If you would have said your future goal was to be a “Youtuber” two decades ago, everyone would have thought you have lost your mind. However nowadays, “about 3 in 10 American children want to be a Youtuber, and lesser preferences include being a teacher, astronaut, and a musician. [1]” Being an established Youtuber is not easy, Youtube has over a billion users, totaling 228575 distinct user channels, 400249 user recommendations, and 400249 Youtube recommendations [2]. Youtube remains to be competitive by utilizing its click-through-rate with its accurate video recommendation algorithms. These numbers are indeed daunting for some of us who would want to start a youtube channel in hopes of earning money. Then, the common question of all YouTubers is “How can I get more views?”
In the study by Yonghyun Ro, Han Lee, and Dennis Wan, nodes in the network represent a youtube video. Two nodes are connected with a directed edge if one video recommends the other. The researchers study the technique of PageRank, which allows them to observe a pattern for a highly viewed video for the categories. The PageRank algorithm works as so: each video on youtube is given a PageRank, which represents the number of other videos it connects to via edges. The more outgoing edges it has, the more links that particular video has, and hence it will have a higher PageRank. However, there is a slight problem to this, it becomes very easy for others to forge the PageRank of a video so that their PageRank is higher. This introduces the Random Surfer Model, which randomly picks a video to visit, and goes to the videos that it’s linked to and then picks another video randomly, so on. It keeps a score of how many times a video has been visited, those videos that have more links are visited more frequently. This brings up another problem: what if all videos are not connected via a link? Then it limits us to only visit that specific cluster of videos and disregard the rest of the videos. This is solved by occasionally resetting the random surfing by its damping factor so that it does not disregard any of the videos.
The researchers analyzed the total PageRank score of Youtube around 750,000, where each video has its own PageRank score. Youtube allows each video to have about 20 recommended videos. Then, if video A is one of the recommended videos of video B, there is a directed edge from B to A. As understood from the PageRank algorithm, if a video has a higher PageRank score means that the video is related to many other videos in the network. Thus, it will have a larger influence than other videos within the network and will be recommended to a larger percentage of the users.
Above is the data that compares the PageRank of a video to its views and its video length. The graphs have x-axis as the PageRank score in log10 base and y-axis is the percentile of each video feature (views and length).
From the graphs, notice that there is a high correlation between the PageRank score and the views associated with the video i.e the higher the PageRank score the higher the views for the particular video. On the other hand, if we take a look at the graph that compares the PageRank of a video to its video length, we notice that there is a relatively low correlation. This tells us that most of the videos that have a high-influence and high PageRank scores are usually a shorter length video compared to the other videos.
So, next time your uploading your youtube video and you find yourself asking “How can I get more views?”, be sure to remember this algorithm and focus on increasing the PageRank of your video!
References: