### Abstract

Classical graph modeling approaches such as Erdos-Renyi (ER) random graphs orBarabasi-Albert (BA) graphs, here referred to as stylized models, aim toreproduce properties of real-world graphs in an interpretable way. Whileuseful, graph generation with stylized models requires domain knowledge anditerative trial and error simulation. Previous work by Stoehr et al. (2019)addresses these issues by learning the generation process from graph data,using a disentanglement-focused deep autoencoding framework, more specifically,a beta-Variational Autoencoder (beta-VAE). While they successfully recover thegenerative parameters of ER graphs through the model's latent variables, graphssuch as BA graphs, due to their oversimplified decoder. We focus on recoveringthe generative parameters of BA graphs by replacing their beta-VAE decoder witha sequential one. We first learn the generative BA parameters in a supervisedfashion using a Graph Neural Network (GNN) and a Random Forest Regressor, byminimizing the squared loss between the true generative parameters and thelatent variables. Next, we train a beta-VAE model, combining the GNN encoderfrom the first stage with an LSTM-based decoder with a customized loss.