@dataset{sehmer_2026_19627517,
  author       = {Sehmer, Felix and
                  Hoffmann, Maximilian and
                  Kloos, Norberth and
                  Bergmann, Ralph},
  title        = {SOWEKI Water Demand Dataset},
  month        = apr,
  year         = 2026,
  publisher    = {Zenodo},
  doi          = {10.5281/zenodo.19627517},
  url          = {https://doi.org/10.5281/zenodo.19627517},
}

@inproceedings{Lenz2025LLsiMLargeLanguage,
  title = {{{LLsiM}}: {{Large Language Models}} for~{{Similarity Assessment}} in~{{Case-Based Reasoning}}},
  shorttitle = {{{LLsiM}}},
  booktitle = {Case-{{Based Reasoning Research}} and {{Development}}},
  author = {Lenz, Mirko and Hoffmann, Maximilian and Bergmann, Ralph},
  editor = {Bichindaritz, Isabelle and L{\'o}pez, Beatriz},
  year = {2025},
  series = {Lecture {{Notes}} in {{Computer Science}}},
  volume = {15662},
  pages = {126--141},
  publisher = {Springer Nature Switzerland},
  address = {Cham},
  doi = {10.1007/978-3-031-96559-3_9},
  abstract = {In Case-Based Reasoning (CBR), past experience is used to solve new problems. Determining the most relevant cases is a crucial aspect of this process and is typically based on one or multiple manually-defined similarity measures, requiring deep domain knowledge. To overcome the knowledge-acquisition bottleneck, we propose the use of Large Language Models (LLMs) to automatically assess similarities between cases. We present three distinct approaches where the model is used for different tasks: (i) to predict similarity scores, (ii) to assess pairwise preferences, and (iii) to automatically configure similarity measures. Our conceptual work is accompanied by an open-source Python implementation that we use to evaluate the approaches on three different domains by comparing them to manually crafted similarity measures. Our results show that directly using LLM-based scores does not align well with the baseline rankings, but letting the LLM automatically configure the measures yields rankings that closely resemble the expert-defined ones.},
  isbn = {978-3-031-96559-3},
  langid = {english},
  url = {https://www.wi2.uni-trier.de/shared/publications/Lenz2025LLsiMLargeLanguage.pdf}
}

In Case-Based Reasoning (CBR), past experience is used to solve new problems. Determining the most relevant cases is a crucial aspect of this process and is typically based on one or multiple manually-defined similarity measures, requiring deep domain knowledge. To overcome the knowledge-acquisition bottleneck, we propose the use of Large Language Models (LLMs) to automatically assess similarities between cases. We present three distinct approaches where the model is used for different tasks: (i) to predict similarity scores, (ii) to assess pairwise preferences, and (iii) to automatically configure similarity measures. Our conceptual work is accompanied by an open-source Python implementation that we use to evaluate the approaches on three different domains by comparing them to manually crafted similarity measures. Our results show that directly using LLM-based scores does not align well with the baseline rankings, but letting the LLM automatically configure the measures yields rankings that closely resemble the expert-defined ones.

@inproceedings{WeichSHB2025,
    author       = {Justin Weich and Alexander Schultheis and Maximilian Hoffmann and Ralph Bergmann},
    title        = {{Integration of Time Series Embedding for Efficient Retrieval in Case-Based Reasoning}},
    booktitle    = {Case-Based Reasoning Research and Development - 33rd International Conference, {ICCBR} 2025, Biarritz, France, June 30 - July 3rd, 2025, Proceedings},
	  series       = {Lecture Notes in Computer Science},
    pages        = {328--344},
    publisher    = {Springer.},
    year         = {2025},
  	doi 		     = {10.1007/978-3-031-96559-3_22},
  	keywords 	   = {Temporal Case-Based Reasoning, Time Series Data, Time Series Embedding, Time Series Similarity Measure, Siamese Neural Networks},
  	abstract 	   = {The increasing volume of time series data in Industry 4.0 applications creates substantial challenges for real-time data analysis. Such analyses that are conducted in the research area of Temporal Case-Based Reasoning (TCBR) face performance problems due to complex similarity measures. One potential approach already proven in other domains for addressing these problems is the usage of embedding techniques for time series data, which map these data into a simplified vector representation. Therefore, this paper investigates the integration of time series embedding techniques in the context of Case-Based Reasoning (CBR) to improve retrieval efficiency. Therefore, requirements for the application of embedding techniques in CBR are derived. A systematic literature study identifies possible approaches that are analyzed based on the requirements, with the result that no approach is suitable for the application. Therefore, a novel embedding architecture is proposed, using a Siamese neural network approach that can be trained with similarity values. The architecture is prototypically implemented in the ProCAKE framework and evaluated in an Internet of Things use case from a smart factory. The results demonstrate that the embedding-based retrieval achieves classification performance comparable to traditional similarity measures while significantly reducing retrieval time.},
    url         = {https://www.wi2.uni-trier.de/shared/publications/2025_ICCBR_WeichEtAL.pdf}
}

The increasing volume of time series data in Industry 4.0 applications creates substantial challenges for real-time data analysis. Such analyses that are conducted in the research area of Temporal Case-Based Reasoning (TCBR) face performance problems due to complex similarity measures. One potential approach already proven in other domains for addressing these problems is the usage of embedding techniques for time series data, which map these data into a simplified vector representation. Therefore, this paper investigates the integration of time series embedding techniques in the context of Case-Based Reasoning (CBR) to improve retrieval efficiency. Therefore, requirements for the application of embedding techniques in CBR are derived. A systematic literature study identifies possible approaches that are analyzed based on the requirements, with the result that no approach is suitable for the application. Therefore, a novel embedding architecture is proposed, using a Siamese neural network approach that can be trained with similarity values. The architecture is prototypically implemented in the ProCAKE framework and evaluated in an Internet of Things use case from a smart factory. The results demonstrate that the embedding-based retrieval achieves classification performance comparable to traditional similarity measures while significantly reducing retrieval time.

