Segmentation and Process Assignment of Semi-Structured Event Logs

Process mining provides valuable insights by discovering process models from execution logs.However, its effectiveness depends heavily on high-quality, well-structured logs.Many real-world systems produce low-level, semi-structured logs lacking clear process identifiers, causing misalignment with their intended process models.This paper introduces a method for structuring raw event logs by segmenting event streams and mapping them to known processes.Using process traces from experienced users, we develop a model that infers process assignments in unstructured logs.Our approach is motivated by a modular enterprise system without predefined workflows, where dynamic processes generate low-level logs requiring interpretation.We validate our method on a semi-synthetic business dataset and a fully synthetic dataset from PLG2.Our results demonstrate that trace segmentation improves process discovery, aligns logs with meaningful structures, and significantly enhances process mining in unstructured environments.