How Does YouTube Choose Suggested Videos?

The video recommendation system for YouTube has only one goal. They want to provide you a high quality, personalized video recommendation that is appropriate to your interests. In doing so, they will keep you on their platform as you click around and watch their targeted ads. Your browsing history on Google, Chrome and YouTube largely influence the search results of your YouTube recommendations. You receive these suggestions on your YouTube home page and also as in-app notifications.

There are several factors that influence your search results and video recommendations. These are fed into YouTube's algorithm which makes these decisions as to what videos you would like automatically. It does so by looking at certain data. Let’s have a look at the different types:

• Content sources: there are two main sources of how YouTube draws your recommended videos and feeds the algorithm. First is the content data you type in the search box, such as titles, description and metadata and second is the data of user activity. It is categorized into a number of attributes, such as your favorites, ratings, number of viewings, etc.

• Your viewing activity: before YouTube suggests you recommended videos, it determines the set of associated videos you are likely to watch after the currently viewing video. To do this, it uses the method of association or co-visitation which is done by identifying a series of videos you have recently watched in a single session. The algorithm finds a connection between the videos you have watched and suggests you the videos you may be interested in viewing.

• Ranking recommendation: once YouTube generates a set of recommendations, it ranks them according to a number of signals. Then, they are organized into different groups, as per their user specification, diversification and video quality. User specification is used to bring forward a video similar to your unique preferences. User specific signals are created based on your view time and count of watching certain type of videos. Diversification is done identifying videos that are very similar to each other. Exactly identical videos are removed and more varied types of video are brought in. The logic behind diversifying is that users have multiple interests and related viewing preferences. Recommended videos that are too similar to each other may not reflect a user’s overall preferences. Video quality includes a number of metrics, such as video ratings, view count, favorites, comments and shares.

• Viewer engagement: YouTube pays due attention to how many comments, likes and dislikes a video draws from the audience. Videos that are most successful in engaging the viewers always top the charts and show on your suggested videos. YouTube suggested videos algorithm works on the performance of certain videos on YouTube. They promote videos that are successful at keeping viewers engaged and encouraging them to like and comment on them. In addition, videos that get lots of likes and comments are always promoted by YouTube. Videos that are successful at receiving long viewer retention and watch time are those that keep people engaged.

• Thumbnails: a thumbnail is the small still image of the video you see with the link of a video. Videos that have thumbnails relevant to the title and the content of the video are most popular among the audience. Videos that have irrelevant thumbnails receive negative response and comments from the audience. So, videos with relevant and attractive thumbnails become popular and YouTube includes them in its suggested videos recommendations.

• YouTube Session Starters: YouTube video suggestions are mostly the ones that start viewing sessions for the audience. It also considers the way a YouTube channel starts a viewing session, and by what frequency. YouTube has data about views and session times of viewers, and it suggests channels and videos that often start viewing sessions based on that data. For instance, if you visit YouTube once in a week, your viewing session may be 20 minutes long. If you come to YouTube twice a week, your viewing session may be 50 minutes long, and 4 visits total up to 150 minutes viewing session. Consequently, more suggestions mean more people coming to YouTube and more videos being featured on suggested video recommendations.

• Outside YouTube: YouTube stays in touch with its users outside of YouTube too via email lists, social profiles, media connections, websites, etc. Even if these are very small platforms, they help in catalyzing starting sessions and knowing users’ preferences, likes and dislikes. Based on this data, they grow their audience by suggesting them relevant videos. On their Facebook and other social media pages, they also schedule their new releases. If a man knows when he should come back for a new video on a channel, he can come and find his video when it is set to go live. How this feeds into the algorithm, however, is not always well understood. This is partly the reason for the recent changes to GDPR which have affected Europe^[1].

YouTube faces a number of challenges while choosing suggested videos for its users. First of all, there is a staggering amount of videos being uploaded to YouTube every single day. Many of them are similar to each other, and choosing one among them is a difficult thing to do. Secondly, many of these videos have irrelevant titles, inappropriate descriptions, extraneous thumbnails and other such signs of poor metadata. Third, the data that YouTube uses to fathom user interest is quite vague and inappropriate at times. For instance, when a person is watching a video about a particular product, YouTube doesn’t know his or her purpose. They may be interested in buying the product, they may be trying to gain knowledge about it, they might be writing marketing content about it, or they just stumbled over that video by chance. Another challenge is that the recommendations need to be refreshed at regular intervals of time. Suggesting certain videos time and again may not interest the viewers.

There are some other challenges that come on the way of dominating the column of suggested videos. One is the ‘up next’ feature that appears on the first position of this column, and it is set to auto play after a video ends. YouTube has programmed this unit algorithmically, and the upcoming video may not be coming from the last channel.

Second challenge is the ‘Recommended’ section in the suggested videos column. Selection of this section highly depends on the personal choices of a viewer. Keeping the viewers engaged through calls to action and annotations is how YouTube secures these spots for their own videos.

Finally, it is a challenge to whether or not include videos on the ‘Exit page’. It is the last page that a viewer hits before he or she decides to leave the website. In YouTube, if one video causes several viewing sessions to terminate, then YouTube would refrain from including that video in their suggested videos list. Probably, that video was low in quality, irrelevant to the subject, or perhaps out of context, due to which the viewer decided to leave.

If you want to read similar articles to How Does YouTube Choose Suggested Videos?, we recommend you visit our Internet category.