In today’s fast-changing era, watercolor painting, as an art form with a long history, still maintains its unique charm and vigorous vitality. In this paper, the dynamic theme model is used to obtain the theme evolution of watercolor painting work style and combined with the TF-IDF algorithm to mine the time series of the style evolution of watercolor painting works and the fluctuation of the vitality of the style of watercolor painting is calculated by evaluating the novelty and inheritance degree of the new word theme of the data. Then, the theme data of watercolor painting works obtained by DTM is input into DSSM and DNN for the output of low-dimensional semantic vectors to solve the similarity of the style themes of watercolor paintings. The enhancement of the style vitality of watercolor paintings between 1986 and 2022 can be up to 3.75 times, and the value of style vitality of watercolor paintings between 2013 and 2022 ranges from 0.31 to 0.49, and the value of coherence is in 0.51 floating around, showing dynamic balance. The lowest thematic similarity of watercolor painting styles between adjacent years after 2016 is only 0.04. There are obvious differences in the styles of watercolor painting works from the pre-1980s and post-21st century. Using the DTM model, TF-IDF algorithm combined with a deep network model, the time series of watercolor paintings can be deeply mined, which helps to clarify the change of watercolor painting style in different periods.