@article{Hoffmann_Process_Augmentation_2025,
author = {Maximilian Hoffmann and Lukas Malburg and Ralph Bergmann},
title = {{Augmentation of Semantic Processes for Deep Learning Applications}},
journal = {Applied Artificial Intelligence},
volume = {39},
number = {1},
pages = {2506788},
year = {2025},
publisher = {Taylor \& Francis},
doi = {10.1080/08839514.2025.2506788},
URL = {https://www.wi2.uni-trier.de/shared/publications/2025_AAI_HoffmannEtAl.pdf}
}

@phdthesis{Hoffmann25,
	title        = {{Hybrid AI for Process Management - Improving Similarity Assessment in Process-Oriented Case-Based Reasoning via Deep Learning}},
	author       = {Hoffmann, Maximilian},
	year         = 2025,
	url          = {https://ubt.opus.hbz-nrw.de/files/2518/Dissertation_Hoffmann.pdf},
	school       = {University of Trier, Germany},
	note = {Licensed under Creative Commons 4.0 BY-NC.}
}

@InProceedings{Schuler.2023_SemiSupervisedTransferLearning,
author="Schuler, Nicolas and Hoffmann, Maximilian and Beise, Hans-Peter and Bergmann, Ralph",
editor="Bramer, Max and Stahl, Frederic",
title="Semi-supervised Similarity Learning in Process-Oriented Case-Based Reasoning",
booktitle="Artificial Intelligence XL",
year="2023",
publisher="Springer Nature Switzerland",
address="Cham",
pages="159--173",
abstract="Supervised learning is typically challenging with insufficient amounts of labeled training data and high costs for label acquisition, creating a demand for unsupervised learning methods. In the research area of Process-Oriented Case-Based Reasoning (POCBR), this demand is created by training data that is manually-modeled and computationally-expensive labeling methods. In this paper, we propose a semi-supervised transfer learning method for learning similarities between pairs of semantic graphs in POCBR with Graph Neural Networks (GNNs). The method aims to replace the fully supervised learning procedure from previous work with an unsupervised and a supervised training phase. In the first phase, the GNNs are pretrained with a triplet learning procedure that utilizes graph augmentation and random selection to enable unsupervised training. This phase is followed by a supervised one where the pretrained model is trained on the original labeled training data. The experimental evaluation examines the quality of the semi-supervised models compared to the supervised models from previous work for three semantic graph domains with different properties. The results indicate the potential of the proposed approach for improving retrieval quality.",
isbn="978-3-031-47994-6",
url="https://www.wi2.uni-trier.de/shared/publications/2023_SGAI_Schuler.pdf"
}

Supervised learning is typically challenging with insufficient amounts of labeled training data and high costs for label acquisition, creating a demand for unsupervised learning methods. In the research area of Process-Oriented Case-Based Reasoning (POCBR), this demand is created by training data that is manually-modeled and computationally-expensive labeling methods. In this paper, we propose a semi-supervised transfer learning method for learning similarities between pairs of semantic graphs in POCBR with Graph Neural Networks (GNNs). The method aims to replace the fully supervised learning procedure from previous work with an unsupervised and a supervised training phase. In the first phase, the GNNs are pretrained with a triplet learning procedure that utilizes graph augmentation and random selection to enable unsupervised training. This phase is followed by a supervised one where the pretrained model is trained on the original labeled training data. The experimental evaluation examines the quality of the semi-supervised models compared to the supervised models from previous work for three semantic graph domains with different properties. The results indicate the potential of the proposed approach for improving retrieval quality.

@inproceedings{SchultheisHMB2023,
  author       = {Schultheis, Alexander and Hoffmann, Maximilian and Malburg, Lukas and Bergmann, Ralph},
  title        = {{Explanation of Similarities in Process-Oriented Case-Based Reasoning by Visualization}},
  booktitle    = {Case-Based Reasoning Research and Development - 31st International Conference, {ICCBR} 2023, Aberdeen, Scotland, July 17-20, 2023, Proceedings},
  series       = {Lecture Notes in Computer Science},
  volume       = {14141},
  pages        = {53--68},
  publisher    = {Springer},
  year         = {2023},
  doi		   = {10.1007/978-3-031-40177-0_4},
  url		   = {https://www.wi2.uni-trier.de/shared/publications/2023_ICCBR_SchultheisHMB.pdf},
  abstract 	   = {Modeling similarity measures in Case-Based Reasoning is a knowledge-intensive, demanding, and error-prone task even for domain experts. Visualizations offer support for users, but are currently only available for certain subdomains and case representations. Currently, there are only visualizations that can be used for local attributes or specific case representations. However, there is no possibility to visualize similarities between complete processes accordingly so far, although complex domains may be present. Therefore, an extension of existing approaches or the design of new suitable concepts for this application domain is necessary. The contribution of this work is to enable a more profound understanding of similarity for knowledge engineers who create a similarity model and support them in this task by using visualization methods in Process-Oriented Case-Based Reasoning (POCBR). For this purpose, we present related approaches and evaluate them against derived requirements for visualizations in POCBR. On this basis, suitable visualizations are further developed as well as new approaches designed. Three such visualizations are created: (1) a graph mapping approach, (2) a merge graph, and (3) a visualization based on heatmaps. An evaluation of these approaches has been performed based on the requirements in which the domain experts determine the graph-mapping visualization as best-suited for engineering of similarity models.},
  keywords 	   = {Visualization, Explanation, Similarity, Process-Oriented Case-Based Reasoning, Explainable Case-Based Reasoning}
}

