Servicenavigation

Detailseite

Zurück

Projekt Druckansicht

Extraktion und Verarbeitung von prozeduralem Erfahrungswissen in Workflows - Qualität, Interaktivität, und Wissenstransfer

Antragstellerinnen / Antragsteller Professor Dr. Ralph Bergmann; Professorin Dr. Mirjam Minor
Fachliche Zuordnung Bild- und Sprachverarbeitung, Computergraphik und Visualisierung, Human Computer Interaction, Ubiquitous und Wearable Computing
Förderung Förderung von 2011 bis 2022
Projektkennung Deutsche Forschungsgemeinschaft (DFG) - Projektnummer 200609093
 
Erstellungsjahr 2022

Zusammenfassung der Projektergebnisse

In the EVER project we investigated new methods in Process-Oriented Case-Based Reasoning (POCBR) and related fields for extracting, representing, and processing procedural experiential knowledge. The knowledge can be represented in a semi-structured manner such as in maintenance instructions or cooking recipes or it can be represented in a more structured form such as in business process models or in executable scientific workflows. During the first funding period of this project, we focused on the extraction of workflows from textual sources, the similarity-based retrieval of workflows for a particular goal of a user, and the automatic adaptation of workflows based on automatically learned adaptation knowledge as the core of automated reasoning tasks. Within the second funding period we extended our work regarding interactivity and we explored more complex workflow domains. In preparation for this, we have further tested the adaptation. We investigated whether the methods are suitable for adequately fulfilling multi-objective queries such as (i) controlling the complexity of generated workflows with desired structure and (ii) considering more complex queries with additional constraints on overall workflow properties. Concerning the interactivity in POCBR, we developed an interactive retrieval approach. The approach makes use of the automatically acquired adaptation knowledge. As a new application domain, we investigated scientific workflows for data mining with a particular focus put on the potentials for digital humanities research. We further developed the adaptation methods for applying them to scientific workflows and addressed the issue of efficient retrieval of workflows that we particularly faced with increased complexity of workflows. Based on our adaptation methods, we developed novel, ontology-based transfer learning methods for procedural knowledge that aim to adapt workflows across the boundaries of application domains. In particular, we contributed with automatic transformation of BPMN workflows into an ontology and utilization of the transfer strategy with the methods of analogical substitution, abstraction, and generalization. We used a knowledge discovery approach based on word embeddings to enrich the ontology by analogical knowledge between workflow tasks. We also operationalised the measurement of transfer distance. Surprisingly, the transfer distance seems to be more difficult to determine for POCBR than for traditional transfer learning approaches based on machine learning. It depends on the content of the workflow repository (data), the ontology (model of the transfer knowledge) and the transfer strategy (transfer process). The costs are calculated post mortem using the edit traces of adaptation operators. We elaborated different transfer strategies to guide the transfer process and implemented a transfer framework for POCBR. Experiments in the application domains passenger and baggage handling at the airport and SAP warehouse management confirmed that transfer learning of workflows - like ”normal” adaptation within one application domain - requires interactivity. Last but not least, we combined the retrieval and adaptation methods in a conversational POCBR approach for large repositories of complex workflows. There are various application domains for applying the methods developed in our work. For instance, we investigate use cases to enhance workflow flexibility in cyber-physical production systems in Industry 4.0. Another promising application domain are argumentation machines to support the synthesis and deliberation of arguments represented as graphs. The project results are of high interest for the data science platforms, also for commercial vendors of such platforms. Together with software partners and application partners, we are currently investigating the transfer of the project results into such a platform.

Projektbezogene Publikationen (Auswahl)

 
 

Zusatzinformationen

© 2024 DFG Kontakt / Impressum / Datenschutz

Textvergrößerung und Kontrastanpassung