Final Project of UCLA STATS101C YouTube is a large scale video-sharing platform owned by Google since late 2006. The site allows users to upload, view, rate, share, create playlists, comment on videos, and subscribe to other users. Of these interactions, views stand out as particularly important because they are a direct measure of engagement with a video and also help to determine how much revenue the content creator will make. A pressing goal for a content creator is to know how fast a video will grow, especially within the first few hours of its lifetime. A video’s early growth pattern can be a broader indicator of the eventual success of the video as well as the overall health of the channel.

In this project, we aim to predict the percentage change in views on a video between the second and sixth hour since its publishing. In order to predict this metric, we have several video features at our disposal, generally fitting into four categories explained below.

Thumbnail Image Features Each video on YouTube has a thumbnail image, which is the first image that a user sees when deciding whether or not to click on the video. Content creators usually spend a great deal of time crafting the thumbnail image to be interesting, eye-catching, and representative of the content in the video. Several features in the data are extracted from the thumbnail image. Some features are more intuitive such as the average pixel value or the percent of nonzero pixels. Others such as the histogram of oriented gradient (HOG) features or convolutional neural network (CNN) features are harder to interpret but might give insights into the underlying structure of the thumbnail image.

Video Title Features Another important factor in a video’s views is the title of the video, which a user also sees when deciding whether or not to watch a video. Content creators might try to craft the title to achieve various goals such as invoking certain emotions, asking a question to be answered, etc. Several features in the data are extracted from the video title. Some features are intuitive such as word count, character count and count of certain punctuation. Others, such as the Doc2Vec features, are harder to interpret but might give insights into the underlying semantic structure of the title text.

Channel Features The third family of features relates not to an individual video but rather the channel which published the video. Indeed, an already very popular channel is likely to record different growth rates on a new video than a less popular channel, other factors held constant. This family of features consists of four features, each of which is measured as low, low_mid, mid_high, or high. For example, one feature is the number of subscribers that the channel has, as of June 8 2020. Another is the number of total views on all videos on the channel as of June 8, 2020.

Other Features Other possibly useful features include the duration of the video, number of views after two hours since publishing, and the timestamp when the video was published.

Please see the 'Feature_Descriptions.xlsx' spreadsheet for full description of each feature.