Abstract: In astrophysical research, accurately accounting for the effects of interstellar extinction and reddening is crucial for optical and near-infrared observations. Information on the interstellar reddening of stars is essential for uncovering their intrinsic properties. The optical survey project of the Chinese Space Station Telescope (CSST) will provide us with a vast amount of very low resolution spectral data for stars, and measuring the reddening information of stars based on this data is of great significance for further determining stellar parameters and understanding the nature of the Milky Way. This study proposes a machine learning method based on random forest regression, which uses the normalized flux of spectra as input parameters to train the intrinsic colors of stars, aiming to accurately estimate the reddening values of stars in very low resolution CSST spectra. This study simulated CSST spectra using the Next Generation Spectral Library (NGSL), predicted the accuracy of the proposed method, and also discussed the impact of different bands and effective temperatures on the accuracy of the results. The comparison between the reddening values E\;(g-i) obtained from the seamless spectra of different CSST bands and the true values shows that at a spectral signal-to-noise ratio of 100, the average deviation for the GU band is 0.0005 mag with a standard deviation of 0.0272 mag; for the GV band, the average deviation is 0.0008 mag with a standard deviation of 0.0286 mag; for the GI band, the average deviation is 0.0008 mag with a standard deviation of 0.0271 mag; and for the full band, the average deviation is 0.0003 mag with a standard deviation of 0.0252 mag. This method, as part of the scientific pre-research for CSST, can be directly applied to CSST data in the future, providing basic support for scientific research on CSST.