Question about paper

[DreamerDW]

Excuse me, I have a question about this paper. In the second step of FeatWalk, select a feature category based on the normalized feature of instance i, Why can an element yim in the feature matrix represent the probability of selecting the m-th feature am?

[xhuang31]

If both node i and node j have m-th attribute category with high values (say 0.8, after normalization), then we would like node i has a high probability to jump to node j.

If both node i and node k have another attribute category (p-th) with low values (say 0.1), then we would let node i have a higher probability to jump to attribute category m-th, instead of p-th.

I hope that this example could answer your question.

Thanks.

[DreamerDW]

Thanks for your reply