Abstract
The present disclosure proposes techniques (e.g., a method and a system) for generating temporal heterophilic graph neural network (THGNN) for event-based graphs. All existing (temporal) GNNs typically follow temporal homophily assumption, while the real-world graphs are largely heterophilic. To solve this problem, the proposed disclosure first defines the graph heterophily based on edge heterophily and then develops a novel method/technique to automatically detect the homophily/heterophily weight of graph nodes that should be regarded as the focal node’s homophilic/heterophilic neighbours. Finally, a neighbourhood sampling based GNN is adopted. In this manner, the proposed disclosure defines heterophily for temporal graphs and proposes a novel adaptive method to select right neighbourhood for graph aggregation.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
YAN, YUCHEN Visa; CHEN, YUZHONG Visa; CHEN, HUIYUAN Visa; LI, XIAOTING Visa; and DAS, MAHASHWETA Visa, "THGNN: Temporal Heterophilic Graph Neural Network for Event-based Graph", Technical Disclosure Commons, (February 22, 2024)
https://www.tdcommons.org/dpubs_series/6705