You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Hi, thanks for this interesting approach! I also read your paper and I still have an open question about the input features and the data set that you use.
Am I right that data sets like AIFB don't naturally have input features? In the documentation on stellargraph, it is indicated that one-hot encoded vectors are used instead. How are these one-hot encoded vectors created? Are they just indices or based on which information is the one-hot encoding made?
I also wonder whether your model takes into account that entities might be of different types (affiliations, person) and have different features so that the dimension of $h_i$ and $h_j$ might be different? And the weight matrices would not only depend on the relation type but also on the type of the node? Or am I missing something here?
Thank you :) Luisa
The text was updated successfully, but these errors were encountered:
Hi, thanks for this interesting approach! I also read your paper and I still have an open question about the input features and the data set that you use.
Am I right that data sets like AIFB don't naturally have input features? In the documentation on stellargraph, it is indicated that one-hot encoded vectors are used instead. How are these one-hot encoded vectors created? Are they just indices or based on which information is the one-hot encoding made?
I also wonder whether your model takes into account that entities might be of different types (affiliations, person) and have different features so that the dimension of$h_i$ and $h_j$ might be different? And the weight matrices would not only depend on the relation type but also on the type of the node? Or am I missing something here?
Thank you :) Luisa
The text was updated successfully, but these errors were encountered: