Bridging the Gap Between Traditional Process Mining and Object-Centric Process Mining ^†

Abstract

Process mining has become an essential technique for analyzing and optimizing business processes by leveraging digital traces recorded by enterprise systems. However, traditional process mining methods rely heavily on the concept of case identifiers, assuming that each event is associated with only one process instance. This assumption often limits their applicability in complex, real-world environments where multiple objects interact concurrently. This study seeks to connect conventional process mining approaches with the growing domain of object-centric process mining, which provides a broader perspective by considering events linked to multiple business entities. We review the conceptual foundations of both approaches and identify the challenges in transitioning from a case-centric to an object-centric perspective. Our findings demonstrate that object-centric process mining provides richer insights into interconnected process behavior. We conclude that object-centric paradigms mark a significant advancement in process analytics, paving the way for more adaptive and intelligent process improvement frameworks. This study not only bridges conventional process mining approaches with the emerging field of object-centric process mining (OC-PM) but also explores how recent advancements, particularly in Generative AI, are being leveraged within OC-PM frameworks. Specifically, we highlight approaches that integrate Generative AI techniques, including Large Language Models (LLMs), to enhance process understanding and prediction. The integration of AI—especially Generative AI—enables researchers and practitioners to move beyond the limitations and challenges of classical, case-centric process mining, offering more flexible, intelligent, and context-aware process analysis capabilities.

1. Introduction

In the era of digital transformation, companies are progressively depending on analytics-based insights to understand and optimize how their operational workflows function. Process Mining (PM) has become a crucial tool in this domain, enabling the derivation of workflow representations from digital traces captured by enterprise platforms. Traditional Process Mining techniques have proven effective in analyzing workflows based on cases that follow a single entity perspective, such as an order or a claim. However, real-world processes, especially in IoT-enabled environments, often involve multiple interacting entities—machines, products, sensors—creating complex relationships that cannot be captured adequately using classical PM approaches.

The limitations of classical process mining have led to the emergence of Object-Centric Process Mining (OC-PM), a new paradigm that models processing from the perspective of multiple objects, each with its own lifecycle and interconnections. Understanding how to transition from traditional to object-centric perspectives is crucial for fully exploiting the potential of modern event data. Building on this shift, recent approaches within OC-PM have started integrating Artificial Intelligence, particularly Generative AI techniques such as Large Language Models (LLMs). This integration enables a deeper interpretation of complex, multi-object processes and helps overcome the inherent challenges of classical process mining, paving the way for more adaptive, intelligent, and context-aware process analysis.

This study aims to investigate this gap by conducting a systematic exploration of the key distinctions and commonalities between traditional PM and OC-PM. The objective is to highlight the limitations of current traditional methods when applied to multi-object environments, evaluate the advances introduced by OC-PM, and also examine the impact of integrating Generative AI into OC-PM to reduce the challenges of classical PM. By bridging this gap, researchers and practitioners will be better equipped to apply process mining techniques in modern data-intensive environments, thus enhancing their capacity to detect inefficiencies, ensuring compliance, and supporting decision-making with greater accuracy and relevance.

To provide a comprehensive understanding of this transition, this article is organized as follows: the Background section defines key concepts, including traditional Process Mining and Object-Centric Process Mining, establishing a foundational understanding for the reader. The Literature Review outlines the methodology used to identify and analyze relevant academic and industrial research, focusing on how existing studies have approached the comparison between these paradigms. This is followed by the Result Analysis, which addresses the research questions by synthesizing findings from the literature and identifying trends, challenges, and innovations. The Discussion section reflects on these results, offering critical insights and potential implications for future work. Finally, the article concludes by summarizing the main contributions and suggesting directions for further research and practical application.

2. Background

To better understand the shift from traditional process mining to more advanced paradigms, it is important to first define the core concepts underlying each approach. The following definitions provide a foundation for distinguishing between classical Process Mining and its object-centric evolution.

