search

NTMC-Community / MatchZoo

Facilitating the design, comparison and sharing of deep text matching models.
Apache License 2.0
3.82k stars 898 forks source link

how to use soft-labels on ranking task? #777

Closed lis-kp closed 3 years ago

lis-kp commented 4 years ago

hi, I wonder if it is possible to use probability values on the label columns when using dssm model for ranking. for example, instead of having something like [1,0,0,0,0], I would have [0.4, 0.1, 0.2, 0.25, 0.25 ].

thanks.

faneshion commented 4 years ago

Right now, we treat all labels as an integer, which is the case of usual ranking or classification task. Could you please tell us in which case you would use the float label?

lis-kp commented 4 years ago

using soft labels can lead to better model generalization, so I was wondering if it would be possible to use them in matchzoo.