Event Log Processing Using Embedding Model Clustering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for processing event logs are inefficient in handling repeating events, which obscure process patterns and require complex analysis, especially in large-scale data, and lack scalability.
Innovation Solution
A method that identifies repeating events in an event log, clusters them using an events embedding model, and replaces them with identifiers, maintaining contextual information while reducing redundancy, employing machine learning and data processing techniques to improve process mining efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If repeating events are maintained in the event log, then information regarding process structure and event contexts is preserved, but redundancy increases and process patterns become obfuscated
Solution Approach 1:
Multiple repeating events that share common characteristics are merged into a single representative event with an associated event type identifier. This consolidation reduces the number of events in the log while preserving the structural information through the event type classification system.
Solution Approach 2:
A universal event type identifier system is introduced that can categorize and represent multiple specific events under common types. This allows the event log to maintain structural information at an abstract level while reducing redundancy, as the same identifier can represent different instances of the same event type.
2Productivity
If repeating events are reduced in the event log, then process mining throughput is improved, but detailed event information may be lost
Solution Approach 1:
Instead of storing multiple detailed event records, the system creates a simplified copy or representation of repeating events using event type identifiers. This representative copy captures the essential structural information needed for process mining while significantly reducing the data volume and improving processing throughput.
3Manufacturing precision
If existing approaches distinguish between events near start and end of process, then repeating events are refined, but scalability is poor for large-scale data
Solution Approach 1:
The system changes the parameter used for event differentiation from positional (start/end of process) to contextual (event type characteristics). By using event type identifiers that capture meaningful distinctions, the system achieves refined event labeling that scales efficiently to large datasets without relying on position-based distinctions.
Data Source
AI summary
Presented are concepts for processing an event log. Once such concept obtains an event log comprising a log of event occurrences for an executed process. It also obtains an events embedding model representative of relationships between a plurality of events of one or more processes. Based on the events embedding model, repeating events in the event log are clustered into one or more groups, and each of the one or more groups are associated with a respective identifier. Repeating events in the event log are then replaced with the identifier associated with the group that the repeating event is a member of.