Abstract: In this paper,a method was proposed,which combined wavelet multi－resolution decomposition and neural networks to forecast network forum topic hot trend.That was first,using wavelet multi－resolution to decompose the original time series formed from click numbers or reply numbers of forum post,to create a series of wavelet coefficients of the sub－series.Then,through BP neural networks to classify the hot post and normal post,coefficients were selected from all wavelet coefficients to form a feature coefficients set,which best enlarged the distance between different classes and reduced the distance within same classes.For the unknown time series,their feature coefficients were sent neural networks to forecast.This proposed method was tested to make better prediction accuracy for network forum topic hot trend．