Prediction of the popularity of artificial intelligence short videos based on MFMA

In this paper, a short video popularity prediction model MFMA is designed based on Multilayer Perceptron (MLP) and artificial intelligence, and the popularity prediction in saturation is modeled as a regression problem. Through multimodal fusion, the four video attributes of visual, auditory, text, and social features are compressed as the input of the network to the greatest extent. Using nMSE and SRC as evaluation indicators, compared with other classical regression algorithms and evaluating the impact of different modal deletions on the performance of the model under the assumption of conditional independence, a large number of experiments on the real dataset of Douyin show that the MFMA model has the best prediction effect and is robust to internal noise and external uncertainty, among which the social mode has the most significant impact on the model performance.