TPlinker-joint-extraction icon indicating copy to clipboard operation
TPlinker-joint-extraction copied to clipboard

Published 20 hours ago • verified publisher icon 131250208

implementation questions

Open world2vec opened this issue 3 years ago • 2 comments

Hi, Thanks for your good idea and paper, have some questions about TPLinkerPlus

  1. I see you use logsum to calculate the loss, is it better than BCE?
  2. Is there any reference/paper for "conditional" layernorm which is used for "cln" shaking_type? Thanks.

world2vec avatar Apr 02 '21 02:04 world2vec

  1. It brought a slight improvement. For more details, see Extend "softmax+cross entropy" to multi-label classification problem
  2. For all I know, no. But you can refer to this blog Conditional text generation based on Conditional Layer Normalization.

131250208 avatar Apr 06 '21 01:04 131250208

Thanks for your quick response and guidance. Looking forward to your paper to know the gains of these 2 component. The first run of my implementation only have about 90.x on the NYT dataset without these 2 component(no hyper parameter tuning, just converge) which looks too low. I will double check the code and tuning a little. Thanks.

world2vec avatar Apr 06 '21 06:04 world2vec