bad result: "man was shot several times"= 99.6% POSITIVE sentiment

[ldmtwo]

I've been debugging bad output from the pretrained model. What corpus is this trained on? How to make it not overfit.

import flair
# !pip install GIT+https://github.com/flairNLP/flair.git

flair_sentiment = flair.models.TextClassifier.load('en-sentiment')

def sentiment(text):
    s = flair.data.Sentence(text)
    flair_sentiment.predict(s)
    ret = {obj.value:obj.score for obj in s.labels}
    return ret

# sentence = 'Texas good samaritan shot several times while saving woman from suspect' 
# sentence = 'Art ran really well, but died.'
sentence = 'man was shot several times'
{'POSITIVE': 0.9961583018302917}

sentiment('I went to Palestine')
{'NEGATIVE': 0.9425508379936218}

sentiment('Black people in the South')
{'NEGATIVE': 0.8325328826904297}

[HallerPatrick]

Hey, I ran you code and getting a different result here:

man was shot several times
{'NEGATIVE': 0.9908545017242432}
I went to Palestine
{'NEGATIVE': 0.9425508379936218}
Black people in the South
{'NEGATIVE': 0.8325329422950745}

Looks like man was shot several times gives quite the opposite result to yours. In your example it is also not passed to the sentiment function.

[stale[bot]]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.