According to [1], Process Mining is a technique that bridges the gap between model-driven process analysis methods (such as simulation and other business process analysis techniques) and data-driven approaches like machine learning and data mining. It enables the discovery, monitoring, and enhancement of real-world processes by extracting valuable insights from event logs commonly available in modern information systems.

According to [2], Object-Centric Process Mining (OC-PM) is an extension of traditional process mining that considers the interactions of multiple object types involved in a process [3]. Unlike classical approaches that focus on a single case notion, OC-PM models processes based on multiple objects—such as people, machines, products, and documents—each with their own lifecycle and interrelations. This allows for a more faithful representation of real-world processes, especially in complex and data-rich environments.

3. Literature Review Methodology

3.1. Research Question

• RQ1: How has object-centric process mining (OC-PM) evolved over time in terms of concepts, methodologies, and practical applications?
• RQ2: What are the key challenges of traditional PM in representing real-world processes that involve multiple interacting objects?
• RQ3: How do traditional (case-centric) process mining techniques compare to object-centric process mining (OC-PM) when applied to the same event logs?
• RQ4: Which algorithms and approaches are most commonly employed in object-centric process mining to solve operational issues of traditional process mining?

3.2. Search String

To gather a comprehensive collection of publications related to object-centric process mining, we defined a general search strategy based on frequently used terminology in the field. Our search string was designed to capture studies addressing both foundational concepts and recent advancements in this area. It combines terms related to event logs, object-centric techniques, and traditional process mining approaches to ensure a broad yet targeted retrieval of relevant literature. The complete search query used is:

(“Object-Centric Event Logs” OR “OCEL logs” OR “multi-object event logs” OR “object-oriented logs”) AND (“Object-Centric Process Mining” OR “OCPM” OR “multi-object process mining” OR “object-aware mining”) AND (“Traditional Process Mining” OR “case-centric process mining” OR “single-case mining”) AND (“LLMs” OR “Generative AI “OR “Large Language Model”)

3.3. Study Selection Criteria

To guide the selection of studies, we defined a set of inclusion and exclusion criteria. These criteria were applied to ensure that only relevant and appropriate research contributions were retained for analysis. A summary of these criteria will be presented in

Table 1. Selection Criteria.

Inclusion Criteria	Exclusion Criteria
Publications in peer-reviewed journals or conference proceedings	Papers are accessible only as abstracts or in PowerPoint.
Articles written in English	Duplicate or retracted papers
Studies published from 2018 onward
Research explicitly addressing object-centric process mining, or object-centric event logs	Studies focusing solely on traditional process mining

.

3.4. Study Sources

To ensure the relevance and quality of the literature analyzed in this study, specific inclusion criteria were established to filter recent, high-quality, and thematically relevant research. The most pertinent criteria are summarized in

Table 2. Sources searched.

No.	Name	URL	Accessed on
1	ArXiv	https://arxiv.org/	12 March 2024
2	Google Scholar	https://scholar.google.com/	1 March 2024
3	IEEE Digital Library	http://ieeexplore.ieee.org	3 April 2024
4	Scopus	https://www.scopus.com/	20 April 2024
5	Web of sciences	https://www.scirp.org/	20 April 2024

.

4. Result Analysis

RQ1: How has object-centric process mining (OC-PM) evolved over time in terms of concepts, methodologies, and practical applications?

Figure 1 illustrates the evolution of Object-Centric Process Mining (OC-PM) over time. The trend shows a clear trajectory of growth and maturity. Initially, between 2018 and 2020, research activity was limited, reflecting the conceptual foundation stage where the core ideas and formal models were being established.

The starting point of 2018 was deliberately chosen, as it marks a turning point in the field where researchers began to critically examine the limitations and challenges of classical process mining. This growing awareness led to the emergence of new approaches, including OC-PM, which explains why most relevant studies in this area have been published from 2018 onward.

From 2021 to 2022, the number of publications increased significantly, indicating a methodological expansion with the introduction of new frameworks, algorithms, and tools tailored to handle multi-object event data. The sharp rise in 2023 and the peak in 2024 highlight a phase of practical application, where OC-PM gained traction in real-world domains such as manufacturing, healthcare, and logistics. This progression illustrates how OC-PM has evolved from a theoretical innovation to a robust analytical approach addressing the complexity of modern, interconnected processes.

