search

nanguoshun / LSR

Pytorch Implementation of our ACL 2020 Paper "Reasoning with Latent Structure Refinement for Document-Level Relation Extraction"
127 stars 21 forks source link

about dis_h_2_t？ #17

Closed yangxia605 closed 4 years ago

yangxia605 commented 4 years ago

hello@nanguoshun : thanks for your excellent work!!! I am confused about the value of dis_h_2_t? And every mention will be the triple's head or tail,so every mention will Pairwise to tirple(h,t,r). So how to calculate the distance between h_2_t? I have view the code , but not for the whole document distance. Thank you Soooo much!!!

nanguoshun commented 4 years ago

Hi @yangxia605 thanks for the question. We use DocRED's original code to compute the distance. The distance is mapped to 1~10 as indicated from line 111 to line 121 in the Config.py. It is quite straightforward if you inspect the value step by step under debugging mode.

yangxia605 commented 4 years ago

Thx for your patient explanation。