waste-wood / e-care Goto Github PK

Hello！It is really an excellent work. I rerun your code for casual reasoning, and obtain accuracy about 66% on the development set, with BERT_base encoder. This is far lower than the performance on the test set, which is more than 75% according to your report. What may be the matter? Or your result? Thanks a lot!

Hi, excited to see such an interesting new dataset. There are some questions about the evaluation metric CEQ.
the CEQ score has a significant positive relationship with the human evaluation results. but I didn't find the corresponding section in the code，nor in the leaderboardleaderboard.
where can i find the CEQ metric.