Figure 2 summarizes the sources of articles used in the study. The largest portion, 43.2%, comes from arXiv, followed by Google Scholar at 24.7%. Web of Sciences and Scopus contributes 13.3%, while the remaining 7.4% are categorized under other sources. This distribution highlights the dominant role of open-access platforms, particularly arXiv, in providing research material related to the study.

RQ2: What are the key challenges of traditional PM in representing real-world processes that involve multiple interacting objects?

The challenges of techniques used in traditional process mining are summarized in

Table 3. Summary of limitations of Traditional Process Mining (Case-Centric).

Article ID	Traditional Process Mining (Case-Centric)
[4]	Traditional process mining treats the process as a whole, making it difficult to interpret in diverse real-life scenarios, and relies on atomic values in process cubes, limiting its ability to handle variability and complexity.
[4,5,6]	Classical process mining assumes that each event relates to only one case notion, which limits its ability to model realistic one-to-many relationships found in real-world scenarios.
[4,5,6]	Traditional process mining struggles with modeling complex one-to-many and many-to-many relationships, as its event logs link each event to a single case notion and traditional metrics like fitness and precision cannot be directly applied in multi-object contexts.
[6]	Traditional process mining provides an incomplete representation of processes by focusing on a single case notion, which omits related object behaviors and fails to capture dependencies across multiple entities.
[7]	Traditional process mining tools are designed for structured data. However, many processes involve unstructured data (e.g., emails, c PDFs, images), posing challenges in data extraction, interpretation, and integration into process models.

. These techniques typically rely on a single case notion, which means each event is linked to only one object or entity. This constraint makes it difficult to represent complex business processes where multiple entities interact—such as orders, customers, and products. As a result, traditional methods struggle with modeling one-to-many and many-to-many relationships. Additionally, they often produce oversimplified models that do not reflect real-life variability or dependencies between different parts of the process, making interpretation challenging and sometimes misleading.

RQ3: How does classical process mining (case centric) techniques compare to object-centric process mining (OC-PM) when applied to the same event logs?

The characteristics of traditional (case-centric) process mining techniques and object-centric process mining (OC-PM) are summarized in

Table 4. Solutions proposed by OCPM to solve issues of Standard process mining.

Article ID	Limit of Traditional Process Mining (Case-Centric)	Solution of Object-Centric Process Mining (OC-PM)
[4]	Limited to one case notion per event and struggles with one-to-many relationships in real-life processes.	Supports multiple interacting case notions (e.g., orders, items) for a more general and realistic representation of complex business processes.
[5]	Case-centric analysis focusing on individual cases and events.	Utilization of Object-Centric Event Data (OCED) for the identification, analysis, and improvement of intricate processes, highlighting interdependencies that traditional methods miss.
[5]	limited to a retrospective analysis centered on a single object, without considering interactions between processes or enabling realistic multi-object simulation.	Object-Centric Process Mining (OC-PM): Allows for accurate process modeling by removing the constraint of a single case, thanks to many-to-many relationships between events and objects (E2O/O2O), thereby integrating all necessary interactions for realistic simulation.
[8]	Focuses on single-case instances with linear event logs, struggles with complex, multi-object interactions, and provides a limited one-dimensional view of processes.	Supports many-to-many relationships among events and objects, captures process interconnectedness, enables richer analysis through object-centric event logs, and offers a broader, three-dimensional perspective of real-world processes.
[9]	Models are limited to linear or isolated case behaviors based on a single case ID.	Enhanced process models reflecting behaviors across different object classes, enabling richer and more realistic simulations.
[10]	Requires flattening or manipulating event data to fit a single-case structure	Graph-based generalizations preserve the integrity of event data using graph structures to represent multi-object interactions without data loss.
[11]	Traditional PM is limited by its single-case focus, rigid process views, fragmented compliance analysis, weak predictive capabilities, poor AI integration, and redundant data transformations.	OCPM overcomes these issues through multi-object analysis, flexible views, comprehensive compliance and performance insights, enhanced predictive modeling, seamless AI integration, and unified data representation.
[12]	Structures data around single cases and struggles with complex, multi-object interactions; provides high interpretability but limited depth in interconnected processes.	Organizes data around multiple objects, capturing rich inter-object interactions and offering higher fitness and precision in complex real-world scenarios.

