✨ Transforms Filters: 🧲 Magnetic labels

🧲 Magnetic labels

 
 

Why magnetic labels

You have already coded your dataset, manually or using AI, and now you want to relabel.
Suppose you already know what labels you want to use, perhaps:
  • you knew before you even started
  • you decided what labels you wanted after reviewing your data and looking at different auto-cluster solutions
Magnetic labels are a really simple solution for these cases.

How to use them

 
notion image
You simply type the list of magnetic labels you want and decide on the power of the magnets (”magnetism”).
Magnetic labels attract existing labels of similar meaning, essentially relabelling these old labels with the new magnetic label. If an existing label is similar to two or more different labels, it is relabelled with the magnetic label it is most similar to.
If you use low magnetism, the magnets are weak and only attract existing labels which are very similar to them.
Increasing the magnetism means that more and more existing labels are attracted to the magnetic labels.
Existing labels which are not attracted to any label are unchanged. This means that you can easily see if your magnetic labels cover most of the original content.
Best practice is then, after applying magnetic labels, to then auto-cluster the links in order to pick out important themes which are not covered by the magnetic labels.
If you want you can even include hierarchical magnetic labels like Health behaviour; hand washing.

Storing your labels

Your magnetic labels are included if you save a bookmark aka saved View.
You can also store your preferred label set in the Codebook in
📂
The Files tab
.
notion image

Use cases

  • Drop in magnetic labels which contain the text from the “official” theory of change.
  • See how much the existing labels get attracted to the magnetic labels, and what material is left over.
  • Conduct “radical zero shot” auto coding with no codebook at all,
    • let the AI decide the best label for each case
    • do some auto clustering until you get a feel for the labels you really want in your story
    • type the labels into the magnetic clusters box

Tips

What if you have a semantically ambiguous label?

An example: you have some research about animals and you want to look for mentions of the organisation Animal Aid. If you use Animal Aid as a label, it might also pick up any mention of helping animals which have nothing to do with the organisation itself.
One way to get round this is to use
🔗 The Manage Links tab
to permanently recode any mention of Animal Aid in your factor labels into something unambiguous like, say, The Archibald Organisation. Choose a meaningless name which is not going to appear in or be related to to the rest of your material.
When doing this “hard recoding” remember to recode AnimalAid, Animals Aid etc as well.

Attracting unwanted material away from your map

You can add an factor into magnetic clusters even if it doesn't appear in the final map.
For example you might have a lot of material about blood donors and you don’t want material about donating clothes. As well as donating blood you might add the labels donating goods and donating money. You can filter these out later, but they will help restrict donating blood to what you want.