Over the past two weeks, I worked through a collection Stardog tutorials and training videos - learning the vocabulary, high-level uses, and interacting with the Stardog UI. Plus, I learned the basics of Turtle and SPARQL language syntax going through examples and testing things. All great stuff. I feel comfortable with the main themes of knowledge graphs and the Stardog UI.
Collectively, the Stardog Tutorials seem to have a common hurdle: the data is already there. Let's say that I wanted to make a knowledge graph of US states from Wikipedia data - there's a lot more involved than installing the knowledge kit and playing around!
From what I gathered, there appears to be two ways to build a knowledge graph: (1) manually (e.g., creating the data, loading the data directly or via virtualization, defining classes and properties, imposing constraints, etc.) or (2) programmatically (e.g., creating data by scraping text with NLP models, converting extracted data for subject-predict-object syntax, creating object properties programmatically (I'm really not sure how people do this, GNNs?) and uploading it to a knowledge graph). How both of those processes in the real world seem opaque to me.
Which leads me to this request: what does a knowledge graph process look like? Also, I feel like seeing a functional, purpose-built knowledge graph - where it comes from, the gist of how it was built, and how it is being maintained - would provide a lot clarity on what is possible in the modern world with knowledge graphs.