Coding: Improve storage of original triples when running OWL-NETS

callahantiff / PheKnowLator

PheKnowLator: Heterogeneous Biomedical Knowledge Graphs and Benchmarks Constructed Under Alternative Semantic Models

https://github.com/callahantiff/PheKnowLator/wiki

Apache License 2.0

159 stars 29 forks source link

Coding: Improve storage of original triples when running OWL-NETS #56

Closed callahantiff closed 3 years ago

callahantiff commented 3 years ago

TASK

Task Type: CODEBASE

Improve the storage of removed OWL semantics when running OWL-NETS version of the build. Remove dictionary constraint for something network-based.

TODO

[ ] Come up with a better solution to handle triples removed from the full KG when creating OWL-NETS
- Information that is compressed when converting a class from OWL to OWL-NETS
- Information that is purposefully ignored or deleted (i.e. specific triples, disjointness)
[ ] Determine if we want to update OWL-NETS to follow any of the OWL to RDF specifications mentioned in the OWL2Vec* paper:
- W3C Defined Graph Mapping (here)
- Projection rule-based approach (here)

@bill-baumgartner - can we talk about the other OWL transformations soon? From what I can tell, OWL-NETS is a more extreme version of the transformations described above.

callahantiff commented 3 years ago

@LEHunter - Would love your opinion on the property graph created by OWL2Vec* and what we do with OWL-NETS (V2 described here). Seems like we are creating a more extreme version. Perhaps we should call OWL-NETS something other than a property graph?

I'm tempted to import their method to derive their property graph version so we can compare it to OWL-NETS and the full graph builds. Since their code is in Python too, it should not be too difficult. Something to think about. 😄

LEHunter commented 3 years ago

I'm not sure I understand your question. A "property graph" as far as I can tell is just a graph with typed and directed edges.

I totally agree about including the OWL2VEC* method in the graph builds for comparison purposes. I'd be surprised if paper reviewers didn't insist on it anyway.

callahantiff commented 3 years ago

I'm not sure I understand your question. A "property graph" as far as I can tell is just a graph with typed and directed edges.

Thanks @LEHunter! I think I was focused on the wrong thing. I agree that we are good on this.

I totally agree about including the OWL2VEC* method in the graph builds for comparison purposes. I'd be surprised if paper reviewers didn't insist on it anyway.

Sounds good. Depending on how easy/hard it is to import their code (it's not the easiest to read), I may not formally include it as a build parameter for the next build. That said, we will plan to include it for the paper.

callahantiff commented 3 years ago

After speaking with @bill-baumgartner - we will be holding off on making further updates to OWL-NETS, especially since the method is functioning just fine. Once the initial paper is complete this issue will be reconsidered.

callahantiff commented 3 years ago

Closing this issue for now, have a temporary workaround. Solution is described on the owl-nets wiki.