Python code to replicate the Twitter Birdwatch Note Ranking algorithm

twitter_birdwatch_note_ranking.py

#Credit: https://twitter.github.io/birdwatch/about/ranking-notes/
import pandas as pd

notes = pd.read_csv("notes-00000.tsv", sep="\t")
ratings = pd.read_csv("ratings-00000.tsv", sep="\t")

ratingsWithNotes = notes.set_index("noteId").join(
    ratings.set_index("noteId"), lsuffix="\_note", rsuffix="\_rating", how="inner"
)
ratingsWithNotes["numRatings"] = 1


def getCurrentlyRatedHelpfulNotesForTweet(
    tweetId,
    noteScoreSmoothingParameter=5,
    maxCurrentlyRatedHelpfulNotes=3,
    minRatingsNeeded=3,
    minSmoothedHelpfulnessScoreNeeded=0.5,
):
    ratingsWithNotesForTweet = ratingsWithNotes[ratingsWithNotes["tweetId"] == tweetId]
    scoredNotes = ratingsWithNotesForTweet.groupby("noteId").sum()
    scoredNotes["smoothedHelpfulnessScore"] = scoredNotes["helpful"] / (
        scoredNotes["numRatings"] + noteScoreSmoothingParameter
    )
    filteredNotes = scoredNotes[
        (scoredNotes["numRatings"] >= minRatingsNeeded)
        & (scoredNotes["smoothedHelpfulnessScore"] >= minSmoothedHelpfulnessScoreNeeded)
    ]
    return filteredNotes.sort_values(by="smoothedHelpfulnessScore", ascending=False)[
        :maxCurrentlyRatedHelpfulNotes
    ]