Modeling similarity measures in Case-Based Reasoning is a knowledge-intensive, demanding, and error-prone task even for domain experts. Visualizations offer support for users, but are currently only available for certain subdomains and case representations. Currently, there are only visualizations that can be used for local attributes or specific case representations. However, there is no possibility to visualize similarities between complete processes accordingly so far, although complex domains may be present. Therefore, an extension of existing approaches or the design of new suitable concepts for this application domain is necessary. The contribution of this work is to enable a more profound understanding of similarity for knowledge engineers who create a similarity model and support them in this task by using visualization methods in Process-Oriented Case-Based Reasoning (POCBR). For this purpose, we present related approaches and evaluate them against derived requirements for visualizations in POCBR. On this basis, suitable visualizations are further developed as well as new approaches designed. Three such visualizations are created: (1) a graph mapping approach, (2) a merge graph, and (3) a visualization based on heatmaps. An evaluation of these approaches has been performed based on the requirements in which the domain experts determine the graph-mapping visualization as best-suited for engineering of similarity models.

@inproceedings{Brand.2023_RLForAdaptiveWorkflows,
  author       = {Florian Brand and
                  Katharina Lott and
                  Lukas Malburg and
                  Maximilian Hoffmann and
                  Ralph Bergmann},
  editor       = {Lukas Malburg and
                  Deepika Verma},
  title        = {{Using Deep Reinforcement Learning for the Adaptation of Semantic Workflows}},
  booktitle    = {Proceedings of the Workshops at the 31st International Conference
                  on Case-Based Reasoning {(ICCBR-WS} 2023) co-located with the 31st
                  International Conference on Case-Based Reasoning {(ICCBR} 2023), Aberdeen,
                  Scotland, UK, July 17, 2023},
  series       = {{CEUR} Workshop Proceedings},
  volume       = {3438},
  pages        = {55--70},
  publisher    = {CEUR-WS.org},
  year         = {2023},
  keywords     = {{Case-Based Reasoning, Semantic Workflows, Deep Learning, Reinforcement Learning}},
  abstract     = {Case-Based Reasoning (CBR) solves new problems by using experience represented by solved cases. The acquisition of adaptation knowledge and its subsequent application remains a classic challenge for CBR applications. In this paper, we present a novel approach for adapting semantic workflows during the reuse phase of the CBR cycle. A reinforcement learning agent is utilized, which applies different actions to change nodes of the workflow. Thereby, changes to the workflow are made by replacing, deleting or adding nodes. The agent is evaluated in a case study outlining its ability to adapt a semantic graph in a smart manufacturing domain. While the approach is detailed for the application in the particular domain, it can be adopted for the usage in other process-oriented domains.},
  url          = {http://www.wi2.uni-trier.de/shared/publications/2023_Brand_RLForAdaptiveWorkflows.pdf}
}

Case-Based Reasoning (CBR) solves new problems by using experience represented by solved cases. The acquisition of adaptation knowledge and its subsequent application remains a classic challenge for CBR applications. In this paper, we present a novel approach for adapting semantic workflows during the reuse phase of the CBR cycle. A reinforcement learning agent is utilized, which applies different actions to change nodes of the workflow. Thereby, changes to the workflow are made by replacing, deleting or adding nodes. The agent is evaluated in a case study outlining its ability to adapt a semantic graph in a smart manufacturing domain. While the approach is detailed for the application in the particular domain, it can be adopted for the usage in other process-oriented domains.

@article{Malburg_MAPEK_Loops_2023,
	title        = {{Applying MAPE-K control loops for adaptive workflow management in smart factories}},
	author       = {Lukas Malburg and Maximilian Hoffmann and Ralph Bergmann},
	year         = 2023,
	journal      = {{Journal of Intelligent Information Systems}},
	pages		 = {83--111},
	volume       = {61},
	number       = {1},
	doi          = {10.1007/s10844-022-00766-w},
	url          = {http://www.wi2.uni-trier.de/shared/publications/2023_MalburgEtAl_MAPEK_Loops.pdf},
	keywords     = {{Complex event processing, Automated planning, Cyber-physical environments, Smart factories, Adaptive workflow management, Process adaptation}},
	abstract     = {Monitoring the state of currently running processes and reacting to ad-hoc situations during runtime is a key challenge in Business Process Management (BPM). This is especially the case in cyber-physical environments that are characterized by high context sensitivity. MAPE-K control loops are widely used for self-management in these environments and describe four phases for approaching this challenge: Monitor, Analyze, Plan, and Execute. In this paper, we present an architectural solution as well as implementation proposals for using MAPE-K control loops for adaptive workflow management in smart factories. We use Complex Event Processing (CEP) techniques and the process execution states of a Workflow Management System (WfMS) in the monitoring phase. In addition, we apply automated planning techniques to resolve detected exceptional situations and to continue process execution. The experimental evaluation with a physical smart factory shows the potential of the developed approach that is able to detect failures by using IoT sensor data and to resolve them autonomously in near real time with considerable results.}
}

Monitoring the state of currently running processes and reacting to ad-hoc situations during runtime is a key challenge in Business Process Management (BPM). This is especially the case in cyber-physical environments that are characterized by high context sensitivity. MAPE-K control loops are widely used for self-management in these environments and describe four phases for approaching this challenge: Monitor, Analyze, Plan, and Execute. In this paper, we present an architectural solution as well as implementation proposals for using MAPE-K control loops for adaptive workflow management in smart factories. We use Complex Event Processing (CEP) techniques and the process execution states of a Workflow Management System (WfMS) in the monitoring phase. In addition, we apply automated planning techniques to resolve detected exceptional situations and to continue process execution. The experimental evaluation with a physical smart factory shows the potential of the developed approach that is able to detect failures by using IoT sensor data and to resolve them autonomously in near real time with considerable results.

