Ask about ConvE model

[oathaha]

I tried to follow code based on concept in the paper but I did not find any code that does dot product with object entity. Can you tell me where is object entity being fed during forward propagation?

[TimDettmers]

ConvE uses 1-N scoring, meaning that we take the subject and relation and score it against all other objects at once (subject, relation, E). This is done by multiplying by the transpose of the entire entity matrix. You can find the relevant line in the code in model.py in line 120.

[YuyangWei]

Hi!
You multiplied [subject, relation] by the transpose of the entire entity matrix, but I wonder how you supervise the learning process of the convolutional network? I am looking forward to your kind reply!

[thsno02]

ConvE uses 1-N scoring, meaning that we take the subject and relation and score it against all other objects at once (subject, relation, E). This is done by multiplying by the transpose of the entire entity matrix. You can find the relevant line in the code in model.py in line 120.

Hi Tim,

Did you mean 1-N scoring trick is to compute the scores in the matrix multiplication manner instead of vector multiplication? If it is, why 1-N can boost the training pass as u mentioned in the paper (Fast Evaluation paragraph 2).