About Multiple Entity Pairs in the Same Sentence about pure HOT 5 CLOSED

Another question: do you consider the directionality of 'hyponym of' in scierc?

from pure.

Hi Haishuo,

For multiple entity pairs in one sentence, do you separately generate a new sentence for each entity pair during training? How do you sample negative samples? I mean entity pairs without relations

Yes, we generate a training instance for each entity pair. Suppose there are n entities in a sentence, we will generate n*(n-1) instances. We didn't downsample negatives and used all of them among those n*(n-1) pairs.

do you consider the directionality of 'hyponym of' in scierc?

Yes. We simply considered all relation types as asymmetric relations in our implementation.

from pure.

Thank you for your reply! How do you implement it? I didn't find the code of creating two different labels. For example, creating two different labels: hyponym of(S, O) and hyponym(O, S)? I guess you process them when generating data?
Just let the latter entity to be subject and the former one to be object

from pure.

We didn't create two different labels for a relation. For example, for a relational triple (S, hyponym of, O), the label of (S, O) will be "hyponym of" and (O, S) will be "no relation". Creating directional labels for a relation type may bring benefits, but we didn't test that.
You can refer to this code for converting entity data to relation data.

from pure.

I got you! Thank you

from pure.