brunneis / ilab-erisk-2020

Repository accompanying the CLEF 2020 eRisk Workshop Working Notes for the iLab team (University Of Strathclyde)

ilab-erisk-2020

Repository accompanying the CLEF 2020 eRisk Workshop Working Notes for the iLab team (University Of Strathclyde)

Work in progress...

ilab-erisk-2020's Issues

I have a question about the Train and Test method

Hi, I've read the paper and have questions about the Train and Test method

1. In the case of Train, is it the same as the method of fine-tuning the BERT model for text classification? In this case, is it correct to understand that only text and tags are used without considering the user?

2. In the case of the test, it was said that the predicted probability is averaged.
   I wonder if 1) Did you take the method of storing the predicted value of the previous post in a specific variable,
   or 2) put the previous posts together as a model input.

   For instance, the input of case 1) is {post 1} {post 2} {post 3} ,,,
   and the input of case 2) is {post 1} {post 1, post 2} {post 1, post 2, post 3},,,,
   Which of the two is correct?

Thank you