I have a complex taxonomy with different level of granularities.

Let’s say I have 4 levels:

• level 1: being the most general
• level 4: being the most granular

This taxonomy has a text label and also an integer code.

For example for:

• level 1: ‘Continent’ code is 01
• level 2: ‘Country’ codes are 01XX
• level 3: ‘Region’ codes are 01XXYY
• level 4: ‘City’ codes are 01XXYYZZ

If I want to classify some texts at the level 3, shall I ask the LLM to return the ‘Region’ as a name or as a code?
My guess is that as LLM have been trained on text data, I am better off to ask him to return a text label and after I could use a lookup table to retrieve the code.

Any thoughts on this?