Where is the embedding for relations?

xiangrongzeng / copy_re

Release for acl18 paper "Extracting Relational Facts by an End-to-End Neural Model with Copy Mechanism"

197 stars 36 forks source link

Where is the embedding for relations? #30

Open jbjeong91 opened 4 years ago

jbjeong91 commented 4 years ago

I just found 'words2id','relations2id' and 'words_id2vector'. where is the embedding for relations?

xiangrongzeng commented 4 years ago

There is no relation embedding. We recognize the relation by "classification".

jbjeong91 commented 4 years ago

Thx for your reply. Then I have one more question. When extracting max 4 triples, 12(0~11) decode sequence steps used and, each '0, 3, 6, 9' sequence step classifies a relation?

xiangrongzeng commented 4 years ago

Exactly

jbjeong91 commented 4 years ago

Hi, I have one more question. in the NYT dataset, I can see one 1-dimension list and two 2-dimension lists in 'test.json'. what is the role of each list? I just infer that the one 2-dimension list is for sentence instances and the other one is for triplets in each sentence. then what is the role 1-dimension list?