Software Lifecycle Management for Knowledge Graphs Workshop

Knowledge graphs are digital artifacts with a complex construction process utilizing numerous tools and data sources. They are generated in elaborate pipelines utilizing a wide variety of semantic technologies, for example mapping languages, such as RML or OTTR, or validation languages, such as SHACL. Further semantic technologies are used to describes the used ontology, such as OWL, and the adjacent queries, such as SPARQL. Far from a linear process, multiple data sources must be mapped into the target knowledge graph.
All the involved artifacts, ontologies, mapping scripts, graph shapes, etc., are interdependent and changes in one of them require the adjustment in others. The building and maintenance of a knowledge graph needs to apply the artifacts and tools in the correct order in the right context, e.g., staging and production contexts, as well as manage the intermediate artifacts generated in substeps.
In current practice, managing the dependencies is a manual process and general management of artifacts and changes is done using ad hoc approaches. Despite the numerous work on knowledge graph construction, there is a focus on the technical aspects of the single steps and little attention has been paid to the practical aspects of (a) organizing and managing knowledge graphs projects in terms of change management, dependencies between semantic artifacts, as well as DevOps for knowledge graphs, and (b) automating building and deploying of the resulting knowledge graph and adjacent artifacts. Similarly, connections to project management in software engineering, where a rich body of experience in DevOps, building, maintaining and deploying of digital artifacts exists are not systematically explored.
The Software Lifecicly Management for KG workshop (SofLiM4KG) aims to collect experiences in successful and abandoned knowledge graph projects from this perspective to (a) carve out the specifics in knowledge graph engineering that pose challenges beyond software engineering practices, (b) to establish best practices and anti-patterns for the community, and (c) build the foundations for the systematic investigation of the connection to software engineering, as well as qualitative and quantitative studies in project management of knowledge graphs.

While Large Language Models have transformed how we process unstructured text, creating structured knowledge graphs from this data remains a significant challenge. This keynote presents an automated pipeline for converting unstructured medical transcripts into queryable knowledge graphs optimized for retrieval-augmented generation (RAG). We demonstrate a Python-based process that handles the complete workflow: from initial text processing using medical-domain LLMs, through automated schema generation incorporating standard medical ontologies, to final graph construction and validation. The presentation includes practical implementations, common pitfalls, and solutions for scaling these systems in production environments. Our approach reduces manual intervention while maintaining medical accuracy and answering temporal questions in a scalable way, providing a blueprint for similar efforts in other domains requiring structured knowledge representation from unstructured sources.

About the speaker

Tom is the Technical Founder of WhyHowAI. He holds a PhD in Computer Science, specializing in Natural Language Processing and Knowledge Graph Embeddings. He has more than 8 years of experience as Machine Learning Engineer. He is most interested in decision making under uncertainty.