How is concept_aliases.json constructed?

How concept_aliases.json is constructed? https://github.com/allenai/scispacy/search?q=“concept_aliases”&unscoped_q=“concept_aliases”

By using this table, I think, if mention is included in this table, mention-entity prior(=similarity) is fixed to 1.0. (If my understanding is incorrect, I’d appreciate if you’d write.)

Currently I’m curious about, how this concept_aliases.json is constructed? It seems that this is actually list, whose length is 4426690, but I can’t understand how this list was constructed. Also, I supposed that alias table is something like,

{'entitiy1':[alias1_1,alias1_2,alias1_3], 'entity2':[alias2_1,alias2_2,...],...}

but concept_aliases.json is list. I’d like to know what this list means.

Still I’m currently reading codes and maybe I missed some codes, which is related to constructing this table.

[DeNeutoy]

Basically, the problem is that if you have some canonical entities like:

"Ford County, Texas": ["Ford County", "Ford County, Texas"]
"Ford County, Alabama": ["Ford County", "Ford County, Alabama"]

etc

And you encounter the string “Ford County” in some text, there is no way to know which canonical entity (which all share the same alias which is an exact match) you should link to. So we default to returning all of them. Obviously this is the point at which Entity Linking becomes difficult and actually interesting, but the good thing is that this doesn’t happen hugely often 😄

[DeNeutoy]

Closing, please open more issues if you need any clarifications, or you want to collaborate or something!