WIP: LLM: identify claim out of context for a keyword

[polomarcus]

Identify claim with a note from 1 to 5 for specific keywords using Scaleway LLM

Steps

1. extract as CSV from metabase all filtered rows (plaintext and id)
2. docker compose run llm
3. export CSV to metabase to analyse (join on keyword_id) using head -n 1 claims-0.csv > combined.csv && tail -n+2 -q *.csv >> combined.csv

Output example

{
 "claims": [
 {
 "claim": "c' est quand même très écologique tout ça",
 "note": 1,
 "sentiment": "positif",
 "theme": "écologie urbaine"
 },
 {
 "claim": "la ville de lyon et très écologiste est en fête des lumières",
 "note": 1,
 "sentiment": "positif",
 "theme": "politique environnementale"
 }
 ]
 }