As a project in progress, we want to share some of the ways we are working with the music video data.
The Monopoly of Entertainment Companies
With the rise of the “Hallyu” wave and the accessibly of video sharing platforms such as YouTube, it is useless to deny the power of the international fanbase. Arguably, not only has their impact enriched certain companies to a great extent, but they also served as another audience for the company’s profit. To a certain extent, this meant that the already established companies like the Top 3 (YG Entertainment, SM Entertainment and JYP Entertainment), who all have English speaking artists, have grown to monopolize the K-Pop industry. Therefore, taking 300 of the most popular K-pop music videos on YouTube, the figure below demonstrates the distribution of the most popular artists and the entertainment companies associated with them.
The Distribution of Artist by Gender
As there seem to be a pattern between the popularity of artist with the entertainment company they are signed with, the next variable that became apparent was the gender of the artists. With the accelerated popularity of groups like BTS, Big Bang and 2NE1, the question of whether the gender of the group was a crucial factor in deciding which bands became the most popular was discussed. Therefore, the graph below demonstrates the first top 100 most viewed music videos and their relationship with gender and even the entertainment companies associated to disclose insight into what types of groups the main companies debuts. Interestingly, although there were more variety of girl groups, the lesser array of boy groups had more of their music videos in the top list. Furthermore, even the top entertainment companies released more boy groups than girl groups.
AI, Deep Machine Learning, and Pose Detection
As the project continues, we will be working with the data that has been produced by applying DensePose, an AI pose detection method developed by Facebook, to the video corpus and rendering an output of body and body part coordinates to JSON files. Below you will see an example music video and a second video that shows the pose detection mesh produced from DensePose. The project team plans to study the data further and attempt to edit the video footage base on identified poses.
Body Figures Detected Per Frame
In order to provide users with a method to quickly glance at the number of bodies detected during each video, we have decided to make interactive visualizations that present number of figures detected per frame. Typically, a video is made up of 24 frames per second, meaning a single music video can have up to 5,000 – 6,000 frames. An example of this visualization can be found below, which was generated from the same music video from the previous section. The y-axis depicts the number of figures, and the x-axis has been normalized to frame percent. Click here to explore the rest of the visualizations.