Feature Request: Organize many classification labels by hierarchy

I have a big data set of questions so i'd like to first get them all classified into top labels and then do a re run and classify them into smaller units, how much re runs will depend on complexity of questions, so in the end i would have an aggregated label like Universe_MilkyWay_Earth_France_Paris

Does it make sense to have hierarchical labels?

It is your task problem, not doccano issue.

So isn't a software solution to resolve problems? I'm not understanding. I've made a question why are you closing it? If you have many labels the ux will be horrible and loose it's easiness. I'm trying to open a debate why can't you allow that? So you're saying that a big data set with many labels is not a doccano issue? It's just for a few labels?

You clearly didn't understand my question.

Hello @tattoodobem. Thank you for your question.

I created a sample hierarchical labels. Does this fit your thought?

Also, I have a question. Do you assume that the task is text classification or sequence labeling?

@tattoodobem Sorry for my inadequate correspondence. I changed the tag from question to feature request. Thank you for reporting the needs of annotation!

Hello @tattoodobem. Thank you for your question.

I created a sample hierarchical labels. Does this fit your thought?

Also, I have a question. Do you assume that the task is text classification or sequence labeling?

text classification. with many topics

Yes something like that. But it needs to support more levels, needs to be easy to use and quick if you have many options. What about something like this? https://vue-treeselect.js.org/ (doccano uses vue right?) You can search for the label. i was also thinking that it would be nice to be able to insert new labels when you are classifying.

I think one problem you have here, if you have many topics, is that it's not easy to immediately know how to label it unless you already know all topics and remember to create them all first. so you want/need help to do that job too. When i was doing a research i had lots of text but no specific idea of what was in them. that's why we need to classify, so i started by classifying in general topics, that gave me a better clue of what was in them and how to get them into smaller topics. So another solution could be using re-runs. So first run would have doccano labels as they are. You see top levels only so there shouldn't be too many ruining the use experience. Next run would let you see second level labels for the topic you chose before. so that again should not ruin the interface. And so on. What would be missing is: When is text finally classified? next button now "says" it is classified right? and the text wont appear again. Understand? this might be easier to do. Wouldn't change the interface. And is probably better for the user, he doesn't need to know the topics so well..

What do you think?

@tattoodobem Sorry for my inadequate correspondence. I changed the tag from question to feature request. Thank you for reporting the needs of annotation!

thanks

Hello @Hironsan

I am currently having a Sequence Labelling project and I'd be interested to add hierarchical labels to my project. I was wondering if there's a way I can add the sample you showed before to my project ?

Thank you

Related: #877.

Hello @tattoodobem. Thank you for your question.

I created a sample hierarchical labels. Does this fit your thought?

Also, I have a question. Do you assume that the task is text classification or sequence labeling?

How to create hierarchical labels?