Labeling tasks

Kern Refinery lets you set up multiple labeling tasks per project. Currently, we support these two options:

  • classification: The digital stamp on your record - be it for binary or multiclass classification.
  • extraction: Your digital text marker - for e.g. named entity recognition, sentence segmentation, or part-of-speech tagging.

🚧 Future task types

We're working tirelessly to add further types in the future, such as relationship labels or native multilabel classifications (which, at the moment, you'd need to build via a workaround of multiple binary classifications).

To add a labeling task, simply click on the "Add labeling task" button.

A modal will open up, which asks you for the attribute you want to label. If you don't care about the attribute selection, you can just go for full record. However, for e.g. extraction tasks, you need to choose the respective attribute. Also, you must specify a name for the labeling task. This is not the name of your later label. Instead, you can imagine this to be the name for a group of labels.

Creating labels

Once you created the task, you can now create labels by clicking on the "+" icon. The label color will automatically be picked in an alternating scheme, but you can also specify a color as you like by clicking on the pipette.

📘 Choosing keyboard shortcuts

You can now also assign keyboard shortcuts to your labels, such that you can label even faster!

Something special to note here is the link between extraction task labels and lookup lists. We've seen that it is highly valuable to collect entities that you label manually in a dedicated list. This has two main benefits:

  • maintenance of your span labeling becomes easier, as you can see all entities you've labeled manually.
  • this data can be used to build labeling functions easily via distant supervision.

We cover this more in the labeling functions section of this documentation.

Now it is time to start labeling!