation of high-dimensional and sparse graphs. Most studies explore the structure and metadata associated with the graph using random walks and employ an unsupervised or semi-supervised learning schemes. Learning in these methods is context-free, because only a single representation per node is learned. Recently studies have argued on the sufficiency of a single representation and proposed a context-sensitive approach that proved to be highly effective in applications such as link prediction and ranking. However, most of these methods rely on additional textual features that require RNNs or CNNs to capture high-level features or rely on a community detection algorithm to identify multiple contexts of a node. In this study, without requiring additional features nor a community detection algorithm, we propose a novel context-sensitive algorithm called GAP that learns to attend on different parts of a node's neighborhood using attentive pooling networks. We show the efficacy of GAP using three real-world datasets on link prediction and node clustering tasks and compare it against 10 popular and state-ofthe-art (SOTA) baselines. GAP consistently outperforms them and achieves up to ≈ 9% and ≈ 20% gain over the best performing methods on link prediction and clustering tasks, respectively.

Graph Embedding with Similarity Metric Learning

“…For example, Epasto et al [12] devised a method for acquiring multiple representations of nodes through self-network decomposition, which enables the encoding of node roles within distinct local communities. Kefato et al [13] employed an attentional pooling network to learn the personalized importance of neighboring nodes. It allows the model to concentrate exclusively on relevant neighbor information and ignore irrelevant neighbor information during interaction processes.…”

“…This form encompasses the modeling of text information and network topology. Expanding upon the research by Kefato et al [13], we propose a novel graph-embedding model that employs a symmetric similarity measurement function that enables encoding a more significant amount of topological structural information of nodes, thereby obtaining high-quality node embedding representations.…”

“…Epasto et al [12] encoded multiple representations of a node by leveraging its roles in distinct local communities through ego-network decomposition. Kefato et al [13] utilized an attentional pooling network to determine the personalized importance of a node's neighbors. Qin et al [14] proposed using GCN to assist in obtaining high-quality node representations.…”

“…We aim to generate high-quality context-sensitive node representations by focusing solely on the structural information of nodes. Our work is based on the Graph Attentional Pooling network (GAP) [13], which could view as a generalization of it. The model suggests the source node's representation using a mutual attention mechanism that responds to changes in the target node, thereby achieving context sensitivity.…”

CommPOOL: An Interpretable Graph Pooling Framework for Hierarchical Graph Representation Learning

“…Most HGP operations in the current HGPNNs (Zhang et al 2019;Ying et al 2018;Lee, Lee, and Kang 2019;Gao and Ji 2019;Kefato and Girdzijauskas 2020;Bianchi, Grattarola, and Alippi 2020) show little interpretability and are difficult to be understood by the users. Moreover, very few studies present the interpretabiliy of their HGP operations in the paper, which may be accounted for by the following two issues: (1).…”