Process Model Refinement via Predictive Density Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Complex process models mined from execution traces can become overly dense and difficult to interpret due to numerous nodes and edges, making it hard to identify relevant pathways, especially when they represent spaghetti-like behavior with rarely-used or redundant paths.
Innovation Solution
A method is introduced that learns a predictive model from execution traces to determine if the process model is too dense or sparse, allowing for refinement by adjusting the model's specificity or generality, thereby simplifying the model by removing irrelevant information and allowing user-guided control over the refinement process.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If process models are mined from execution traces to capture comprehensive process behavior, then the model completeness is improved, but the model complexity increases making it difficult to interpret
Solution Approach 1:
The patent segments the process model into multiple layers: a high-level abstract model showing main pathways and a detailed model showing complete behavior. This allows users to view only the necessary level of detail, reducing perceived complexity while preserving all information in the detailed layer.
Solution Approach 2:
The patent introduces a new dimension of abstraction level, allowing the same process model to be viewed at different granularities. The mining algorithm operates in this multi-dimensional space, automatically adjusting the level of detail based on user preferences or analytical needs.
2Reliability
If the process model includes all execution paths to ensure completeness, then the model accuracy is improved, but the interpretability deteriorates due to spaghetti-like behavior
Solution Approach 1:
The patent extracts and separates rarely-used or redundant paths from the main process flow. These extracted paths can be viewed independently or hidden by default, allowing the main model to remain clean and interpretable while preserving complete behavioral information in the extracted components.
Solution Approach 2:
The patent makes the process model dynamically configurable, allowing users to interactively adjust which paths are visible and at what level of detail. The model can transition between simplified views for interpretation and detailed views for analysis, adapting to different user needs.
3Measurement precision
If the predictive model is made more specific to improve outcome prediction, then the prediction accuracy is improved, but the model density increases making it harder to identify relevant pathways
Solution Approach 1:
The patent applies different levels of specificity to different parts of the model. High-density detailed modeling is applied only to regions relevant to the target outcome, while other regions use lower-density abstract representations. This local differentiation maintains prediction accuracy for critical pathways while reducing overall model density.
Solution Approach 2:
The patent changes the parameter of model density locally based on relevance to the outcome. The mining algorithm identifies which pathways are most relevant to the predicted outcome and applies higher density only to those regions, adjusting the density parameter dynamically based on outcome relevance.
Data Source
AI summary
Methods and systems for refining a process model include determining whether the process model is too dense or too sparse. A predictive model is learned from execution traces to predict an outcome. The predictive model is modified responsive to the determination of whether the process model is too dense or too sparse. A refined process model is refined from updated traces based on attributes present in the modified predictive model.