Hacker News Reader

Comment by fnordpiglet | Hacker News Reader

I’m struggling to see why this is better than OWL/RDF and SparQL. Given the breadth and depth of semantic web knowledge in the corpus they do an amazing job discovering well established ontologies and building new ones on presented data and querying it with sparql. In fact better than almost anything I’ve seen. What an irony that the semantic web after all these years is likely to find a real use - it’s original one in fact.

View on HN

Maturity of the storage and query engines:

SQL > Cypher > SPARQL

You can do RDF/SHACL/OWL with LPG/Cypher based databases.

Ontolite (sqlite) and ontobug (ladybug) are examples.

Few thoughts:

- you need some engine to support your owl/rdf/sparql, and all available are lower quality/perf/features than pg

- its easier for LLMs to talk standard SQL compared so some version of sparql

There are plenty of graph databases, and RDF is expressed in triple stores. OWL is just for validation.

I’ve used AWS Neptune as a pretty reasonable managed sparql query engine but for most my needs actually just local jsonld mastered in git is plenty for almost APO knowledge graphs I have.

I’ve found LLMs are exceptionally proficient at authoring sparql, and quite clever. This isn’t surprising given there’s plenty of sparql in the corpus, and it’s semantics aren’t HUGELY different than SQL. Likewise they do fine writing datalog etc.

> I’ve used AWS Neptune as a pretty reasonable managed sparql query engi

it is obviously not OSS like postgres.

I am new to this field and our org struggle with authoring and documenting existing implicitly given ontologies. It is like working with archeology, and proposed automated solutions like nvidia’s seem alluring in that regard. What would be your suggestion for a way to extract the ontology/RDF/OWL/graph from existing underdocumented fields?

I found LLMs are pretty good at extracting ontologies through a) researching ontologies that exist for adjacent fields, there are a lot published after decades of work, b) doing multi pass ontological classifications using existing ontologies as baselines and refining, c) given ontologies attempting to represent knowledge in them and querying them along various paths with agents tasked with building recipes and techniques and evaluating them across synthetic benchmarks developed along side.

There’s also a lot of books on how to develop ontologies as well as online resources. They are all in the corpus and tasking a frontier pro level model with building a process and mental model document with references inline for grounding works really well as well for priming.

Aren’t these data formats and query engines? You still need something to extract and maintain the ontology no?

Yes, but my question wasn’t the utility of automated ontology building but the fact this seems to use pretty naive ontological and semantic graph tooling, making it less practically useful than if it used the industry standards of 30 years.

Reply on news.ycombinator.com