Closed cthoyt closed 1 year ago
For my ears, Entity Alignment Matching does not make sense;
An alignment is a set of mappings, for example between two ontologies, or knowledge graphs. "entity matching" is what all this hierarchy is about. Can you provide a link to what you mean here? Do you mean: EmbeddingSimilarityBasedMatching?
Well you can do some arbitrary embedding method then compare the embeddings in embedding space with a metric, but KGEMs are a bit different. Here's a paper that describes a few methodswith KGEMs: https://www.dbs.ifi.lmu.de/~tresp/papers/978-3-030-45442-5_Chapter_1.pdf
I could be wrong, but the paper you suggest would result in semapv:GCNBasedMatching
. The fact that it is applied to knowledge graphs is not really relevant (same would apply to OntologyMatching
), and the entity
is redundant. The Matching activity should focus on the how
to provide maximum value for provenance.
I do see however your point: this is a controlled vocabulary, and your hope seems to be that all terms relevant to our domain should be captured there. For now I would cautiously reject this term (or leave it open until we know better), as we do not want people to give semapv:KnowledgeGraphEntityMatching as a justification (its not), but rather semapv:GCNBasedMatching
. I think!
I'm fine to demote this to an issue where we can have further discussion, thanks for the feedback
If I predict mappings using an entity alignment methodology (e.g., via a knowledge graph embedding model), I think it makes sense to have a specific term