Bad predictions of the models about absa-pytorch HOT 1 CLOSED

I think you are doing the right thing, and you could refer to issue #25.
Here I try to give you two reasons why most models suffer from the problem you mentioned:
Firstly, you are testing models with sentences from the domain which they never met with in training dataset, i.e., in laptop or restaurant domain;
Secondly, models occasionally make wrong prediction in such samples as they don't consider the relationships between different aspects within one sentence.
From my perspective, above-mentioned two flaws are the directions we should concentrate in later works.

from absa-pytorch.