Hi Kevin, Thanks for linking the paper. Using LSTMs for predicting next state in user journey looks promising. I will take an in-depth look.

For the purposes of this article, the intent behind using DFG was to enable some sort of EDA. As you rightly said, journey of interaction with the podcasts in itself do not reflect any conscious decision making process on the end-user's part, (unlike sequential processes involved in, say the supply chain sector).