@article{Guldner_AICPPS_2023,
	url = {https://doi.org/10.1515/itit-2023-0001},
	title = {A framework for AI-based self-adaptive cyber-physical process systems},
	author = {Achim Guldner and Maximilian Hoffmann and Christian Lohr and Rüdiger Machhamer and Lukas Malburg and Marlies Morgen and Stephanie C. Rodermund and Florian Schäfer and Lars Schaupeter and Jens Schneider and Felix Theusch and Ralph Bergmann and Guido Dartmann and Norbert Kuhn and Stefan Naumann and Ingo J. Timm and Matthias Vette-Steinkamp and Benjamin Weyers},
	journal = {it - Information Technology},
	volume = {65},
	number = {3},
	url          = {http://www.wi2.uni-trier.de/shared/publications/2023_Guldner_AICPPS.pdf},
	doi = {doi:10.1515/itit-2023-0001},
	keywords     = {{Artificial Intelligence, Business Process Managemenet, Cyber-Physical Systems, Framework, Green AI, Process-Aware Information System}},
	abstract     = {{Digital transformation is both an opportunity and a challenge. To take advantage of this opportunity for humans and the environment, the transformation process must be understood as a design process that affects almost all areas of life. In this paper, we investigate AI-Based Self-Adaptive Cyber-Physical Process Systems (AI-CPPS) as an extension of the traditional CPS view. As contribution, we present a framework that addresses challenges that arise from recent literature. The aim of the AI-CPPS framework is to enable an adaptive integration of IoT environments with higher-level process-oriented systems. In addition, the
framework integrates humans as actors into the system, which is often neglected by recent related approaches. The framework consists of three layers, i.e., processes, semantic modeling, and systems and actors, and we describe for each layer challenges and solution outlines for application. We also address the requirement to enable the integration of new networked devices under the premise of a targeted process that is optimally designed for humans, while profitably integrating AI and IoT. It is expected that AI-CPPS can contribute significantly to increasing sustainability and quality of life and offer solutions to pressing problems such as environmental protection, mobility, or demographic change. Thus, it is all the more important that the systems themselves do not become a driver of resource consumption.}},
	year = {2023}
}

Digital transformation is both an opportunity and a challenge. To take advantage of this opportunity for humans and the environment, the transformation process must be understood as a design process that affects almost all areas of life. In this paper, we investigate AI-Based Self-Adaptive Cyber-Physical Process Systems (AI-CPPS) as an extension of the traditional CPS view. As contribution, we present a framework that addresses challenges that arise from recent literature. The aim of the AI-CPPS framework is to enable an adaptive integration of IoT environments with higher-level process-oriented systems. In addition, the framework integrates humans as actors into the system, which is often neglected by recent related approaches. The framework consists of three layers, i.e., processes, semantic modeling, and systems and actors, and we describe for each layer challenges and solution outlines for application. We also address the requirement to enable the integration of new networked devices under the premise of a targeted process that is optimally designed for humans, while profitably integrating AI and IoT. It is expected that AI-CPPS can contribute significantly to increasing sustainability and quality of life and offer solutions to pressing problems such as environmental protection, mobility, or demographic change. Thus, it is all the more important that the systems themselves do not become a driver of resource consumption.

@inproceedings{pauli_transfer_learning_2023,
	title        = {{Similarity-Based Retrieval in Process-Oriented Case-Based Reasoning Using Graph Neural Networks and Transfer Learning}},
	author       = {Pauli, Johannes and Hoffmann, Maximilian and Bergmann, Ralph},
	year         = 2023,
	booktitle    = {Proceedings of the 36th International Florida Artificial Intelligence Research Society Conference, FLAIRS 2023, Clearwater Beach, Florida, USA},
	doi          = {10.32473/flairs.36.133040},
	url          = {http://www.wi2.uni-trier.de/shared/publications/2023_Pauli_FLAIRS.pdf},
	abstract     = {Similarity-based retrieval of semantic graphs is a crucial task of Process-Oriented Case-Based Reasoning (POCBR) that is usually complex and time-consuming, as it requires some kind of inexact graph matching. Previous work tackles this problem by using Graph Neural Networks (GNNs) to learn pairwise graph similarities. In this paper, we present a novel approach that improves on the GNN-based case retrieval with a Transfer Learning (TL) setup, composed of two phases: First, the pretraining phase trains a model for assessing the similarities between graph nodes and edges and their semantic annotations. Second, the pretrained model is then integrated into the GNN model by either using fine-tuning, i.e., the parameters of the pretrained model are further trained, or feature extraction, i.e., the parameters of the pretrained model are converted to constants. The experimental evaluation examines the quality and performance of the models based on TL compared to the GNN models from previous work for three semantic graph domains with various properties. The results show the great potential of the proposed approach for reducing the similarity prediction error and the training time.}
}

Similarity-based retrieval of semantic graphs is a crucial task of Process-Oriented Case-Based Reasoning (POCBR) that is usually complex and time-consuming, as it requires some kind of inexact graph matching. Previous work tackles this problem by using Graph Neural Networks (GNNs) to learn pairwise graph similarities. In this paper, we present a novel approach that improves on the GNN-based case retrieval with a Transfer Learning (TL) setup, composed of two phases: First, the pretraining phase trains a model for assessing the similarities between graph nodes and edges and their semantic annotations. Second, the pretrained model is then integrated into the GNN model by either using fine-tuning, i.e., the parameters of the pretrained model are further trained, or feature extraction, i.e., the parameters of the pretrained model are converted to constants. The experimental evaluation examines the quality and performance of the models based on TL compared to the GNN models from previous work for three semantic graph domains with various properties. The results show the great potential of the proposed approach for reducing the similarity prediction error and the training time.

