flyingdoog/awesome-graph-explainability-papers

1. This repository is a static curated bibliography, not an application or library. Its “product” is a maintained index of graph neural network explainability papers, with hyperlinks out to papers, code, surveys, and documentation.

2. The primary entry point is README.md. It acts as the landing page and main catalog, starting with a title and short scope statement, then listing papers grouped by category and year.

3. The content model is taxonomy-first. The README organizes the corpus into major sections such as Surveys, Platforms, and then year-based collections from 2026 back to “2020 and before.” This is the main navigation structure a reader would use.

4. The most important architectural boundary is between the repository and external sources. Almost every item links out to external paper hosts such as arXiv, OpenReview, ACM/IEEE/Springer pages, and occasionally project code repositories or platform documentation.

5. The “Platforms” section is a second subsystem distinct from the paper index. It highlights tooling and benchmark ecosystems around GNN explainability, including PyTorch Geometric explainability docs and external libraries/benchmarks like DIG, GraphXAI, GraphFramEx, GNNExplainer/PGExplainer code, and BAGEL.

6. The year sections function as a chronological changelog of the research field. They do not contain executable logic; instead they encode curation data and provenance for papers, enabling the repo to serve as a historical map of the area.

7. There is also a topic-oriented view implied by bytopics.md. Even though only the filename is visible here, it likely serves as an alternate index that reorganizes the same bibliography by method family, task type, or explanation style rather than by publication year.

8. Because the repository is documentation-only, there are no runtime services, build pipelines, databases, or APIs. The only “data flow” is manual curation: papers are added or reclassified in Markdown, and readers follow the embedded links to external sources.

9. The repository’s content boundary is broad but focused: it covers explainability for graph neural networks and closely related graph learning topics, but excludes implementation details except where a paper or platform is explicitly referenced.

10. For diagramming purposes, the repo should be modeled as a two-view knowledge base: a chronological bibliography (README.md) plus a topic index (bytopics.md), both backed by external scholarly sources and platform docs.