Temporal exponential random graph models (TERGM) are powerful statistical1models that can be used to infer the temporal pattern of edge formation and2elimination in complex networks (e.g., social networks). TERGMs can also be used3in a generative capacity to predict longitudinal time series data in these evolving4graphs. However, parameter estimation within this framework fails to capture5many real-world properties of social networks, including: triadic relationships,6small world characteristics, and social learning theories which could be used to7constrain the probabilistic estimation of dyadic covariates. Here, we propose triadic8temporal exponential random graph models (TTERGM) to fill this void, which9includes these hierarchical network relationships within the graph model. We10represent social network learning theory as an additional probability distribution11that optimizes Gibbs entropy in the graph vector space. The new parameters are12then approximated via Markov chain Monte Carlo maximum likelihood estimation.13We show that our TTERGM model achieves improved fidelity and more accurate14predictions compared to several benchmark