Dangling auto-links
dginev opened this issue · comments
If we enhance the NNexus term recognition with a statistic term extraction algorithm, such as the C-value approach, we should consider linking to the highest scoring phrases even if they are not yet in the NNexus database.
That would result in dangling auto-links, which authors could later fill in, a-la Wikipedia.
Dangling links would also need to be treated by the auto-linking and invalidation mechanisms as they are an interesting breed of an annotation. As the dangling links will produce, well - links, they shouldn't be a part of the invalidation mechanism, rather they should be treated as already defined concepts (the link/annotation has already been deposited).
But this also opens a question of how do we reliably figure out requests for deletions - if a page is no longer found to be indexed it could be because of temporary downtime, rather than the concept being removed. Should there be an explicit delete request? Maybe it makes sense that the way we have a "index this URL" request we should have a "purge this URL" request.
Still a feature I'd like to have, and one that a general model over scientific discourse that is trained on concept recognition would be able to automatically deliver.