This task is closely related to Task 3: add other LLMs to Harmony
Harmony often matches items that should not be matched, or vice versa, e.g.
Feeling tired or having little energy? was incorrectly matched to Having upsetting dreams that replay part of the experience or are clearly related to the experience?
We have a Github repo where we have experimented with this: https://github.com/harmonydata/matching
So how can we improve the matching?
- Probably the easiest way would be to switch this underlying model, rather than trying anything clever. See also Task 3: add other LLMs to Harmony
- Longer term an LLM could be fine-tuned but that needs lots of resources and lots of data!
You will notice that on the web UI there are options for users to tag a harmonisation as incorrect.
We store this data. I exported it on 18 April 2024. Maybe we can use this data to improve the tool?
All saved harmonisations
https://harmonyapistorage.z33.web.core.windows.net/harmonisations.json
All mismatches tagged by users in the web UI:
https://harmonyapistorage.z33.web.core.windows.net/mismatches.json
All user ratings:
https://harmonyapistorage.z33.web.core.windows.net/ratings.json