wikilinks / conll03_nel_eval

Python evaluation scripts for AIDA-formatted CoNLL data
Apache License 2.0
20 stars 4 forks source link

Mappings to Freebase #35

Open jnothman opened 10 years ago

jnothman commented 10 years ago

We should be able to build mappings to Freebase IDs. The original data includes this information, which is more robust to e.g. a title becoming a disambiguation page.

hoffart commented 10 years ago

Is there any description of how Freebase solves the problem of page moves and keeps its IDs consistent?

jnothman commented 10 years ago

I haven't seen one. But I believe Wikipedia ids (not titles) move with page moves, and certainly sufficient information is available by processing the revision history (which Freebase has the resources to do). And Freebase does indicate timestamps when relating one of their entities to a Wikipedia entry.

FB may or may not effectively handle trickier cases such as a person's page being dissolved into their organisation or product page, after editors decide the person is not especially notable. Since this looks identical to a merging of entities, will Freebase merge their ids, or maintain an old ID that no longer has a Wikipedia correspondence?

In any case, I've not seen it documented.