Overaggressive color filtering on coverguess tags

TalkTalk about LibraryThing

Join LibraryThing to post.

Overaggressive color filtering on coverguess tags

1lorax
Jul 18, 9:12 am

Pulling this out of the thread on the introduction of the feature in the hopes it gets seen.

There's some sort of automated logic to filter out colors from being shown as coverguess tags - this is intended to reduce duplication between them and the automatically extracted color tags, even though they aren't always from the same cover and the extraction doesn't always label colors the way humans would. While it's a reasonable idea in principle in practice it's a bit over-aggressive; there are plenty of reasonable nouns for things appearing on colors that are filtered out by this process. Examples I've encountered:

sand
snow
coral
lemon
orange (referring to the fruit)

I'm sure there are others. Any possibility that someone could take a peek at the list of words and remove those which are obvious English nouns, or is that all part of the Syndetics black box?

2knerd.knitter
Jul 18, 9:20 am

>1 lorax: This was done intentionally, using all the color names, but I would agree that excluding some of those does not make sense. I'm not sure we're going to get rid of orange, so it might have to be tagged "orange (fruit)"; but I would prefer to have it only exclude the main color names: red, beige, yellow, green, gray, white, orange, purple, pink, blue, and brown.

3lorax
Jul 18, 9:47 am

I'd be more than happy with that compromise!

4knerd.knitter
Jul 18, 10:38 am

Updated to only exclude red, beige, yellow, green, gray, grey, white, orange, purple, pink, blue, and brown

5lorax
Jul 18, 1:45 pm

Thank you! I'll go back and retag those winter and beach scenes.