Scale-free Graphs

Dataset Overview

This dataset focuses on deep graph translation problems, providing various synthetic and real‑world graph datasets for studying and exploring mapping rules between graphs. The dataset includes the following sections:

1. Scale‑free Graphs

• Task Type: Suitable for one‑to‑many graph translation.
• Content: Five subsets of different sizes (10, 20, 50, 100, 150 nodes).
• File Format: Input and output graphs are stored in .csv files named scale-(graph_size)-input-index.csv and scale-(graph_size)-target-index.csv respectively.
• Data Links: Scale_free_150 etc.

2. Erdos‑Renyi Graphs

• Task Type: Suitable for one‑to‑one graph translation.
• Content: Three subsets of sizes 20, 40, 60 nodes.
• File Format: Files named ER-(graph_size)-input-index.csv and ER-(graph_size)-target-index.csv.
• Data Links: ER_20 etc.

3. Barabási‑Albert Graphs

• Task Type: Suitable for one‑to‑one graph translation.
• Content: Three subsets of sizes 20, 40, 60 nodes.
• File Format: Files named BA-(graph_size)-input-index.csv and BA-(graph_size)-target-index.csv.
• Data Links: BA_20 etc.

4. IoT

• Context: Malware isolation in the Internet of Things.
• Content: Three subsets of sizes 20, 40, 60 nodes.
• File Format: Files named IoT-[graph_size]-[input/output]-[infection rate]-[recovery rate]-[decay rate]-[index].csv.
• Data Links: IoT_20 etc.

5. User Authentication

• Context: Predicting malicious behavior in enterprise network user authentication.
• Content: Two subsets of sizes 50 and 300 nodes.
• File Format: Files named Auth-[graph_size]-[input/output]-[index].csv.
• Data Links: Auth_50 etc.

6. Chemistry Reaction

• Context: Chemical reaction prediction.
• Content: 7,180 pairs of reactant and product molecular graphs.
• File Format: Data stored in multiple folders, e.g., mol_edge and mol_nodes.
• Data Links: Mol_edge etc.

7. Molecule Optimization

• Context: Molecular optimization via matched molecular pair analysis (MMPA) to improve chemical properties.
• Content: Multiple tasks such as improving penalized logP and drug‑likeness (QED).
• Data Links: Penalized logP etc.

These datasets provide rich resources for research on deep graph translation, supporting various mapping rules and application scenarios.