README.md

AVOID

KDD_AVOID is a simulation and detection framework for rumor propagation based on social networks and personalized agents.

Project Structure

KDD_AVOID/
├── Env_Rumor_politic_persona1/           # Stores agent environments including long- and short-term memory
├── gcn_data_politic_img_persona/         # Stores propagation graphs generated for each news article
├── persona/                              # Persona extraction and modeling code
├── dataset_processing.py                 # Dataset loading and preprocessing utilities
├── earlystopping.py                      # Early stopping mechanism for model training
├── gcn.py                                # GCN model 
├── node_feature.py                       # Node-level feature extraction
├── LLM_prompt.py                         # LLM prompt construction and message simulation
├── Retriever.py                          # LLM-based retrieval for agent decision-making
├── self-map.py                           # Self-mapping logic for propagation
├── pro_mul.py                            # Multi-agent propagation simulation
├── main.py                               # Entry point for misinformation detection

Module Descriptions

• Environment Construction:

  • Env_Rumor_politic_persona1/: Contains the full agent memory structure (short-term + long-term).
  • persona/: Extracts and assigns persona traits for realistic agent behavior.
  • gcn_data_politic_img_persona/: Contains propagation graphs (as adjacency matrices, features, etc.) per article.

• dataprocessing:

  • dataset_processing.py, node_feature.py: Used for data parsing, feature extraction, and graph preparation.
  • gcn.py: Implements the GCN for learning over propagation graphs.
  • earlystopping.py: Prevents overfitting during training by monitoring validation loss.

• Agent-based Simulation:

  • LLM_prompt.py, Retriever.py, self-map.py, pro_mul.py: Use LLMs and rules to simulate multi-agent rumor propagation, decisions, and interactions.

• Detection Logic:

  • main.py: Main detection pipeline that integrates all components and outputs results.

This project is developed for research purposes related to rumor propagation modeling and early misinformation detection in social environments.