Question about Feature dimensions

[LuisaWerner]

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