Predicting the characteristics of weibo reposts through time series analysis is very important topic in the research area of social network. An improved method by introducing the wavelet transform mechanism based on the ARMIA model is used in this paper. We propose an improved data processing method based on wavelet theory to remove noise in time series and to enhance the ability of predicting the weibo reposts. Experimental results show that multi-scale analysis combined with high-frequency zeroing can retain the effective part of the data and remove the noise part of the high-frequency, making the curve smooth.