Speaker
Description
In modern industrial environments, understanding how data flows through complex ETL pipelines is critical for traceability, auditing, and compliance. While traditional lineage tracking tools rely on static metadata or log-based introspection, they often lack semantic expressiveness and offer limited support for automated validation.
This work presents a semantic-aware framework for ETL data lineage tracking, integrating property graph modeling (Neo4j), lightweight ontologies (RDF/OWL), and SPARQL querying. Each transformation step is modeled as a semantic entity enriched with metadata such as inputs, outputs, timestamps, and execution order. The resulting lineage graph is exported to RDF, enabling validation via SHACL and pattern-based queries over transformation chains.
The proposed solution was implemented and tested on the AdventureWorks DW2022 dataset, demonstrating low-latency querying, structural correctness, and enhanced traceability. Comparative analysis shows that, unlike traditional tools such as Apache Atlas or OpenLineage, our approach supports fine-grained reasoning, constraint checking, and semantic completeness verification.
This contribution offers a lightweight and reproducible solution using only open-source components and is particularly suited for evolving, compliance-heavy industrial data environments. Future work includes extending the ontology, integrating streaming ETL sources, and deploying the model over scalable RDF triple stores.
