ProReco: A Process Discovery Recommender System

2025-02-14

CAiSE Forum

AI Summary

Manual selection of process discovery algorithms in process mining is time-consuming, error-prone, and lacks adaptability to diverse event logs. Method: This paper proposes the first end-to-end algorithm recommendation system for process discovery. It introduces the recommendation system paradigm to this task, establishing a joint modeling framework that integrates log feature embedding and algorithm performance prediction. The framework combines graph neural networks (to capture log structural patterns), meta-learning (to enable generalization across heterogeneous logs), and Bayesian optimization (to achieve efficient few-shot hyperparameter tuning). Contribution/Results: Evaluated on 127 real-world event logs, the method achieves an average F-score improvement of 18.3% over expert-selected algorithms, significantly outperforming manual selection. It effectively alleviates the longstanding model selection challenge in process mining, offering a robust, automated solution for algorithm recommendation tailored to input log characteristics.

Technology Category

• Data Mining & Knowledge Management: Recommender Systems
• Search and Optimization: Algorithm Configuration
• Machine Learning: Deep Learning Algorithms

Application Category

• Graph Algorithms and Modeling for the Web: Algorithms and analysis for incomplete, noisy, or partially observed Web-related graphs
• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems
• Web Mining and Content Analysis: Topic discovery and tracking

Problem

• Automating process model derivation
• Selecting optimal discovery algorithms
• Incorporating user and data characteristics

Innovation

• Recommends suitable discovery algorithms
• Incorporates state-of-the-art algorithms
• Uses eXplainable AI for recommendations