Note: This repository contains the official code implementation for the paper "EHHN: An Event-driven Heterogeneous Hypergraph Network for Object-Centric Next Activity Prediction" .

This project proposes an event-driven Heterogeneous Hypergraph Learning approach specifically designed for Object-Centric Event Logs (OCEL) to tackle the Next-Activity Prediction task in process mining.

The core code and datasets are organized as follows:

• data/: Directory for storing raw datasets
• newTry/: Core implementation code
  • Data Analysis & Preprocessing
    • ana.py: Script for exploring/analyzing raw datasets
    • construct_PE.py: Helper functions for feature construction
    • preprocess.py: Helper functions for general preprocessing
    • pipeline_OTC.py: Dedicated pipeline for OTC dataset
    • pipline_2017.py: Dedicated pipeline for BPI 2017 dataset
    • pipline_inter.py: Dedicated pipeline for Inter dataset
    • pipline_p2p.py: Dedicated pipeline for P2P dataset
  • Model Architecture
    • OCELhg.py: Custom core structure (OCEL Heterogeneous Hypergraph)
    • encoder.py: Encoder module
    • model.py: Overall model architecture
  • Training & Evaluation
    • Trainer.py: Model training script
    • test.py: Model testing and evaluation script
  • Configuration & Utilities
    • config.py: Configuration file
    • utils.py: General utility functions

We recommend using Conda to create a virtual environment:

conda create -n ocel_hg python=3.8
conda activate ocel_hg

Install required packages:

pip install torch torchvision torchaudio --index-url [https://download.pytorch.org/whl/cu118](https://download.pytorch.org/whl/cu118)
pip install torch_geometric pandas numpy pm4py

Place your raw OCEL datasets into the data/ directory. Ensure the filenames match the paths specified in your preprocessing pipelines.

Before running any scripts, open newTry/config.py to configure your settings (Target Dataset, Hyperparameters, Paths, etc.).

We provide customized preprocessing pipelines for each dataset. Run the specific pipeline tailored to your dataset to clean data and construct the hypergraph.

For the OTC dataset:

python newTry/pipeline_OTC.py

For other datasets:

python newTry/pipline_2017.py
python newTry/pipline_inter.py
python newTry/pipline_p2p.py

(Tip: If you are introducing a new dataset, run newTry/ana.py first to analyze the raw data structure.)

Once the preprocessing is complete, you can start training the model:

python newTry/Trainer.py

After the model finishes training, run the test script to evaluate its performance:

python newTry/test.py

• OCELhg.py: Maps object-centric event logs into a heterogeneous hypergraph structure to capture complex many-to-many relationships.

• Customized Pipelines: Emphasizes refined feature engineering tailored to different business contexts.

• OCELhg.py: Maps object-centric event logs into a heterogeneous hypergraph structure to capture complex many-to-many relationships.

• Customized Pipelines: Emphasizes refined feature engineering tailored to different business contexts.