In order to improve the matching degree of short video media resource push, a personalized short video media resource push method based on cloud computing is proposed. A Hadoop-based personalized push framework for short video media resources is designed, based on the collected information, the latent Dirichlet allocation model is used to classify the types of short videos according to the theme, and the convolutional neural network model based on the attention mechanism is used to identify the theme of short videos at wonderful moments; The short video media resource recommendation module determines the interest value of the user in the unbrowsed short video according to the user’s historical short video browsing behavior, generates a recommendation list according to the value, and presents the push result to the user through the data presentation layer. The experimental results show that the method can realize the personalized push of short video media resources with the topics that users are interested in, and when the optimal number of topics extracted from each bullet comment text is 2 and the length of the recommendation list is 3, the push effect is the most prominent; the method can improve the performance of personalized push of short video media resources, and the pushed content is more in line with the user’s interest.