neuraly

/

bert-base-italian-cased-sentiment

like 8

I tried using your model for classifying multiple sencences:

• I encoded the sentences
• I did padding / truncating for having encoding of the same dimension (512, which is the maximum length for your model)
• Then I classified all the sentences

Since it was an attempt, I used small phrases such as "Amo tutti, vi voglio bene", or "Odio tutti, vi detesto".
I received as a result that the sentences were neutral (I think it's because of the padding). When I classify them one by one, I then get correct results.
Why do padding influences the results??