@inproceedings{hoffmann_ltr_2023,
	title        = {{Ranking-Based Case Retrieval with Graph Neural Networks in Process-Oriented Case-Based Reasoning}},
	author       = {Hoffmann, Maximilian and Bergmann, Ralph},
	year         = 2023,
	booktitle    = {Proceedings of the 36th International Florida Artificial Intelligence Research Society Conference, FLAIRS 2023, Clearwater Beach, Florida, USA},
	doi          = {10.32473/flairs.36.133039},
	url          = {http://www.wi2.uni-trier.de/shared/publications/2023_Hoffmann_FLAIRS.pdf},
	abstract     = {In Process-Oriented Case-Based Reasoning (POCBR), experiential knowledge from previous problem-solving situations is retrieved from a case base to be reused for upcoming problems. The task of retrieval is approached in previous work by using Graph Neural Networks (GNNs) to learn workflow similarities which are, in turn, used to find similar workflows w.r.t. a query workflow. This paper is motivated by the fact that these GNNs are mostly used for predicting the similarity between two workflows (query and case), while the retrieval in CBR is only concerned with the ranking of the most similar workflows from the case base w.r.t. the query. Thus, we propose a novel approach to extend the GNN-based workflow retrieval by a Learning-to-Rank (LTR) component where rankings instead of similarities between cases are predicted. The main contribution of this paper addresses the changes to the GNNs from previous work, such that their model architecture predicts pairwise preferences between cases w.r.t. a query and that they can be trained using labeled preference data. In order to transform these preferences into a case ranking, we also describe rank aggregation methods with different levels of computational complexity. The experimental evaluation compares different models for predicting similarities and rankings in case retrieval scenarios. The results indicate the potential of our ranking-based approach in significantly improving retrieval quality with only small impacts on the performance.}
}

In Process-Oriented Case-Based Reasoning (POCBR), experiential knowledge from previous problem-solving situations is retrieved from a case base to be reused for upcoming problems. The task of retrieval is approached in previous work by using Graph Neural Networks (GNNs) to learn workflow similarities which are, in turn, used to find similar workflows w.r.t. a query workflow. This paper is motivated by the fact that these GNNs are mostly used for predicting the similarity between two workflows (query and case), while the retrieval in CBR is only concerned with the ranking of the most similar workflows from the case base w.r.t. the query. Thus, we propose a novel approach to extend the GNN-based workflow retrieval by a Learning-to-Rank (LTR) component where rankings instead of similarities between cases are predicted. The main contribution of this paper addresses the changes to the GNNs from previous work, such that their model architecture predicts pairwise preferences between cases w.r.t. a query and that they can be trained using labeled preference data. In order to transform these preferences into a case ranking, we also describe rank aggregation methods with different levels of computational complexity. The experimental evaluation compares different models for predicting similarities and rankings in case retrieval scenarios. The results indicate the potential of our ranking-based approach in significantly improving retrieval quality with only small impacts on the performance.

@inproceedings{hoffmann_hyperparameters_2022,
	title        = {{Improving Automated Hyperparameter Optimization with Case-Based Reasoning}},
	author       = {Maximilian Hoffmann and Ralph Bergmann},
	year         = 2022,
	booktitle    = {Case-Based Reasoning Research and Development - 30th International Conference, {ICCBR} 2022, Nancy, France, September 12-15, 2022, Proceedings},
	publisher    = {Springer},
	series       = {Lecture Notes in Computer Science},
	volume       = 13405,
	pages        = {273--288},
	doi          = {10.1007/978-3-031-14923-8\_18},
	url          = {http://www.wi2.uni-trier.de/shared/publications/2022_ICCBR__Hyperparameter_Optimization_with_CBR.pdf},
	editor       = {Mark T. Keane and Nirmalie Wiratunga},
	abstract     = {The hyperparameter configuration of machine learning models has a great influence on their performance. These hyperparameters are often set either manually w. r. t. to the experience of an expert or by an Automated Hyperparameter Optimization (HPO) method. However, integrating experience knowledge into HPO methods is challenging. Therefore, we propose the approach HypOCBR (Hyperparameter Optimization with Case-Based Reasoning) that uses Case-Based Reasoning (CBR) to improve the optimization of hyperparameters. HypOCBR is used as an addition to HPO methods and builds up a case base of sampled hyperparameter vectors with their loss values. The case base is then used to retrieve hyperparameter vectors given a query vector and to make decisions whether to proceed trialing with this query or abort and sample another vector. The experimental evaluation investigates the suitability of HypOCBR for two deep learning setups of varying complexity. It shows its potential to improve the optimization results, especially in complex scenarios with limited optimization time.}
}

The hyperparameter configuration of machine learning models has a great influence on their performance. These hyperparameters are often set either manually w. r. t. to the experience of an expert or by an Automated Hyperparameter Optimization (HPO) method. However, integrating experience knowledge into HPO methods is challenging. Therefore, we propose the approach HypOCBR (Hyperparameter Optimization with Case-Based Reasoning) that uses Case-Based Reasoning (CBR) to improve the optimization of hyperparameters. HypOCBR is used as an addition to HPO methods and builds up a case base of sampled hyperparameter vectors with their loss values. The case base is then used to retrieve hyperparameter vectors given a query vector and to make decisions whether to proceed trialing with this query or abort and sample another vector. The experimental evaluation investigates the suitability of HypOCBR for two deep learning setups of varying complexity. It shows its potential to improve the optimization results, especially in complex scenarios with limited optimization time.

