Documents classification with variable set of labels - django

Question

I want to create a model where a user can define a set of labels for the purpose of classification of documents. Each label for that document should also have a probability (float) that that document belongs to that label.
So this is what I have for my model with 3 labels (news, business, sports):
class Document(models.Model):
document_name = models.CharField(max_length=100)
proba_label_news = models.DecimalField(max_digits=5)
proba_label_business = models.DecimalField(max_digits=5)
proba_label_sports = models.DecimalField(max_digits=5)
...
But of course there could be more labels what a user can define at the start of a classification project and what a developer can't foresee.
I tried manytomany as an alternative but when I create a Label model and define a new label as "news" than I can't have a different probability for each document label, right?
So what is the django way of solving this problem?

Answer 1

Yes, a many-to-many relationship could work. What you need to add is an intermediatiary 'through' model to store the probability for each document-label relationship. Something like this:
class Label(models.Model):
name = models.CharField(max_length=50)
class LabelProbability(models.Model):
label = models.ForeignKey(Label, on_delete=models.CASCADE)
document = models.ForeignKey("Document", on_delete=models.CASCADE)
probability = models.DecimalField(max_digits=5)
class Document(models.Model):
document_name = models.CharField(max_length=100)
labels = models.ManyToManyField(Label, through=LabelProbability)
Then when you want to add a label to a document, you would use the through_defaults argument to assign a probability. Do something like the following:
news = Label.objects.get(name="news")
document_1.labels.add(news, through_defaults={"probability":0.57})
You can read more about many-to-many fields, and using a through table in django docs here
An alternative approach
Alternatively, you could store the label and the probability together, something like this:
class Label(models.Model):
name = models.CharField(max_length=50)
probability = models.DecimalField(max_digits=5)
document = models.ForeignKey("Document", on_delete=models.CASCADE)
class Document(models.Model):
document_name = models.CharField(max_length=100)
This way, you just have two tables, and each label stores both the name of the label (allowing you to use any name), and the probability. You can add as many labels as you want pointing to the same document (just as before).