search

chiayewken / Span-ASTE

Code Implementation of "Learning Span-Level Interactions for Aspect Sentiment Triplet Extraction".
MIT License
169 stars 45 forks source link

evaluation and predict test #31

Closed arroyoaaa closed 2 years ago

arroyoaaa commented 2 years ago

Excuse me, how to you calculate the precision, recall and FScore?您好，请问您是用什么方法（公式）计算精确率、召回率和F1值？ And what do the last two numbers mean in predict test of each sample?（as shown below）以及在每个样本的预测结果中的最后两个数字代表什么意思呢？

捕获

chiayewken commented 2 years ago

Hi, the two numbers are raw scores and softmax scores: https://github.com/chiayewken/Span-ASTE/issues/14#issuecomment-1034495370

Regarding the precision/recall/f1, we use micro f1 as shown here: https://github.com/chiayewken/Span-ASTE/blob/9e2b93d09b7e7ef70e2a8285cc9da5ea41b20f1d/aste/wrapper.py#L99