@inproceedings{hoffmann_gpu_astar_2022,
	title        = {{GPU-Based Graph Matching for Accelerating Similarity Assessment in Process-Oriented Case-Based Reasoning}},
	author       = {Maximilian Hoffmann and Lukas Malburg and Nico Bach and Ralph Bergmann},
	year         = 2022,
	booktitle    = {Case-Based Reasoning Research and Development - 30th International Conference, {ICCBR} 2022, Nancy, France, September 12-15, 2022, Proceedings},
	publisher    = {Springer},
	series       = {Lecture Notes in Computer Science},
	volume       = 13405,
	pages        = {240--255},
	doi          = {10.1007/978-3-031-14923-8\_16},
	url          = {http://www.wi2.uni-trier.de/shared/publications/2022_ICCBR__A_Star_GPU.pdf},
	editor       = {Mark T. Keane and Nirmalie Wiratunga},
	abstract     = {In Process-Oriented Case-Based Reasoning (POCBR), determining the similarity between cases represented as semantic graphs often requires some kind of inexact graph matching, which generally is an NP-hard problem. Heuristic search algorithms such as A* search have been successfully applied for this task, but the computational performance is still a limiting factor for large case bases. As related work shows a great potential for accelerating A* search by using GPUs, we propose a novel approach called AMonG for efficiently computing graph similarities with an A* graph matching process involving GPU computing. The three-phased matching process distributes the search process over multiple search instances running in parallel on the GPU. We develop and examine different strategies within these phases that allow to customize the matching process adjusted to the problem situation to be solved. The experimental evaluation compares the proposed GPU-based approach with a pure CPU-based one. The results clearly demonstrate that the GPU-based approach significantly outperforms the CPU-based approach in a retrieval scenario, leading to an average speedup factor of 16.}
}

In Process-Oriented Case-Based Reasoning (POCBR), determining the similarity between cases represented as semantic graphs often requires some kind of inexact graph matching, which generally is an NP-hard problem. Heuristic search algorithms such as A* search have been successfully applied for this task, but the computational performance is still a limiting factor for large case bases. As related work shows a great potential for accelerating A* search by using GPUs, we propose a novel approach called AMonG for efficiently computing graph similarities with an A* graph matching process involving GPU computing. The three-phased matching process distributes the search process over multiple search instances running in parallel on the GPU. We develop and examine different strategies within these phases that allow to customize the matching process adjusted to the problem situation to be solved. The experimental evaluation compares the proposed GPU-based approach with a pure CPU-based one. The results clearly demonstrate that the GPU-based approach significantly outperforms the CPU-based approach in a retrieval scenario, leading to an average speedup factor of 16.

@inproceedings{Hoffmann.2022_ProGAN,
	title        = {{ProGAN: Toward a Framework for Process Monitoring and Flexibility by Change via Generative Adversarial Networks}},
	author       = {Maximilian Hoffmann and Lukas Malburg and Ralph Bergmann},
	year         = 2022,
	booktitle    = {Business Process Management Workshops - {BPM} 2021, Rome, Italy, September 6 - 10, 2021},
	publisher    = {Springer.},
	series       = {Lecture Notes in Business Information Processing},
	volume       = 436,
	pages        = {43--55},
	doi          = {10.1007/978-3-030-94343-1\_4},
	url          = {https://doi.org/10.1007/978-3-030-94343-1\_4},
	url          = {http://www.wi2.uni-trier.de/shared/publications/2022_AI4BPM_HoffmannEtal_ProGAN.pdf},
	note         = {The original publication is available at www.springerlink.com},
	editor       = {Andrea Marrella and Barbara Weber},
	keywords     = {{Business process prediction, Generative Adversarial Networks, Flexibility by change, Process adaptation}},
	abstract     = {Monitoring the state of currently running processes and reacting to deviations during runtime is a key challenge in Business Process Management (BPM). The MAPE-K control loop describes four phases for approaching this challenge: Monitor, Analyze, Plan, Execute. In this paper, we present the ProGAN framework, an idea of an approach for implementing the monitor, analyze, and plan phases of MAPE-K. For this purpose, we leverage a deep learning architecture that builds upon Generative Adversarial Networks (GANs): The discriminator is used for monitoring the process in its environment by using sensor data and for detecting deviations w.r.t. the desired process state (monitor phase). The generator is used afterwards for analyzing the detected deviation and its symptoms as well as for adapting the current process to resolve the deviation and to restore the desired state. Both components are trained together by utilizing each other's feedback in a self-supervised way. We demonstrate the application of our approach for an exemplary scenario in the manufacturing domain.}
}

Monitoring the state of currently running processes and reacting to deviations during runtime is a key challenge in Business Process Management (BPM). The MAPE-K control loop describes four phases for approaching this challenge: Monitor, Analyze, Plan, Execute. In this paper, we present the ProGAN framework, an idea of an approach for implementing the monitor, analyze, and plan phases of MAPE-K. For this purpose, we leverage a deep learning architecture that builds upon Generative Adversarial Networks (GANs): The discriminator is used for monitoring the process in its environment by using sensor data and for detecting deviations w.r.t. the desired process state (monitor phase). The generator is used afterwards for analyzing the detected deviation and its symptoms as well as for adapting the current process to resolve the deviation and to restore the desired state. Both components are trained together by utilizing each other's feedback in a self-supervised way. We demonstrate the application of our approach for an exemplary scenario in the manufacturing domain.

