Here you can find links to code that reproduces the note scoring/ranking code that Birdwatch runs in production.
If you download the notes, ratings, and note status history files made available on the Birdwatch Data Download page and put them in the same directory as the following files, you can then run python main.py to produce a scoredNotes.tsv file that contains note scores, statuses, and explanation tags that will match what’s running in production (as of the time the data was from).
- main.py
- algorithm.py
- constants.py
- explanation_tags.py
- helpfulness_scores.py
- matrix_factorization.py
- process_data
You can also see this code on Github.