Object-Centric Process Constraints using Variable Bindings

Aaron Küsters

Abstract

With increasing interest and availability of Object-Centric Event Data (OCED), the focus of process mining research is shifting towards object-centric techniques. OCED removes the requirement of a single case notion, i.e., that all events only belong to exactly one case. Thus, OCED can capture real-life processes much more accurately. In this thesis, we introduce a declarative querying and constraint approach for OCED, focusing specifically on very high expressiveness while still allowing for efficient execution in practice. We first present and formalize a way to formulate nested queries of combinations of objects and events, so-called variable bindings. In contrast to prior work, our approach allows for querying combinations of multiple objects and events of any types. For example, also permitting queries for two orders placed by the same customer, one placed after the other. Constraints are presented as an extension to the querying approach, additionally specifying for each queried binding if it should be considered satisfied or violated. We introduce a visual notation for the queries and constraints of our approach and present a supporting tool implementation. Apart from the approach formalization, we also describe how the proposed types of declarative queries and constraints can be efficiently algorithmically evaluated on input OCED. Additionally, we outline how some types of constraints, which are very relevant for real-life processes, can be discovered automatically based on input OCED. Finally, we also evaluate the presented query and constraint approach by showcasing example constraints, demonstrating the high expressiveness and convenient visual representation of simple and complex constraints. Moreover, we explore the scalability and runtime performance of our approach implementation, showing excellent performance even for large real-life datasets with more than one million events.

Topics
process mining machine learning data science event data
Research Methods

Publication Data

Author: Aaron Küsters
Thesis Type: Master's Thesis
Pages: 0
Language: English
DOI:
About the Author:
Major / Study Program: Computer Science
Primary Field of Study:
Additional Study Interests:
License: CC BY-NC-SA 4.0
Date of Publication: 10/25/24
Status: Available
Date of Grading: 09/13/24
Institution: Chair of Process and Data Science (PADS) RWTH Aachen University (Chair of Process and Data Science (PADS) RWTH Aachen University, Germany)

Endorsements

# Name Details Endorsement
1
prof.dr.ir. wil van der Aalst
Supervisor
chair of Process and Data Science (PADS)
10/25/24
01:00:00 AM
Show comment

Thesis Documents and Supplemental Materials

11/07/24 03:30:02 AM
# Description Type Upload Date Location
1 Thesis Document PDF (47.82MB) 10/24/24 01:00:00 AMIPFS Download Raw
2 Research Data Repository 10/24/24 01:00:00 AM GitHub Open Repository