. From the following table, we can see that traditional techniques are limited to single case notions, struggle with one-to-many relationships, and often require data manipulation. In contrast, OC-PM supports multiple interacting objects, preserves the integrity of event data through graph-based structures, and enables richer, more realistic process models. Traditional methods focus on linear event logs and isolated case behaviors, while OC-PM offers a broader, multidimensional view of complex business processes.

Figure 3 provides a visual summary of the transition from traditional, case-centric process mining to advanced, object-centric approaches enhanced by Artificial Intelligence. It contrasts the limitations of classical models—such as linearity, single-case focus, and poor relationship representation—with the capabilities of OC-PM, which introduces multi-object tracking, richer interdependencies, and AI-powered analysis. This evolution reflects the core argument of the paper: embracing object-centric and AI-driven methods is essential for capturing the true complexity of modern business processes.

RQ4: Which algorithms and approaches are most commonly employed in object-centric process mining to solve operational issues of traditional process mining?

To answer research question RQ4 regarding the most commonly employed algorithms and approaches in OC-PM, [9] presents a comprehensive framework that includes techniques for exploring and filtering object-centric event logs, automatically discovering workflow representations, and enhancing them with meaningful annotations. It also introduces conformance checking metrics to assess alignment between models and logs. Additionally, the article presents the OC-PM tool, available as both a web application and a ProM plugin, which supports uploading, filtering, and analyzing object-centric data. These approaches aim to make OCPM more accessible and effective for analyzing complex business processes, especially in real-world systems like SAP ERP.

The research work in [13] presents the eXtensible Object-Centric (XOC) [14] event log method. This approach captures events at a higher, more intuitive level (e.g., “create order”) and links them to evolving database states using object models. By removing the need for a predefined case notion, it supports complex relationships like one-to-many and many-to-many. The authors also introduce a ProM [15] plugin that automates the generation of these logs, enhancing both the clarity and richness of process data.

HOEG (Heterogeneous Object Event Graph encoding), introduced in [16], is a predictive approach in object-centric process mining that models events and objects as nodes in a heterogeneous graph. By leveraging Graph Neural Networks, it preserves detailed object-level information and captures complex interactions, resulting in improved predictive performance on real-world OCEL datasets.

The article [1] proposes an unsupervised anomaly detection approach for object-centric process mining using a Graph Convolutional Autoencoder (GCNAE) [17]. The method models event dependencies as attributed graphs and applies a two-layer Graph Convolutional Network (GCN) as the encoder to learn latent node representations, followed by a GCN-based decoder that reconstructs the original node features. The anomaly score is calculated by measuring the mean squared error (MSE) between original and reconstructed features, with higher errors indicating anomalies. To automate anomaly labeling, the approach uses an inter-quartile range (IQR) heuristic to set the threshold (τ = Q₃ + k · IQR). This algorithm enables effective detection of abnormal behavior in object-centric event logs without requiring labeled training data.

Table 5. The relevance of using LLMs and Object-Centric Process Mining.

