Description
It's nice to be able to select the source nodes for pageRank, but in reallity the probability of starting to a given node is not even across all source nodes. As an example, if you create a markov chain from customer behavior, each customer will start from a specific channel, so you could have 10 customers starting from channel A et 4 from channel B.
To overcome the limitation of not being able to set a starting score in each sourceNode, you can create a node zero as a unique sourceNode and link it with weighted relationships to your actual source nodes which represent where customer will start their journey in an uneven way. The weight will act like the initial score if I'm right.
But this requires to create a node zero which is not natural in the workflow.
Here is an example of this in cuGraph from NVIDIA, section at the bottom of the page