Media

YouTube Shares Overview of its Content Recommendation Systems, and the Key Factors that Define Reach


YouTube has published a new overview of how its content recommendations system works, which is one of the central drivers of video reach and views on the platform, and may help YouTube marketers get a better understanding of what guides optimal response.

YouTube actually published a similar overview earlier in the year, as part of its ongoing effort to maximize transparency, with this new explainer giving a little more historical insight as to how its systems have evolved, and how it’s working to improve its processes.

As explained by YouTube:

Our recommendation system is built on the simple principle of helping people find the videos they want to watch and that will give them value.”

Of course, ‘value’ is a fairly vague term in social media metrics, and especially in measurement, but the idea, according to YouTube, is to show people more of what they like, based on not only their own behaviors, but other, similar users as well.

“You can find recommendations at work in two main places: your homepage and the “Up Next” panel. Your homepage is what you see when you first open YouTube – it displays a mixture of personalized recommendations, subscriptions, and the latest news and information. The Up Next panel appears when you’re watching a video and suggests additional content based on what you’re currently watching, alongside other videos that we think you may be interested in.

YouTube Recommendations surfaces

The ‘Up Next’ panel has been one of the more scrutinized elements of the platform in recent years, with some users saying that these recommendations can lead them down conspiracy-fueled rabbit holes, and even radicalize them based on the content they find.

So how might that happen?

Here are some of the key notes on exactly how YouTube’s recommendations process works.

Foundationally, YouTube’s recommendations are based on four key elements:

  • Clicks  The videos you click on provide YouTube with a direct indicator of your interest in the content. But it’s not always the thing that defines your experience. For example, you might click through on a video looking for something, then not find it in that specific clip, so that click, in itself, is not a strong indicator of what you want. Which is why YouTube also measures ‘Watchtime’ as an additional qualifier. 
  • Watchtime As it sounds, watchtime measures how long you actually watch each video you click on for, which helps YouTube recommend more specific content aligned with your interests: “So if a tennis fan watched 20 minutes of Wimbledon highlight clips, and only a few seconds of match analysis video, we can safely assume they found watching those highlights more valuable.”
  • Sharing, Likes, Dislikes YouTube also measures your share and like activity, another direct response measurement in the app. “Our system uses this information to try to predict the likelihood that you will share or like further videos. If you dislike a video, that’s a signal that it probably wasn’t something you enjoyed watching.”
  • Survey Responses Finally, and in addition to these explicit response indicators, YouTube also conducts regular viewer surveys to find out if users are having a good experience in the app. For example, if you watch a clip for 20 minutes, YouTube may ask you if you enjoyed the clip, and to give it a star rating to better guide its recommendation systems.

All of these elements you likely could have guessed would be factored in, so there’s no major insight, necessarily. Though it is also interesting to note that YouTube additionally seeks to help you find content that you might not even know exists, based on the content that other people with similar viewing profiles to you watch.

“So if you like tennis videos and our system notices that others who like the same tennis videos as you also enjoy jazz videos, you may be recommended jazz videos, even if you’ve never watched a single one before.”

That’s likely how people come across those conspiracy theory trails – you look up one video on a topic that you’re interested in, then YouTube hits you with a range of related viewing that other people have watched as a result. If you fall into the wrong viewer profile, that could lead to a host of questionable stuff – though YouTube does also note that it is working to address such recommendations, and limit exposure to what it identifies ‘low quality content’.

So what qualifies as ‘low quality’ in this context?

“We’ve used recommendations to limit low-quality content from being widely viewed since 2011, when we built classifiers to identify videos that were racy or violent and prevented them from being recommended. Then in 2015, we noticed that sensationalistic tabloid content was appearing on homepages and took steps to demote it. A year later, we started to predict the likelihood of a video to include minors in risky situations and removed those from recommendations. And in 2017, to ensure that our recommendation system was fair to marginalized communities, we began evaluating the machine learning that powers our system for fairness across protected groups – such as the LGBTQ+ community.

In addition to these, YouTube also bans content that includes false health claims (like COVID conspiracy clips), while it’s also taking more steps to address political misinformation. Some of this type of material still gets through, of course, but YouTube is working to improve its systems to ensure that such material is not recommended via its discovery tools.

A key consideration in this element relates to “authoritative” or “borderline” content.

In seeking to limit the reach of borderline clips – those that don’t necessarily break the platform’s rules, but do present potentially harmful material – YouTube uses human evaluators to assess the quality of information in each channel or video.

“These evaluators hail from around the world and are trained through a set of detailed, publicly available rating guidelines. We also rely on certified experts, such as medical doctors when content involves health information.”

To determine ‘authoritativeness’, YouTube says that its evaluators answer a few key questions:

  • Does the content deliver on its promise or achieve its goal?
  • What kind of expertise is needed to achieve the video goal?
  • What’s the reputation of the speaker in the video and the channel it’s on?
  • What’s the main topic of the video (eg. News, Sports, History, Science, etc)?
  • Is the content primarily meant to be satire?

YouTube’s evaluators assess the reputation of a channel/creator based on a range of qualifiers, including online reviews, recommendations by experts, news articles and Wikipedia entries (you can check out the full listing of potential qualifiers here).

All in all, the system is designed to utilize explicit and implicit signals to highlight more of what each person wants to see, while also filtering out the worst kinds of content, in order to limit potential harm. The actual specifics of harm are a factor in this calculation, and limiting that reach – but again, YouTube says that it is working to update its recommendation tools to ensure higher quality content, based on these qualifiers at least, ends up getting more exposure in the app.

YouTube has also shared this overview of how its recommendations algorithms have evolved over time.

YouTube Recommendations development history

In assessing the various measures from a marketing and performance perspective, the key consideration is audience response, and creating content that appeals to your target viewers.

You can measure this in your YouTube analytics, and with users able to directly subscribe to your channel, there are some strong, key indicators that you can use to assess your performance, and ensure you’re aligning with viewer interests. That will then see your content also shown to other people with similar audience traits, while ensuring that you have a good website reputation, and a strong general web presence, will also limit potential penalties in YouTube moderators’ assessment.

It’s also worth checking your content against the above listing of ‘authoritativeness’ as a quick measure that you’re adhering to YouTube’s goals.

None of these elements will guarantee ultimate reach success, but failing to tick the right boxes will limit your potential. It’s worth noting these keys, and considering each aspect in your marketing effort.



READ SOURCE

See also  Snapchat base to decline for first time as users defect to Instagram

Leave a Reply

This website uses cookies. By continuing to use this site, you accept our use of cookies.