CascadeKG

This repository contains the code for the paper "Risk-Controlled Event-Driven Cascading Updates for Knowledge Graph Consistency Restoration".

Core Components

Models (models/)

• RGCNEncoder.py - Text-conditioned R-GCN variants (additive, gated, FiLM, concat+MLP)
• DistMultDecoder.py, ComplExDecoder.py - KG scoring decoders
• Baseline models: DistMult, TransE, CompGCN, R-GAT, HittER, SimKGC, KGT5

Training & Evaluation

• main_cascade.py - Main training script for CASCADEKG
• run_text_ablation.py - Text fusion ablation study (Table in rebuttal)
• run_kgc_baselines.py - KGC baseline adaptation experiments
• eval_kgc_baselines.py - KGC baseline evaluation
• benchmark_calibration_time.py - LTT calibration timing analysis

Core Logic

• trainers/calibration.py - LTT calibration implementation
• trainers/cascade_prediction.py - Cascade prediction pipeline
• trainers/train_loop.py - Training loop
• utils/data_utils.py - Data loading and preprocessing
• data/CascadeDataset.py - PyTorch Dataset implementation

Usage

Training CASCADEKG

python main_cascade.py --dataset icews14 --alpha1 0.6 --alpha2 0.3

Text Fusion Ablation

python run_text_ablation.py --variant [text_conditioned|text_gated|text_film|text_concat_mlp]

KGC Baseline Evaluation

python run_kgc_baselines.py --model [simkgc|kgt5]

Calibration Time Benchmark

python benchmark_calibration_time.py --granularity [coarse|medium|fine]

Requirements

torch>=1.10
transformers>=4.20
numpy
tqdm

Datasets

The code expects processed ICEWS14-Event and YAGO-Event datasets in ./data/. Data processing scripts are included in data/icews14/ and data/YAGO3-10/.