A Novel Graph Aggregation Method Based on Feature Distribution Around Each Ego-node for Heterophily

Shuichiro Haruta, Tatsuya Konishi, Mori Kurokawa
Proceedings of The 14th Asian Conference on Machine Learning, PMLR 189:452-466, 2023.

Abstract

In this paper, we propose a novel graph aggregation method based on feature distribution around each ego-node (a node to which features are aggregated) for heterophily. In heterophily graphs, labels of neighboring nodes can be uniformly distributed. In such case, aggregated features by existing GNNs will be always similar regardless of the label of ego-node and fail to capture useful information for a node classification task. Since the existing methods basically ignore label distribution around the ego-node, we attempt to handle heterophily graphs through dynamic aggregations so that nodes with similar vicinity characteristics exhibit similar behavior. In particular, we adjust the amount of aggregation based on the features generated by higher-order neighbors, since they reflect the label distribution around each ego-node. By doing this, we can take the influence of distant nodes into account while adapting local structures of each node. Extensive experiments demonstrate that the proposed method achieves higher performance in heterophily graphs by up to 14.68% compared with existing methods.

Cite this Paper

BibTeX
@InProceedings{pmlr-v189-haruta23a, title = {A Novel Graph Aggregation Method Based on Feature Distribution Around Each Ego-node for Heterophily}, author = {Haruta, Shuichiro and Konishi, Tatsuya and Kurokawa, Mori}, booktitle = {Proceedings of The 14th Asian Conference on Machine Learning}, pages = {452--466}, year = {2023}, editor = {Khan, Emtiyaz and Gonen, Mehmet}, volume = {189}, series = {Proceedings of Machine Learning Research}, month = {12--14 Dec}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v189/haruta23a/haruta23a.pdf}, url = {https://proceedings.mlr.press/v189/haruta23a.html}, abstract = {In this paper, we propose a novel graph aggregation method based on feature distribution around each ego-node (a node to which features are aggregated) for heterophily. In heterophily graphs, labels of neighboring nodes can be uniformly distributed. In such case, aggregated features by existing GNNs will be always similar regardless of the label of ego-node and fail to capture useful information for a node classification task. Since the existing methods basically ignore label distribution around the ego-node, we attempt to handle heterophily graphs through dynamic aggregations so that nodes with similar vicinity characteristics exhibit similar behavior. In particular, we adjust the amount of aggregation based on the features generated by higher-order neighbors, since they reflect the label distribution around each ego-node. By doing this, we can take the influence of distant nodes into account while adapting local structures of each node. Extensive experiments demonstrate that the proposed method achieves higher performance in heterophily graphs by up to 14.68% compared with existing methods.} }
Endnote
%0 Conference Paper %T A Novel Graph Aggregation Method Based on Feature Distribution Around Each Ego-node for Heterophily %A Shuichiro Haruta %A Tatsuya Konishi %A Mori Kurokawa %B Proceedings of The 14th Asian Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2023 %E Emtiyaz Khan %E Mehmet Gonen %F pmlr-v189-haruta23a %I PMLR %P 452--466 %U https://proceedings.mlr.press/v189/haruta23a.html %V 189 %X In this paper, we propose a novel graph aggregation method based on feature distribution around each ego-node (a node to which features are aggregated) for heterophily. In heterophily graphs, labels of neighboring nodes can be uniformly distributed. In such case, aggregated features by existing GNNs will be always similar regardless of the label of ego-node and fail to capture useful information for a node classification task. Since the existing methods basically ignore label distribution around the ego-node, we attempt to handle heterophily graphs through dynamic aggregations so that nodes with similar vicinity characteristics exhibit similar behavior. In particular, we adjust the amount of aggregation based on the features generated by higher-order neighbors, since they reflect the label distribution around each ego-node. By doing this, we can take the influence of distant nodes into account while adapting local structures of each node. Extensive experiments demonstrate that the proposed method achieves higher performance in heterophily graphs by up to 14.68% compared with existing methods.
APA
Haruta, S., Konishi, T. & Kurokawa, M.. (2023). A Novel Graph Aggregation Method Based on Feature Distribution Around Each Ego-node for Heterophily. Proceedings of The 14th Asian Conference on Machine Learning, in Proceedings of Machine Learning Research 189:452-466 Available from https://proceedings.mlr.press/v189/haruta23a.html.

Related Material