Questions about the data preprocessing method

Open mingyang-wang26 opened this issue 2 years ago • 1 comments

Hi,

thanks for your excellent paper! I have a questions about the data preprocessing in the implementation code.

A small observation about data processing, for the 'yahoo', 'amazon', 'agnews' and 'dbpedia' dataset, besides the 'content' column, they also include the 'title' (and 'answer' for 'yahoo') column. In the implementation in IDBR and MbPA++, they only use 'content' for training and discard other information such as 'title' and 'answer'. In Progressive Prompts all texts are taken into account, I'm wondering if this will leads to some performance difference and did you compare them?

Many thanks in advance for your clarification!

Jul 10 '23 10:07 mingyang-wang26

Hi, @DDD26

have you addressed the problem above for the value of "select_k_per_class"?

Jul 26 '23 09:07 JHL328