search

thunlp / PL-Marker

Source code for "Packed Levitated Marker for Entity and Relation Extraction"
MIT License
260 stars 35 forks source link

re_classifier #31

Closed wsh2836741 closed 2 years ago

wsh2836741 commented 2 years ago

感谢您杰出的工作。想问下，关系预测里为什么需要有两个re_classifier（re_classifier_m1和re_classifier_m2）？相当于是subject和object分别计算的吗？传统做法不应该是拼接到一起然后过一个re_classifier多分类吗？

YeDeming commented 2 years ago

你好，效果是一样的， W_classifer[x;y] = W_x x+W_y * y