A Study Of Popularity Prediction For Online Video Services

As the growth of the large video data explosion, online video service faces a sequence of serious problem, such as, resource overload, accurate prediction to online video service attaches more and more importance. Based on the research about big data collected from one large-scale online video service provider, and according to the different stages of video show, the popularity of online video services divided into “accurate prediction” and “sync prediction”.1) Before release, concerned the issue of traditional prediction model, a method based on a kind of deep belief networks(DBNs) for video popularity prediction is proposed. Firstly, combing attention of social network with search interest of video keywords, modeling and quantitative processing is performed for the influence factor. Secondly, the structure of each network is confirmed on the basis of input and output variable, optimizing the networks parameter and prediction model. Finally, Though using the date collected from online video service provider, the paper performs training for DBN. After numerous crossover experiments and comparisons, the result shows that DBN-based prediction approach can acquire good performance, implementing the primary prediction for video popularity.2) After release, by means of statistical analysis of the early video views sequence, an approach to predict views of the online video is presented in order to predict the video popularity synchronously. According to the difference of sequence feature existing in video views, various models can be chosen which achieve higher accuracy on synchronous prediction for non-stationary domestic video and seasonal foreign video. Compared with the moving average, exponential smoothing method and the forecast method of least squares, ARMA model prediction approach obtains better performance.Through the study of deep belief networks and time series model, the paper implements timely, continuous and accurate predictions for video views in different period. In contrast with traditional forecasting methods, the prediction strategy not only can provide decision-making information for risk assessment, publicity affair, and investment of online videos, but also can get accurate amount of video views range after the video release, provide support for reasonable advertising, the resource storage and business decisions.