A Semantic Encoding of Object Centric Event Data

The Object-Centric Event Data (OCED) is a novel meta-model aimed at providing a common ground for process data records centered around events and objects. One of its objectives is to foster interoperability and process information exchange. In this context, the integration of data from different providers, the combination of multiple processes, and the enhancement of knowledge inference are novel challenges. Semantic Web technologies can enable the creation of a machine-readable OCED description enriched through ontology-based relationships and entity categorization. In this paper, we introduce an approach built upon Semantic Web technologies for the realization of semantic-enhanced OCED, with the aim to strengthen process data reasoning, interconnect information sources, and boost expressiveness.

💡 Research Summary

The paper addresses the emerging need for a unified, machine‑readable representation of object‑centric event data (OCED) in the field of process mining. Traditional process mining relies on activity‑centric logs such as XES, which model processes as linear sequences of events tied to a single case identifier. This approach struggles to capture the rich interactions among multiple business objects that occur in real‑world processes. The IEEE Task Force on Process Mining has proposed the OCED meta‑model to overcome these limitations by treating events and objects as first‑class citizens. However, the OCED specification itself remains at a meta‑model level, lacking concrete semantics, domain‑specific constraints, and mechanisms for linking disparate data sources.

To fill this gap, the authors propose a three‑layer semantic encoding built on Semantic Web standards (RDF, RDF Schema, OWL). The layers are:

1. Meta‑Level (OCED Ontology, OCEDO) – The core OCED concepts (Event, Object, Attribute, Relation) are modeled as OWL classes and object properties. Two namespaces are introduced: oced: for core entities and aux: for auxiliary constructs such as event types, object types, and qualifier relations. For example, oced:Event is an owl:Class and oced:observes is an owl:ObjectProperty linking an event to the objects it observes. This layer provides a formal, machine‑interpretable description of the OCED meta‑model.

2. Intensional Level (OCED Domain‑Specific Extension, OCEDD) – Domain experts can extend the core ontology with sub‑classes and sub‑properties that capture business‑specific semantics. The paper demonstrates this with the BPIC 2013 incident‑management dataset, defining classes such as ext:Incident, ext:Product, and ext:TeamMember, and properties like ext:has_impact, ext:is_part_of, and ext:works_in. These extensions allow the encoding of constraints (e.g., every incident must be linked to a product) and enrich the graph with domain vocabulary without altering the core OCED ontology.

3. Extensional Level (OCED Resources, OCEDR) – Actual event logs are transformed into RDF triples, forming a knowledge graph that instantiates the ontologies. The authors built an automated pipeline that parses XES logs, extracts events, objects, and their attributes, and maps them to RDF resources using standard XML‑Schema datatypes (e.g., xsd:dateTime for timestamps). The resulting graph can be published via a SPARQL endpoint, enabling queries, reasoning, and integration with other linked data sources.

The implementation is provided as an open‑source prototype (GitHub repository) and is demonstrated by converting a real‑world XES log into an OCEDR knowledge graph. The authors argue that this approach yields several benefits:

• Interoperability – RDF/OWL are widely adopted standards; the same graph can be linked to other ontologies (e.g., product catalogs, customer relationship data) using shared URIs, facilitating cross‑domain analytics.
• Reasoning and Validation – OWL DL reasoning can automatically detect inconsistencies (e.g., an incident lacking a product link) and infer implicit relationships, supporting more sophisticated process mining tasks such as conformance checking with business rules.
• Expressiveness – By explicitly modeling many‑to‑many object relations, qualifiers, and timestamps, the representation captures complex object‑centric behaviors that activity‑centric logs cannot express.
• Extensibility – New domains can be accommodated by adding domain‑specific extensions without modifying the core ontology, promoting reuse and modularity.

The paper also acknowledges limitations. Designing domain extensions requires expertise in both the business domain and ontology engineering, which may increase upfront effort. Converting large‑scale logs into RDF can generate massive triple stores, raising concerns about storage, indexing, and query performance; the current evaluation is limited to a prototype scale. Finally, because the OCED meta‑model is still undergoing standardization, versioning and compatibility between different OCED‑based ontologies may become an issue.

In conclusion, the work demonstrates how Semantic Web technologies can provide a rigorous, layered representation of object‑centric process data, enabling richer analysis, better integration, and automated reasoning. The open‑source tools and the concrete BPIC case study illustrate practical feasibility, while future research directions include scalability testing, automated generation of domain extensions, and tighter alignment with the evolving OCED standard.