A Framework for Comprehensive Evaluations of Graph Neural Network based Community Detection using Node Clustering

Will Leeney, Ryan McConville

Keywords: graph neural networks, clustering, community detection

Abstract: Graph Neural Networks (GNNs) have shown promising performance across a number of tasks in recent years. Unsupervised community detection using GNNs involves the clustering of nodes of a graph given both the features of nodes as well as the structure of the graph, and has many applications to real world tasks from social networks to genomics. Unfortunately, there has been relatively little research using GNNs for commOunity detection, and even less that evaluates the systems rigorously and fairly. A comprehensive evaluation of the performance of GNNs requires an suitable environment within which they are evaluated. This is exacerbated by the fact that community detection is primarily an unsupervised task, and that (graph) neural networks are used which contain many hyperparameters, discovered by inconsistent procedures. We argue that there is currently a gap in the literature that establishes a sufficient benchmarking environment for the consistent evaluation of GNN based community detection, thereby impeding progress in this nascent field. In this work we propose and evaluate an environment for the consistent evaluation of neural community detection. With this we show the strong dependence of the performance to the experimental settings , thereby motivating the use of this framework to facilitate research into GNN based community detection.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics

Submission Guidelines: Yes

Please Choose The Closest Area That Your Submission Falls Into: Unsupervised and Self-supervised learning