Article ID	Solved Issues	Description
[11]	Enhancing prediction, diagnosis, and understanding in OCPM	Integrating Generative AI with OC-PM enables advanced predictive capabilities, actionable insights, real-time diagnostics, and nuanced understanding of complex object interactions. It improves model generation and empowers organizations to anticipate issues, recommend solutions, and respond swiftly, making operational management more proactive, holistic, and data-driven.
[18]	Robust performance of LLMs	Integration of LLMs like GPT-4 and Google’s Bard into the OC-PM ecosystem goes beyond traditional descriptive analysis, enabling a contextual, explanatory, and predictive understanding of processes—thereby meeting the demands of intelligent and complex industrial environments.
[19]	Advanced anomaly detection	LLMs provide an intelligent layer that complements OC-PM by offering flexible, model-agnostic anomaly detection in highly interconnected and dynamic environments, such as those found in IoT-enabled systems and smart factories.
[20]	Contextual Understanding And Adaptive Responses	The integration of LLMs into operational processes enhances querying, supports data abstraction, offers contextual insights, fosters adaptive learning, and improves the efficiency of problem-solving, all while posing new challenges that must be managed effectively.

provides a concise overview of recent advancements in the integration of Artificial Intelligence in Object-Centric Process Mining (OC-PM), particularly through the Large Language Models (LLMs). Each entry highlights a specific issue traditionally faced by OC-PM and the corresponding contribution of AI-driven approaches to address it. The table illustrates how LLMs enhance process understanding, enable intelligent and model-agnostic anomaly detection in complex systems, and improve prediction, diagnosis, and decision-making through real-time analytics and holistic analysis of object interactions. This demonstrates the growing synergy between AI and OC-PM in solving operational challenges in traditional process mining.

5. Discussion

This discussion highlights how Object-Centric Process Mining (OC-PM) overcomes the key challenges of classical process mining by enabling multi-object analysis, capturing complex interactions, and supporting realistic simulations. It also shows how integrating advanced algorithms such as Large Language Models (LLMs) enhances OC-PM with predictive capabilities, real-time diagnostics, and intelligent anomaly detection, making it a robust framework for analyzing and optimizing complex, dynamic processes.

The analysis clearly demonstrates that Object-Centric Process Mining (OC-PM) emerges as a robust response to the fundamental constraints of standard, case-centric process mining. Where traditional PM is confined to linear, single-case views and struggle to model real-world processes involving multiple, interacting entities, OC-PM introduces a paradigm shift. By enabling many-to-many relationships between events and objects, and by structuring data around multiple interacting case notions, OCPM captures the true interconnectedness of complex systems. It preserves data richness through graph-based representations, enhances interpretability across multiple dimensions, and facilitates more realistic simulations, predictive insights, and AI integration. As such, OC-PM does not merely extend traditional methods—it redefines process mining for modern, data-rich environments.

In addressing RQ4, the reviewed studies highlight how advanced algorithms—particularly those involving Large Language Models (LLMs) and Generative AI—are revolutionizing Object-Centric Process Mining (OC-PM). Traditional techniques often fall short when faced with the complexity and variability of modern, multi-object environments. However, by embedding LLMs within the OCPM ecosystem, researchers have achieved robust predictive performance, contextual analysis, and anomaly detection that go far beyond static, descriptive process mining. These intelligent systems not only enhance the interpretability and flexibility of OC-PM models but also support real-time diagnostics and proactive decision-making in dynamic settings such as IoT-enabled infrastructures and smart factories. Ultimately, these innovations solidify OC-PM as a powerful, forward-looking framework for analyzing, predicting, and optimizing complex processes, closing the gap between static frameworks and adaptive, intelligent operational systems.

6. Conclusions

The article reviews key studies on bridging standard Process Mining and Object-Centric Process Mining (OC-PM), highlighting classical process mining’s limitations in handling complex object interactions and presenting OC-PM as a promising solution. It also notes the potential of integrating Large Language Models (LLMs) to improve data understanding and automation in OC-PM.

OC-PM is still a relatively new and underexplored field, with few dedicated publications, revealing a significant research gap. Future research is expected to focus on using LLMs to tackle challenges such as log extraction, semantic enrichment, and generating explainable insights.

A promising research direction is comparative studies applying both traditional process mining and OC-PM on the same dataset. This would evaluate their strengths and limitations regarding accuracy, scalability, flexibility, and handling complex, real-world process dynamics, providing evidence of OC-PM’s added value in multi-entity scenarios.