Feedback control is widely applied to the campaign management in online advertising. Learning the pattern of user traffic on Internet plays an important role in solving the control problem. In this paper, we focus on characterizing the seasonality, e.g., time of day (TOD) pattern of Internet user traffic for individual ad campaign. We model the seasonality using a truncated Fourier series with a set of amplitude and phase parameters. These seasonality parameters are estimated in a Bayesian framework using a minimum mean square error (MMSE) estimator, with their prior distribution learnt from historical data of a large number of campaigns. The proposed Bayesian method is shown to be robust and renders sensible seasonality for campaigns of disparate noise levels.