Parallelism-Based Session Creation to Identify High-Level Activities in Event Log Abstraction
Küçük Resim Yok
Tarih
2024
Yazarlar
Dergi Başlığı
Dergi ISSN
Cilt Başlığı
Yayıncı
Springer International Publishing Ag
Erişim Hakkı
info:eu-repo/semantics/closedAccess
Özet
Process mining utilizes event data to gain insights into the execution of processes. While techniques are valuable, their effectiveness may be hindered when dealing with highly complex processes that have a vast number of variants. Additionally, because the recorded events in information systems are at a low-level, process mining techniques may not align with the higher-level concepts understood at the business level. Without abstracting event sequences to higher-level concepts, the outcomes of process mining, such as discovering a model, can become overly complex and challenging to interpret, rendering them less useful. Some research has been conducted on event abstraction, often requiring significant domain knowledge that may not be readily accessible. Alternatively, unsupervised abstraction techniques may yield less accurate results and rely on stronger assumptions. This paper introduces a technique that addresses the challenge of limited domain knowledge by utilizing a straightforward approach. The technique involves dividing traces into batch sessions, taking into account relationships between subsequent events. Each session is then abstracted as a single high-level activity execution. This abstraction process utilizes a combination of automatic clustering and visualization methods. The proposed technique was evaluated using a randomly generated process model with high variability. The results demonstrate the significant advantages of the proposed abstraction in effectively communicating accurate knowledge to stakeholders.
Açıklama
5th International Conference on Process Mining (ICPM) -- OCT 23-27, 2023 -- Rome, ITALY
Anahtar Kelimeler
Event log abstraction, Parallel relationships, Clustering, Visual analytics, Process discovery
