User feedback for class is not taken into account

[tomas-knap]

1. Run books-v1 with dbpedia Jan

2. Change the classification for series to something which does not relate to that topic, e.g. City as below. (so search for "City", select it)
   

3. Re-run

4. There should not be any disambiguation results, but there are !

[brodecva]

There should not be any disambiguation results, but there are !

Shouldn't? Not sure about that, maybe @josef-janousek might say something about this. And at least write down the basic rules that the annotations follow. This should become a part of the documentation.

Concerning the unwanted disambugations, the user has always the option to not disambiguate, so this might not matter that much.

But it might in case when the custom class is not taken into account where two equally good disambiguation results might be decided by the custom classification. Please try that, @josef-janousek as well. thx!

[tomas-knap]

What I mean is that user fixed a class to "city" (locked). Dissambiguations are unlocked. So that means that when the algorithm is re-executed, there should not be any disambiguation for the class "city", right?

[tomas-knap]

(we can also finetune this as part of master theses), if no time. But it seems to me that this was working before (1.0) correctly

[brodecva]

@tomas-knap As @josef-janousek explained this is behavior that is present from the first version. What happens is that when the classification does not produce any viable disambiguations, the algorithm attempts to ignore it in order to produce something. @josef-janousek will document it. And then he will close it.