The research focuses on solving the challenges of traditional cultural educational resources in digital applications, such as dull generation effects, fragmented knowledge organization, and cold-start recommendation difficulties. Firstly, a style migration model based on adaptive instance normalization (AdaIN) is proposed. Then, we disassemble and associate cultural resources through knowledge elements to construct a structured knowledge graph containing more than 10 kinds of granularity, such as terms and events. Finally, a one-stop community learning system based on the multimodal meta-learning recommendation algorithm (UMMeLE) is designed. The statistics of 500 traditional cultural resources found that although the overall quality is fair, with most of the resources rated >70, there is a typical long-tail effect, with up to 70.4% of the resources having less than 1,000 views and only 17.2% of the resources having a download rate of more than 20%. In this paper, the multimodal meta-learning UMMeLE algorithm reduces the prediction error RMSE to as low as 0.697 on the real TCE dataset.
