TL;DR
The paper introduces GECo, a novel explainability method for GNNs that leverages graph communities to improve interpretability in graph classification tasks, demonstrating superior performance over existing methods.
Contribution
GECo is a new community-based explainability approach for GNNs that highlights relevant graph structures, enhancing interpretability in graph classification.
Findings
GECo outperforms existing explainability methods on artificial datasets.
GECo achieves competitive results on real-world datasets.
The method effectively identifies key community structures influencing classification.
Abstract
Graph Neural Networks (GNNs) are powerful models that can manage complex data sources and their interconnection links. One of GNNs' main drawbacks is their lack of interpretability, which limits their application in sensitive fields. In this paper, we introduce a new methodology involving graph communities to address the interpretability of graph classification problems. The proposed method, called GECo, exploits the idea that if a community is a subset of graph nodes densely connected, this property should play a role in graph classification. This is reasonable, especially if we consider the message-passing mechanism, which is the basic mechanism of GNNs. GECo analyzes the contribution to the classification result of the communities in the graph, building a mask that highlights graph-relevant structures. GECo is tested for Graph Convolutional Networks on six artificial and fourā¦
Peer Reviews
DecisionĀ·Submitted to ICLR 2025
1. GECo enhances the interpretability of GNNs by identifying the most influential subgraphs, providing a more intuitive and targeted explanation for classification results. 2. GECo achieves significant explainability performance demonstrating effectiveness across both synthetic and real-world datasets.
1. This paper does not adequately highlight the advantages it has over other explainability models. Specifically, it should analyze the limitations of existing explainability models mentioned in the Related Work section (such as GNNExplainer, PGExplainer, SubgraphX, and PGMExplainer) and convincingly argue the advantages and necessity of a community-based approach for explainability, based on these limitations. 2. The approach in this study is straightforward and lacks novelty. In particular, u
1. The paper is well-organized and clearly written. 2. This paper introduces a novel community-based method for explaining GNNs, focusing on identifying key subgraphs rather than just individual nodes or edges.
1. GECo uses Blondel et al.'s modularity optimization algorithm for community detection, which performs well on large sparse matrices. However, it does not discuss how different community detection algorithms might impact the explanation results, leading to a lack of robustness verification. 2. GECo determines the threshold š by calculating the probability values of communities, using the mean or median as the threshold. However, this method may not be suitable for all cases, especially when th
1. The paper demonstrates that GECo outperforms existing methods in explainability across various datasets. It also highlights GECo's efficiency, offering faster computation times compared to other approaches.
1. Lack of novelty compared to the conventional graph community method. The proposed explainability method is based on detecting contributory substructures, specifically graph communities. However, the authors do not clearly distinguish the novelty of this approach from conventional graph community-based methods. Since substructure discovery is a widely adopted strategy for explaining GNNs, a more detailed demonstrationāeither theoretical or empiricalāof how the proposed method advances existing
Code & Models
Videos
No videos yet. Explain this paper in a talk, walkthrough, or lecture? Add one.
Taxonomy
MethodsGeneralized ELBO with Constrained Optimization
