a question about the key 'label' in the test dataset

[GuanNiPiShi123]

Thanks for publishing your code to public.
I wonder how you obtain the " label " in test_CNNDM_roberta.jsonl. Do you use the greedy selction algorithm mentioned in SumRunner or use the BertSum prediction to obtain it? If the former, I think it is cheated to some degree when using MatchSum to test, because the cadidate summary is based on the selected labels ,and the selected labels is caopared with the gold summary.
If the latter ,I suspect the oracle is not as higher as R_1:52.59 R_2: 31.23 R_L: 48.87 ,because the BertEXT model can't predict as accurate as labels using the greedy selction algorithm.
Looking forward to your reply.

[buster120]

@GuanNiPiShi123 hello，do you have the code of greedy selction algorithm mentioned in SumRunner ,Can you send me a copy if you have? thank you.
Looking forward to your reply.