YouTube: Enhancing the user experience

It is all about using our technology and research to help enrich people’s lives. Like YouTube – and its mission to give everyone a voice and show the world.

Our work with YouTube and engineering product teams helped improve decision -making processes, increase safety and participation, and enhance experience for all types of users.

Make pants more search

YouTube-short-shaped video shorts are watched less than a minute-more than 50 billion times a day.

Everything from the emerging K-POP stars covers local food guides, rushing to watch, and soon they make it-and it gets more popular all the time. But since the short pants are created in just a few minutes, it often does not include descriptions and addresses that make it easy to find by searching. So we presented Flamingo, our visual language model to help generate descriptions.

Flamingo analyzes primary video tires, explains what is displayed on the screen (for example, “a dog balances a group of crackers on his head”). It preserves this text as a definition in YouTube, creating clear content categories and matching user searches to improve results.

YouTube develops this technology via short pants, with video descriptions that were automatically created on all new downloads. Viewers can now find and watch more relevant videos, from a more varied collection of international creators.

Improving video pressure

The video has exploded in recent years, and with the expectation that the traffic will grow only in the future – video pressure is an increasingly urgent problem.

We have worked with YouTube to test our artificial intelligence model capabilities, Muzero to improve the VP9 coding program, which is a coding format that helps to compress and transfer video online. Then, we applied Muzero on some live traffic in YouTube.

At launch, we saw a 4 % average decrease in a variety of videos. Bitrate helps to determine the ability of computing and display the frequency range needed to play and store videos – which affects everything from the download time, to accuracy, temporary storage, and data use.

By improving the VP9 coding program on YouTube, we helped us reduce the traffic on the Internet and use the data and time needed to download videos. By improving video pressure, millions of people around the world can see more videos while using less data.

Brand safety protection

Since 2018, our cooperation on YouTube has helped educate creators about this type of videos that can earn ads revenues, and to ensure that the appropriate ads appear in the right place.

We have developed the LQM quality model with YouTube to name more accurate videos, and overlap with the advertiser’s friendly guidelines in YouTube. In addition to improving the accuracy of ads that work on videos, it helps to ensure the appearance of ads in addition to the content that follows YouTube.

By improving the way to identify and classify videos, we have improved confidence in the platform for both viewers, creators and advertisers.

Improving automatic luxury

As the way we develop and watch the video, creators began adding chapters to their videos. Their fans are easy to find the content they want – but it can be a slow process.

We have worked with the YouTube research team to develop the Amnesty International system that suggests separation segments and creative addresses on YouTube, by automatically processing video texts and audio and visual features. Through automatic luxury, viewers spend less time searching for content, and creators save time to create chapters for their videos.

Since the introduction of the feature in Google I/O in 2022, chapters were automatically created on tens of millions of videos (and count) via YouTube.

Technologies and advanced products

We are constantly looking for ways to improve alphabet products through our research in artificial intelligence.

Our cooperation with YouTube has already caused a major impact on people’s lives – and with more ongoing projects, we continue to improve the user experience.