@article{Hoffmann.2022_GraphEmbeddingPOCBR,
	title        = {{Using Graph Embedding Techniques in Process-Oriented Case-Based Reasoning}},
	author       = {Maximilian Hoffmann and Ralph Bergmann},
	year         = 2022,
	journal      = {Algorithms},
	volume       = 15,
	number       = 2,
	doi          = {10.3390/a15020027},
	issn         = {1999-4893},
	url          = {https://www.mdpi.com/1999-4893/15/2/27/pdf},
	abstract     = {Similarity-based retrieval of semantic graphs is a core task of Process-Oriented Case-Based Reasoning (POCBR) with applications in real-world scenarios, e.g., in smart manufacturing. The involved similarity computation is usually complex and time-consuming, as it requires some kind of inexact graph matching. To tackle these problems, we present an approach to modeling similarity measures based on embedding semantic graphs via Graph Neural Networks (GNNs). Therefore, we first examine how arbitrary semantic graphs, including node and edge types and their knowledge-rich semantic annotations, can be encoded in a numeric format that is usable by GNNs. Given this, the architecture of two generic graph embedding models from the literature is adapted to enable their usage as a similarity measure for similarity-based retrieval. Thereby, one of the two models is more optimized towards fast similarity prediction, while the other model is optimized towards knowledge-intensive, more expressive predictions. The evaluation examines the quality and performance of these models in preselecting retrieval candidates and in approximating the ground-truth similarities of a graph-matching-based similarity measure for two semantic graph domains. The results show the great potential of the approach for use in a retrieval scenario, either as a preselection model or as an approximation of a graph similarity measure.}
}

Similarity-based retrieval of semantic graphs is a core task of Process-Oriented Case-Based Reasoning (POCBR) with applications in real-world scenarios, e.g., in smart manufacturing. The involved similarity computation is usually complex and time-consuming, as it requires some kind of inexact graph matching. To tackle these problems, we present an approach to modeling similarity measures based on embedding semantic graphs via Graph Neural Networks (GNNs). Therefore, we first examine how arbitrary semantic graphs, including node and edge types and their knowledge-rich semantic annotations, can be encoded in a numeric format that is usable by GNNs. Given this, the architecture of two generic graph embedding models from the literature is adapted to enable their usage as a similarity measure for similarity-based retrieval. Thereby, one of the two models is more optimized towards fast similarity prediction, while the other model is optimized towards knowledge-intensive, more expressive predictions. The evaluation examines the quality and performance of these models in preselecting retrieval candidates and in approximating the ground-truth similarities of a graph-matching-based similarity measure for two semantic graph domains. The results show the great potential of the approach for use in a retrieval scenario, either as a preselection model or as an approximation of a graph similarity measure.

@inproceedings{kumar_dependencyretrieval_2022,
	title        = {{Considering Inter-Case Dependencies During Similarity-Based Retrieval in Process-Oriented Case-Based Reasoning}},
	author       = {Kumar, Rahol and Schultheis, Alexander and Malburg, Lukas and Hoffmann, Maximilian and Bergmann, Ralph},
	year         = 2022,
	booktitle    = {Proceedings of the 35th International Florida Artificial Intelligence Research Society Conference, FLAIRS 2022, Hutchinson Island, Jensen Beach, Florida, USA},
	doi          = {10.32473/flairs.v35i.130680},
	url          = {http://www.wi2.uni-trier.de/shared/publications/2022_Kumar_FLAIRS.pdf},
	abstract     = {In Case-Based Reasoning (CBR), knowledge gained from previously experienced problem-solving situations is stored as cases that can be used to solve similar upcoming problems. Although these cases act as independent knowledge entities, dependencies between cases are common in real-world scenarios, despite being only rarely considered during case retrieval or other CBR phases. In this paper, we introduce so-called inter-case dependencies, which are considered in the context of Process-Oriented CBR (POCBR). Therefore, we 1) derive requirements that must be satisfied for considering dependencies during the retrieval phase, 2) analyze which knowledge representations are suitable for representing dependencies between cases, and, 3) present our approach for Dependency-Guided Retrieval (DGR) that considers these dependencies between cases during the retrieval phase. In the experimental evaluation, the proposed DGR approach is compared to a regular CBR approach in case retrieval scenarios from the cooking domain. The results demonstrate that the use of the DGR approach leads to significantly reduced times for human problem-solving compared to regular CBR.}
}

In Case-Based Reasoning (CBR), knowledge gained from previously experienced problem-solving situations is stored as cases that can be used to solve similar upcoming problems. Although these cases act as independent knowledge entities, dependencies between cases are common in real-world scenarios, despite being only rarely considered during case retrieval or other CBR phases. In this paper, we introduce so-called inter-case dependencies, which are considered in the context of Process-Oriented CBR (POCBR). Therefore, we 1) derive requirements that must be satisfied for considering dependencies during the retrieval phase, 2) analyze which knowledge representations are suitable for representing dependencies between cases, and, 3) present our approach for Dependency-Guided Retrieval (DGR) that considers these dependencies between cases during the retrieval phase. In the experimental evaluation, the proposed DGR approach is compared to a regular CBR approach in case retrieval scenarios from the cooking domain. The results demonstrate that the use of the DGR approach leads to significantly reduced times for human problem-solving compared to regular CBR.

@inproceedings{malburg_gpuretrieval_2021,
	title        = {{Improving Similarity-Based Retrieval Efficiency by Using Graphic Processing Units in Case-Based Reasoning}},
	author       = {Malburg, Lukas and Hoffmann, Maximilian and Trumm, Simon and Bergmann, Ralph},
	year         = 2021,
	booktitle    = {Proceedings of the 34th International Florida Artificial Intelligence Research Society Conference, FLAIRS 2021, North Miami Beach, Florida, USA},
	doi          = {10.32473/flairs.v34i1.128345},
	url          = {https://doi.org/10.32473/flairs.v34i1.128345},
	url          = {http://www.wi2.uni-trier.de/shared/publications/2021_MalburgEtAl_ImprovingRetrievalByGPUs.pdf},
	note         = {Best Student Paper.},
	abstract     = {The accelerated growth of available data causes case bases of increasing sizes and thus lowers efficiency during the case retrieval phase in Case-Based Reasoning (CBR) systems. Even though, many complex and data-intensive tasks are solved by using Graphic Processing Units (GPUs), its application in CBR research has yet to advance past the early stage phase. In this paper, we present an approach to use CUDA-compatible GPUs for similarity assessment of structural, feature vector based cases. Our approach supports several syntactic and semantic similarity measures and is implemented in the open-source case-based reasoning framework ProCAKE. When comparing to current retrieval techniques that calculate similarities on the CPU, our GPU-based approach outperforms them by a factor of up to 37. In addition, our evaluation indicates that the performance gains increase with higher case complexity.}
}

