Datasets:

fancyzhx

/

yelp_polarity

like 14

The original paper seems to contain a typo, referring to stars 1/2 for negative and 3/4 for positive.

I checked the data like so:

import datasets

ds_yp = datasets.load_dataset("yelp_polarity", split="test")
ds_yf = datasets.load_dataset("yelp_review_full", split="test")

df_yp = ds_yp.to_pandas().set_index("text").rename(columns={"label": "Sentiment"})
df_yf = ds_yf.to_pandas().set_index("text").rename(columns={"label": "Stars"})

df = df_yp.join(df_yf).dropna().sort_values(["Sentiment", "Stars"])
uniques = df.groupby(["Sentiment", "Stars"]).groups.keys()

print(uniques)

And can see that what they meant to say is that
1/2 stars = negative
3 stars = excluded
4/5 stars = positive.

In the data it's all zero-indexed so the raw values are 1 lower.

Hi @Davidg707 , thanks a lot for the investigation and the fix.

Definitely, it seems the authors meant to say something different from what they finally said...

I'm adding here the output of your investigation for completeness:

dict_keys([(0, 0.0), (0, 1.0), (1, 3.0), (1, 4.0)])