Understanding the inference output

[LaurentVeyssier]

Hi, many thanks for this suberb living tool and research!

I would like to have your thoughts which could help me understand the output of aspect extraction inference. I am using a model I trained on a custom dataset (v2.x). I noticed during inference a few sentences not displaying any aspect nor sentiment (although there is one which should be detected). However examining the atepc result output, the aspect is properly extracted, a sentiment predicted and a confidence level. The aspect/sentiment is just not showing up unlike the other samples submitted during batch inference.

any idea why this could happen? Additionally, is the confidence level related to the aspect extraction or the sentiment class and does it mean (ie how to use it)?

many thanks in advance for your reply