Seeliger, Alexander (2020)
Intelligent Computer-assisted Process Mining.
Technische Universität Darmstadt
doi: 10.25534/tuprints-00011915
Dissertation, Erstveröffentlichung
Kurzbeschreibung (Abstract)
Most enterprises and organizations have digitized their work by implementing process-aware information systems. These systems typically record a large amount of event data about the organization's day-to-day business. Organizations are interested in analyzing this data to optimize their business processes. Process mining is a research field that analyzes event data to obtain valuable knowledge about how processes are executed in reality. Different from conventional interview-based methods, process mining analyzes the process performance and compliance solely based on recorded event data.
Currently, the work of analysts using process mining tools is characterized as largely manual, leading to many ad-hoc tasks. With the growth of available event data and the increasing complexity of processes, several issues emerge in practice. Many process mining methods are not specifically designed for large event logs, leading to high computation time and a substantial amount of manual work. Consequently, suitable subsets of cases must be selected before applying certain analysis tasks successfully. Furthermore, applying process mining techniques to unknown event data often requires extensive domain knowledge and process mining expertise to obtain valuable insights. Although efforts were made to systematize the manual work of analysts in process mining projects, current tools lack intelligent computer-assisted guidance.
This dissertation introduces several contributions to different steps along an analyst's workflow to address the above issues. It is divided into three major parts: The first part introduces a process knowledge artifact framework that simplifies data extraction and processing of heterogeneous data sources as an enabling preparatory step towards process mining. In the second part, the dissertation introduces three algorithms specifically designed for large and complex real-life event logs. A novel compliance checking algorithm significantly improves runtime performance for large event logs to identify compliance rule violations. A parameter-free process drift detection algorithm automatically detects changes in the process execution over time to reveal potential process issues. A novel multi-perspective trace clustering algorithm automatically detects patterns in the control flow and the data perspective of a process to reveal the different process behaviors in event logs, aiming at simplifying process discovery. The third part of the dissertation integrates the contributions made in the first two parts and introduces an interactive visual recommendation approach to enhance process analysis guidance. Lastly, a process improvement approach is presented, suggesting modifications to process models for a given improvement goal.
The proposed algorithms and methods were evaluated with synthetic and real-life event logs, demonstrating their superior qualitative performance and practical feasibility. A user study conducted with process mining experts evaluated the prototype implementation of the interactive visual recommendation system to assess the usefulness in the process analysis workflow.
Typ des Eintrags: | Dissertation | ||||
---|---|---|---|---|---|
Erschienen: | 2020 | ||||
Autor(en): | Seeliger, Alexander | ||||
Art des Eintrags: | Erstveröffentlichung | ||||
Titel: | Intelligent Computer-assisted Process Mining | ||||
Sprache: | Englisch | ||||
Referenten: | Mühlhäuser, Prof. Dr. Max ; Rosemann, Prof. Dr. Michael | ||||
Publikationsjahr: | 18 August 2020 | ||||
Ort: | Darmstadt | ||||
Datum der mündlichen Prüfung: | 26 Juni 2020 | ||||
DOI: | 10.25534/tuprints-00011915 | ||||
URL / URN: | https://tuprints.ulb.tu-darmstadt.de/11915 | ||||
Kurzbeschreibung (Abstract): | Most enterprises and organizations have digitized their work by implementing process-aware information systems. These systems typically record a large amount of event data about the organization's day-to-day business. Organizations are interested in analyzing this data to optimize their business processes. Process mining is a research field that analyzes event data to obtain valuable knowledge about how processes are executed in reality. Different from conventional interview-based methods, process mining analyzes the process performance and compliance solely based on recorded event data. Currently, the work of analysts using process mining tools is characterized as largely manual, leading to many ad-hoc tasks. With the growth of available event data and the increasing complexity of processes, several issues emerge in practice. Many process mining methods are not specifically designed for large event logs, leading to high computation time and a substantial amount of manual work. Consequently, suitable subsets of cases must be selected before applying certain analysis tasks successfully. Furthermore, applying process mining techniques to unknown event data often requires extensive domain knowledge and process mining expertise to obtain valuable insights. Although efforts were made to systematize the manual work of analysts in process mining projects, current tools lack intelligent computer-assisted guidance. This dissertation introduces several contributions to different steps along an analyst's workflow to address the above issues. It is divided into three major parts: The first part introduces a process knowledge artifact framework that simplifies data extraction and processing of heterogeneous data sources as an enabling preparatory step towards process mining. In the second part, the dissertation introduces three algorithms specifically designed for large and complex real-life event logs. A novel compliance checking algorithm significantly improves runtime performance for large event logs to identify compliance rule violations. A parameter-free process drift detection algorithm automatically detects changes in the process execution over time to reveal potential process issues. A novel multi-perspective trace clustering algorithm automatically detects patterns in the control flow and the data perspective of a process to reveal the different process behaviors in event logs, aiming at simplifying process discovery. The third part of the dissertation integrates the contributions made in the first two parts and introduces an interactive visual recommendation approach to enhance process analysis guidance. Lastly, a process improvement approach is presented, suggesting modifications to process models for a given improvement goal. The proposed algorithms and methods were evaluated with synthetic and real-life event logs, demonstrating their superior qualitative performance and practical feasibility. A user study conducted with process mining experts evaluated the prototype implementation of the interactive visual recommendation system to assess the usefulness in the process analysis workflow. |
||||
Alternatives oder übersetztes Abstract: |
|
||||
URN: | urn:nbn:de:tuda-tuprints-119154 | ||||
Sachgruppe der Dewey Dezimalklassifikatin (DDC): | 000 Allgemeines, Informatik, Informationswissenschaft > 004 Informatik | ||||
Fachbereich(e)/-gebiet(e): | 20 Fachbereich Informatik 20 Fachbereich Informatik > Telekooperation |
||||
Hinterlegungsdatum: | 18 Aug 2020 12:06 | ||||
Letzte Änderung: | 25 Aug 2020 08:27 | ||||
PPN: | |||||
Referenten: | Mühlhäuser, Prof. Dr. Max ; Rosemann, Prof. Dr. Michael | ||||
Datum der mündlichen Prüfung / Verteidigung / mdl. Prüfung: | 26 Juni 2020 | ||||
Export: | |||||
Suche nach Titel in: | TUfind oder in Google |
Frage zum Eintrag |
Optionen (nur für Redakteure)
Redaktionelle Details anzeigen |