hi! thanks for your great job! i have a question, why you don't compute the transition probabilities directly? you have the graph but you use the Transition Attention Map. i dont understand you. another question is that when you initialize the node in GCN, you use one-hot code for category, and continuous value for poi_ID and longitude and latitude, how did you came up with that, and does it really work?
hi! thanks for your great job! i have a question, why you don't compute the transition probabilities directly? you have the graph but you use the Transition Attention Map. i dont understand you. another question is that when you initialize the node in GCN, you use one-hot code for category, and continuous value for poi_ID and longitude and latitude, how did you came up with that, and does it really work?