Abstract
This paper introduces a robust framework for declarative process mining using Answer Set Programming (ASP). By leveraging modern ASP solvers, the authors provide a unified solution for log generation, conformance checking, and query checking in data-aware business processes. The proposed method significantly outperforms existing Alloy-based tools, offering up to a 60x speedup in log generation and providing the first viable solution for data-aware query checking.
Introduction
Process Mining (PM) serves as the bridge between data science and business process management, aiming to discover and improve real-world processes by extracting knowledge from event logs. In declarative PM, processes are described as a set of logical constraints (e.g., "Activity A must eventually be followed by Activity B") rather than rigid flowcharts.
A major challenge in this field is the "multi-perspective" or "data-aware" setting, where events carry payloads such as price or priority. Most existing tools struggle to scale when these data constraints are added to temporal logic. This paper addresses three fundamental problems in this context: log generation, conformance checking, and query checking.
Methodology
The authors utilize $L-LTL_f$ (Local Linear-time Temporal Logic over finite traces), a highly expressive logic that subsumes industry standards like DECLARE. The technical contribution is a three-step pipeline:
- Automata Construction: $L-LTL_f$ formulas are converted into Finite State Automata (FSA), which represent the rules the process must follow.
- ASP Encoding: These automata and associated data constraints are translated into an ASP program. Predicates like
state(I, S, T)track the automaton's state over time. - Solving with Clingo: The state-of-the-art ASP solver Clingo is used to find "Answer Sets," which represent generated logs, query assignments, or compliance confirmations.
This approach benefits from specification reuse: the same core rules for automaton transitions are used for all three PM tasks, with only input and output predicates changing.
Evaluation and Results
The framework was evaluated using synthetic models and real-life logs from the 4TU.ResearchData repository, including BPI2012 and Sepsis.
Performance Gains
The ASP-based tool demonstrated remarkable efficiency. On synthetic models, it was approximately 60 times faster than the previous state-of-the-art Alloy-based generator. For instance, it could generate 10,000 traces of length 30 in under 2 seconds.
Real-World Application
For complex real-life models like the PL log, the ASP tool generated 10,000 traces in about 9 seconds, compared to nearly 4 minutes for the Alloy-based tool. While specialized tools like Declare Analyzer are slightly faster for simple control-flow checks, the ASP approach is more general and handles data-aware constraints that other tools cannot.
Query Checking
The authors successfully demonstrated that query checking—previously considered too complex for data-aware logs—is feasible. Execution times scale linearly with trace length, making it practical for real-world use.
Conclusion
By modeling process mining tasks as automaton executions within ASP, the researchers have created a versatile "Swiss Army Knife" for the field. This unified framework eliminates the need for task-specific algorithms. Future work will focus on incorporating non-integer timestamps and extending the logic to allow correlation conditions that compare data across different process states.
Summary updated