Research on Video Ratings Prediction and Survival Analysis

Aiming at the poor accuracy of the existing Internet video ratings prediction, this paper proposes a self-correlation prediction based on the view amount and a survival analysis scheme based on the video component factors. The self-correlation prediction scheme uses BP neural network and autoregressive moving average (ARMA) to predict the short video, and a multi-direction hybrid prediction method based on the ladder data to predict the long video. The experimental results show that the prediction accuracy of the above methods can reach up to 90%. The survival analysis scheme adopts the COX proportional hazard model (COX model) to analyze the correlation between the component factors of the variety shows and their ratings, then predicts the ratings according to these factors. The experimental results show that the concordance index (C-index) can be up to 0.8.