Node Classification in Temporal Graphs through Stochastic Sparsification and Temporal Structural Convolution

Publication Date: 9/18/2020

Event: ECML-PKDD 2020 – The European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases

Reference: 1-17. 2020

Authors: Cheng Zheng, NEC Laboratories America, Inc.; University of California, Los Angeles; Bo Zong, NEC Laboratories America, Inc.; Wei Cheng, NEC Laboratories America, Inc.; Dongjin Song, NEC Laboratories America, Inc.; Jingchao Ni, NEC Laboratories America, Inc.; Wenchao Yu, NEC Laboratories America, Inc.; Haifeng Chen, NEC Laboratories America, Inc.; Wei Wang, University of California, Los Angeles

Abstract: Node classification in temporal graphs aims to predict node labels based on historical observations. In real-world applications, temporal graphs are complex with both graph topology and node attributes evolving rapidly, which poses a high overfitting risk to existing graph learning approaches. In this paper, we propose a novel Temporal Structural Network (TSNet) model, which jointly learns temporal and structural features for node classification from the sparsified temporal graphs. We show that the proposed TSNet learns how to sparsify temporal graphs to favor the subsequent classification tasks and prevent overfitting from complex neighborhood structures. The effective local features are then extracted by simultaneous convolutions in temporal and spatial domains. Using the standard stochastic gradient descent and backpropagation techniques, TSNet iteratively optimizes sparsification and node representations for subsequent classification tasks. Experimental study on public benchmark datasets demonstrates the competitive performance of the proposed model in node classification. Besides, TSNet has the potential to help domain experts to interpret and visualize the learned models.

Publication Link: