Data Analysis and Prediction Based on Video Platform

Abstract

Nowadays, there are video platforms emerging and data virtualization and prediction on those platforms draw little attention. Related research on it is beneficial for both content creators and advertisers as it helps decision-making. In this study, a video platform in China called Bilibili was selected. This paper acquired the dataset from a number of top-rated video uploaders in their field from a data share website and preprocess them to combine them for each month. For each frame of the data, the number of likes, shares, comments, archives, and bullet charts was used to predict the number of views. In the analysis, this study mainly used random forest and sklearn to achieve the prediction. 368 data points were utilized, and in each data point the amount of favorite, share, save and play are combined to make the prediction. The prediction that is made by using random forest is much more realistic than that made by using sklearn because some of the statistics in the sklearn graph reach negative, which is not realistic in the real situation. According to the experimental results, it can be found that the play rate has a positive relationship with the amount of favorite which is the most important factor in play rate. The amount of share is the secondary factor, and the least effective factor is the amount of collecting.