Skip to content

Recipe: comments → pandas → sentiment

pip install ytscrape pandas vaderSentiment

import pandas as pd
from vaderSentiment.vaderSentiment import SentimentIntensityAnalyzer
from ytscrape import CommentSort, YouTube, to_dict

with YouTube() as yt:
    comments = list(yt.comments("dQw4w9WgXcQ", max_results=500, sort=CommentSort.TOP))

df = pd.DataFrame([to_dict(c) for c in comments])
analyzer = SentimentIntensityAnalyzer()
df["sentiment"] = df["text"].fillna("").map(lambda t: analyzer.polarity_scores(t)["compound"])
print(df[["author", "like_count", "sentiment"]].sort_values("like_count", ascending=False).head())
print("Mean sentiment:", df["sentiment"].mean())

import asyncio
import pandas as pd
from ytscrape import AsyncYouTube, to_dict

async def main() -> pd.DataFrame:
    async with AsyncYouTube() as yt:
        thread = await yt.comments("dQw4w9WgXcQ", max_results=500)
        return pd.DataFrame([to_dict(c) async for c in thread])

df = asyncio.run(main())

like_count and reply_count are integers, so you can sort and aggregate directly.