20-06-2024 — Standardized Developer-Friendly Data Modeling with LinkML
Abstract
In this small presentation I share with the LinkML Community how LinkML is currently being used to model data products at Alliander, and especially at Netbeheer Nederland, a national collaborative effort between Dutch grid operators.
16-05-2024 — Building CIM-based Data Products with LinkML: the Linked Data Modeling Language
Abstract
Alliander has been working on implementing a decentralized Data Mesh architecture. Data is produced bottom-up, empowering teams and removing centralization bottlenecks. To make sure we can manage and govern the data, teams need to treat their data as a product, providing metadata to describe things such as ownership, versioning, quality annotations and privacy and security classifications. Crucially, this also includes a logical data model that describes the structure of and constraints on the data.
Many of these so called data products could benefit from having their logical data models being based on the CIM, i.e. use the CIM as a vocabulary. In this talk we will look at several ways of doing that, introducing and tackling challenges as we encounter them.
Since the normative CIM model is developed in Sparx EA, many CIM users use EA for information modeling. Despite the support of powerful features such as profiling and schema generation, EA has serious limitations, most notably: (i) Linked Data support is minimal; (ii) interoperability with other software is non- trivial and cumbersome; and (iii) the model is not represented as human-readable plain text, causing users to be unable to use text editors and other text manipulation tools to maintain it.
Next we will look at the Semantic Web technology stack and see that it solves the aforementioned problems with EA: Linked Data is part of its philosophy, it is plain text and also well-standardized. Furthermore, the technology is very flexible and mature. However, it comes with challenges of its own, most notably that the technology is notoriously abstract, complex and inaccessible to both developers and information architects/modelers.
Finally, we will look at LinkML, a modeling language that is both human and machine readable, designed to be accessible to write or generate, and expressive enough to support Linked Data and both conceptual and logical data modeling. Out of the box it ships with a bunch of generators to create physical data models, documentation and visualization. Moreover, since it is open-source it can be extended easily.
Will LinkML be our pragmatic savior?
15-03-2024 — Entity and Relation Extraction using OntoGPT and SPIRES
Abstract
At the Alliander Research Center for Digital Technologies I did a proof of concept with OntoGPT, a library which helps with extracting entities and relations from unstructured text using a LLM and predefined schema.