In Bayesian statistics, a hyperprior is a prior distribution placed on the hyperparameters of another distribution, which is itself the prior for the parameters of a model. To clarify, the Bayesian framework involves using prior distributions to quantify our beliefs about parameters before observing data. When these parameters have their own parameters, which we don't know and want to estimate, we refer to those as hyperparameters. The distribution assigned to these hyperparameters is what's known as a hyperprior.
New to topics? Read the docs here!