The accelerated growth of available data causes case bases of increasing sizes and thus lowers efficiency during the case retrieval phase in Case-Based Reasoning (CBR) systems. Even though, many complex and data-intensive tasks are solved by using Graphic Processing Units (GPUs), its application in CBR research has yet to advance past the early stage phase. In this paper, we present an approach to use CUDA-compatible GPUs for similarity assessment of structural, feature vector based cases. Our approach supports several syntactic and semantic similarity measures and is implemented in the open-source case-based reasoning framework ProCAKE. When comparing to current retrieval techniques that calculate similarities on the CPU, our GPU-based approach outperforms them by a factor of up to 37. In addition, our evaluation indicates that the performance gains increase with higher case complexity.

@article{Hoffmann.2021_InformedMLCBR,
	title        = {Informed Machine Learning for Improved Similarity Assessment in Process-Oriented Case-Based Reasoning},
	author       = {Maximilian Hoffmann and Ralph Bergmann},
	year         = 2021,
	journal      = {CoRR},
	volume       = {abs/2106.15931},
	url          = {https://arxiv.org/pdf/2106.15931.pdf},
	note         = {Presented at the IJCAI-21 workshop on Deep Learning, Case-Based Reasoning, and AutoML: Present and Future Synergies, Montreal, August 21-26},
	archiveprefix = {arXiv},
	eprint       = {2106.15931}
}

@inproceedings{hoffmann_graph_embedding_2020,
	title        = {{Using Siamese Graph Neural Networks for Similarity-Based Retrieval in Process-Oriented Case-Based Reasoning}},
	author       = {Maximilian Hoffmann and Lukas Malburg and Patrick Klein and Ralph Bergmann},
	year         = 2020,
	booktitle    = {Case-Based Reasoning Research and Development - 28th International Conference, {ICCBR} 2020, Salamanca, Spain, June 8-12, 2020, Proceedings},
	publisher    = {Springer},
	series       = {Lecture Notes in Computer Science},
	volume       = 12311,
	pages        = {229--244},
	doi          = {10.1007/978-3-030-58342-2\_15},
	url          = {http://www.wi2.uni-trier.de/shared/publications/2020_ICCBR__Workflow_Graph_Embedding.pdf},
	note         = {The original publication is available at www.springerlink.com},
	editor       = {Ian Watson and Rosina O. Weber},
	abstract     = {Similarity-based retrieval of semantic graphs is widely used in real-world scenarios, e.g., in the domain of business workflows. To tackle the problem of complex and time-consuming graph similarity computations during retrieval, the MAC/FAC approach is used in Process- Oriented Case-Based Reasoning (POCBR), where similar graphs are extracted from a preselected set of candidate graphs. These graphs result from a similarity computation with a computationally inexpensive similarity measure. The contribution of this paper is a novel similarity measure where vector space embeddings generated by two siamese Graph Neural Networks (GNNs) are used to approximate the similarities of a precise but therefore computationally complex graph similarity measure. This includes a special scheme for encoding semantic graphs to be used in the neural networks. The evaluation examines the quality and performance of these models in preselecting retrieval candidates and in approximating the ground-truth similarities of the graph similarity measure. The results show great potential of the approach for being used in a MAC/FAC scenario, either as a preselection model or as an approximation of the graph similarity measure.}
}

Similarity-based retrieval of semantic graphs is widely used in real-world scenarios, e.g., in the domain of business workflows. To tackle the problem of complex and time-consuming graph similarity computations during retrieval, the MAC/FAC approach is used in Process- Oriented Case-Based Reasoning (POCBR), where similar graphs are extracted from a preselected set of candidate graphs. These graphs result from a similarity computation with a computationally inexpensive similarity measure. The contribution of this paper is a novel similarity measure where vector space embeddings generated by two siamese Graph Neural Networks (GNNs) are used to approximate the similarities of a precise but therefore computationally complex graph similarity measure. This includes a special scheme for encoding semantic graphs to be used in the neural networks. The evaluation examines the quality and performance of these models in preselecting retrieval candidates and in approximating the ground-truth similarities of the graph similarity measure. The results show great potential of the approach for being used in a MAC/FAC scenario, either as a preselection model or as an approximation of the graph similarity measure.

@inproceedings{zeyen_nutrients_2018,
	title        = {{Considering Nutrients during the Generation of Recipes by Process-Oriented Case-based Reasoning}},
	author       = {Zeyen, Christian and Hoffmann, Maximilian and M{\"u}ller, Gilbert and Bergmann, Ralph},
	year         = 2018,
	booktitle    = {Case-{Based} {Reasoning} {Research} and {Development}: 26th {International} {Conference}, {ICCBR} 2018, {Stockholm}, {Sweden}, {July} 9-12, 2018, {Proceedings}},
	publisher    = {Springer},
	series       = {Lecture {Notes} in {Artificial} {Intelligence}},
	volume       = 11156,
	pages        = {464--479},
	url          = {http://www.wi2.uni-trier.de/shared/publications/2018_ZeyenHoffmannMuellerBergmann_ICCBR.pdf},
	note         = {The original publication is available at www.springerlink.com},
	editor       = {Cox, Michael T. and Funk, Peter and Begum, Shahina}
}