In network analyses, it is often necessary to sample a portion of the network data due to its large size or limited access permissions. The quality of a network sample can be assessed by estimating some basic properties of the original network. Random walks have been used for network sampling because they require only local neighborhood data of nodes, and various random walks have been developed specifically for hypergraphs. In this study, we introduce a general framework for random walks on hypergraphs. This framework allows for describing different random walks by adjusting the probability of selecting a hyperedge to which a node belongs during the sampling process. We use this framework to evaluate an estimator for hyperedge size distribution and compare the sampling efficiency of two random walks in empirical hypergraphs.