Research on short video content recognition and recommendation algorithm based on deep learning

This paper discusses the method of short video content recognition and personalized recommendation by using deep learning technology. Different from traditional behavior recognition techniques, deep learning shows its strong ability in feature extraction, which can automatically learn adaptive and highly distinguishable features. First of all, considering the dual properties of time and space of video content, this paper adopts deep learning model to extract spatial and temporal features of video respectively, and combines them to form spatiotemporal fusion features. A video behavior recognition algorithm based on spatiotemporal convolutional network is designed to process video behavior recognition as the acquisition of spatial and temporal information of video. Based on the two-flow convolutional network, the effects of optical flow image and salience image on behavior recognition are compared and analyzed. On this basis, the attention mechanism inspired by the human visual system is introduced, and the space-time fusion model based on the attention mechanism is constructed. By assigning higher weights to key frames in video clips, the model focuses more on the important parts of the video, thus reducing the impact of redundant information on behavior analysis. Finally, the reliability of the model is verified on a real world data set.