inception-project / inception

INCEpTION provides a semantic annotation platform offering intelligent annotation assistance and knowledge management.
https://inception-project.github.io
Apache License 2.0
589 stars 148 forks source link

Accept only recomendations from the current view #2066

Open david-waterworth opened 3 years ago

david-waterworth commented 3 years ago

I currently have documents with a large number of sentences (i.e. hundreds - thousands), I generally set the view to show ~10. I'm not sure this is best practice and I may split them somehow. However, feedback from potential annotators is they prefer it this way since they can easily go back and forward page by page.

I find the string matching recommender works pretty well, but I prefer not to click Accept All since it applies across the entire document so I have no idea what it's going to accept.

Is there any way of accepting only the visible recommendations (i.e. only from the sentences currently shown in screen)? If not could it be added? Not only does accept all sometimes create more work, it also teaches the recommender "bad habits"

reckart commented 3 years ago

It sounds like a good idea, yes.

reckart commented 3 years ago

Please make sure you are using the latest INCEpTION 0.18.3 as it fixes a bug that can lead to the loss of annotations that was observed to happen more often in larger documents in conjunction with recommenders.

As for splitting: I'd recommend to split large documents down into smaller documents resembling "units of work" (i.e. what can be usually done in optimally a single annotations session) - that way you can mark a them as "finished" more often and more easily keep track of progress in your project.

david-waterworth commented 3 years ago

I've followed your suggestion and it works really well. I did run into an issue where I named by files based on a field which I used to denote the "units of work". Problem is some of the files contains characters which aren't valid in URI's (i.e. >) and failed to import. So I URI encoded the files but that also failed with a similar error. For now I manually removed the invalid chars from a handful of files - I'll log a separate issue later.

A couple of related enhancements:

The ability to "Undo" after "Accept All" and/or the ability to bulk remove specific annotations would be nice. I find now most of my time is taken up deleting annotations that I accidentally accepted because I didn't notice that they recommended had incorrect recommendations.

reckart commented 3 years ago

Best open new issues for new ideas, otherwise the ideas get lost. E.g. I think there is no issue for "undo" yet.