Inductive Link Prediction with Interactive Structure Learning on Attributed Graph
Lists
Only you see this
Abstract
Link prediction is one of the most important tasks in graph machine learning, which aims at predicting whether two nodes in a network have an edge. Real-world graphs typically contain abundant node and edge attributes, thus how to perform link prediction by simultaneously learning structure and attribute information from both interactions/paths between two associated nodes and local neighborhood among node’s ego subgraph is intractable. To address this issue, we develop a novel Path-aware Graph Neural Network (PaGNN) method for link prediction, which incorporates interaction and neighborhood information into graph neural networks via broadcasting and aggregating operations. And a cache strategy is developed to accelerate the inference process. Extensive experiments show a superior performance of our proposal over state-of-the-art methods on real-world link prediction tasks.
BibTeX
@inproceedings{yang2021inductive,
author = {Shuo Yang and Binbin Hu and Zhiqiang Zhang and Wang Sun and Yang Wang and Jun Zhou and Hongyu Shan and Yuetian Cao and Borui Ye and Yanming Fang and Quan Yu},
title = {{Inductive Link Prediction with Interactive Structure Learning on Attributed Graph}},
booktitle = {ECML/PKDD},
year = {2021},
doi = {10.1007/978-3-030-86520-7\_24},
}Loading related content...