Inconsistency between paper description and code implementation regarding fuzzy rules

[zhiqing0205]

Dear authors,

Thank you for sharing this interesting work on combining fuzzy rules with GNNs for image classification. I've been studying the codebase and found some inconsistencies between the paper
description and the actual implementation that I'd like to clarify.

Paper vs. Code Discrepancies

1. Number of rules and features

• Paper mentions: 3 rules with 3 features
• Code implementation: 10 rules with 6 topological features (utils/graph_utils.py:296-297, fuzzy_models/fuzzy_models.py:85)

2. Rule generation mechanism

• Paper describes: Hand-crafted threshold rules (e.g., "IF degree ≥ 7.28 THEN high_connectivity")
• Code implementation: K-means clustering to automatically generate rule centers (utils/graph_utils.py:155)

3. THEN clause mapping

• Paper mentions: Semantic concepts like "high_connectivity", "high_cliquishness", "high_label_consistency"
• Code implementation: No corresponding semantic mapping found; rules directly produce activation values that are concatenated with GNN embeddings

4. Rule-to-classification pathway
   The paper's examples suggest a clear interpretable path from topological patterns to image categories, but I couldn't find the corresponding implementation that maps concepts like
   "high_connectivity" to specific image classes (e.g., organs, blood cells).

Questions

1. Are the rule examples in the paper (Rules 1-3) from a different experiment or configuration?
2. Could you clarify how the topological fuzzy rules semantically relate to medical image classification?
3. Is there additional code or configuration that implements the specific rules mentioned in the paper?

I really appreciate the novel idea of integrating fuzzy rules with GNNs, and I'm eager to better understand the complete framework. Any clarification would be greatly helpful for reproducing
and building upon your work.

Thank you for your time and consideration.

Best regards