With the rapid growth of user base, the scale of short videos is also exploding. When faced with a large amount of video content, users often feel confused and find it difficult to find the content they want. Real-time collection of user browsing, liking, commenting, sharing, and other behaviors on short video platforms was conducted using data collection tools such as site logs or application interfaces. Image features were extracted from video frames and convolutional neural networks were used to extract visual features from each frame of the image. A personalized recommendation list based on user interests and the characteristics of video content was built. On this basis, using deep learning-based multi task learning methods or attention mechanism models, joint modeling of user interests and video content was carried out to improve recommendation accuracy and personalization level. When the user sample size was 1000, the actual number of clicks was 850; the recommended content was 1200; the user’s actual interest rate was 70.83%. This article presented an artificial intelligence (AI)-based intelligent short video recommendation algorithm technology in the context of digitalization. Through intelligent recommendation methods, it helps users find video content that meets their preferences.