SEO For YouTube: How To Classify YouTube Videos

YouTube has become the first search engine for information, after Google, of course.

YouTube is one of the online channels that currently has the most activity, along with Instagram and Facebook. And it is also a platform that helps a lot to publicize a brand in addition to being used to generate trust.

Therefore, it is so important to know how the YouTube algorithm works and thus be able to apply a correct SEO to our videos.

In this article, I want to review one of the Google patents that exist on the optimization of videos uploaded to YouTube and how this information could help us with our SEO strategy for videos.

According to published studies on the type of information that Internet users currently prefer, YouTube or video content has been positioned as the first option.

In addition, an interesting fact for all users who want to sell services or are the webmaster of an online store is that video can make the difference between buying or not doing it.

It is the zero moments of truth or known as ZMOT, which is defined as that single moment when the decision to buy is made. Therefore, video is an excellent way to “convince” the user on the one hand and to earn money on the other.

The avalanche of YouTubers that is coming out is nothing more than the result of videos being consumed at high speed by users of all ages.

Any topic that comes to mind or questions about something, we will surely find an explanatory video on YouTube that can provide an answer.

YouTube Trend Analyst Earnest Pettie explains 2019 YouTube user trends:

If users want to know about the stars on the big screen, brands must find new ways to collaborate with and present celebrities.

The support of a famous person for a product is much more valuable if that person creates videos for their fans to see. It’s what Rihanna does for her Fenty makeup brand with Tutorial Tuesdays videos.

Earnest has also given examples and clues as to why and how to use YouTube, with an example from Will Smith:

Will Smith is a good example. In 2018 we did not see him in any movie, but he was constantly present with the videos he shared on YouTube about his life and leisure.

Related: Best YouTube Tools To Boost Views

The Hegemony Of YouTube 

About a month ago. I came across a complaint from some SEOs on Twitter about why Google only took video content from its YouTube platform and did not give the option to output videos from other high-quality video platforms.

Although the answer seems obvious, I do not think it is too accurate on the part of google.

To just give a couple of examples:

Daily Motion has 300 million users worldwide who watch 3.5 billion videos every month.

Related: YouTube Channel Name Ideas

Vimeo is a high-quality video platform that has more than 100 million unique visitors per month. Therefore your content could appear on Google along with YouTube. Don’t you think that?

What Everyone Says About SEO On Youtube

If you are looking for how to position a video on YouTube, probably all the contents you find have the same advice in common, such as:

• Place the keyword within the title
• Create a long description of the video
• Place keywords within the description 
• The video file must contain the keyword
• Upload a cover image with the keyword as the filename
• Create a multitude of tags (as if they were keywords).
• Create internal links from the video description to the video itself.
• Make link-building to videos.

Well, they are not really bad tips to position a video or a YouTube channel, but they are scarce or shallow.

To position videos on YouTube, you need something else. You need to know how the YouTube algorithm works.

That is the system that can really improve the ranking of a video channel within YouTube.

Related: Tools to Perfect Your YouTube SEO

Google Patent On Videos

This 2011 google patent describes how the search engine can improve video indexing by identifying and indexing images and audio clips associated with specific keywords in videos.

The YouTube algorithm needs more than just a title and description to understand the content of a video.

In addition, on many occasions, users upload videos with hardly any description or with a title that does not even include a keyword.

Another problem the YouTube search team faced was ” semantic variations ” of the tags.

To do this, the YouTube team had to implement a LogitBoost- based latent learning framework.

Another added problem can be when very long videos are published, in which cases it can be more difficult to identify all the content of a video through a title or description.

The YouTube algorithm “mistrusts” the textual content surrounding the video, and then uploaded thumbnail may not correspond too much with the content of the video then …

Related: Top 15 YouTube Rank Checkers

What Elements Does Youtube Take Into Account To Classify The Videos?

It may be easier for a search engine to be able to extract information from the frames and from the audio to know the real content of a video, and based on this information, decide which search terms are relevant to classify that video.

The patent mentions:

The video hosting system uses a machine learning process to learn a keyword and feature model that associates functions of the media content from a training data set tagged with keywords that describe its content. The system uses the learned model to provide relevant video search results for a keyword query based on the characteristics found in the videos.

There is some data associated with this patent that can help to understand the algorithm behind YouTube and thus be able to better classify a video or a part of the video for a specific type of query.

The video hosting system uses a machine learning process to learn a keyword and feature model that associates functions of the media content from a training data set tagged with keywords that describe its content. The system uses the learned model to provide relevant video search results for a keyword query based on the characteristics found in the videos.

Related: YouTube Keyword Research Tools

SEO Strategies That Work 

The YouTube process to know the relevance of a video.

According to this patent, the process they follow to identify the relevance of a video is the following:

• The video hosting system receives a keyword search query from a user.
• Select one or more videos with relevant content for the keyword query.
• Selects a frame of the video as representative of the video content using a video index that stores the keyword association scores between frames of a plurality of videos and keywords associated with the frames.
• YouTube presents the selected frame as a thumbnail for the video.
•  YouTube receives a tagged training data set that includes a set of media items (for example, images or audio clips ) along with one or more descriptive keywords for the content of the media items.
• The system extracts the characteristics that characterize the content of the multimedia elements.
• The video hosting system finds and presents search results based on the actual content of the videos rather than relying solely on textual metadata.
• A video annotation engine generates feature vectors from sampled portions of the video (for example, video frames or short audio clips ) in the video database.
• A video annotation engine applies a learned keyword model to the extracted feature vectors to generate a set of keyword scores.
• The learning engine uses machine learning to train the keyword function model that associates characteristics of images or short audio clips with descriptive keywords of the audio or visual content.
• The engine displays a list of relevant videos for the query with thumbnails.

As we have seen, the video learning engine extracts a set of characteristics from the tagged training data (images, video, or audio) and analyzes the extracted characteristics to determine statistical associations between particular characteristics and the tagged keywords.

A clear example can be found already available in the organic results. Google is able to identify what is inside a video according to a query made within its search engine.

Related: How To Start A YouTube Channel

For instance:

A-frame in which a person appears speaking on a stage can be associated with keywords such as the name of that person, the name of the conference, or the topic of the conference.

The CTR in The Videos

The percentage of CTR that a video receives on a specific query is also stored and tagged in a database as a positive signal.

However, the CTR can also be a negative signal if a video is displayed for a query and receives no clicks.

Audio 

The algorithm has a feature extraction module that segments audio clips into “short sounds” and extracts functions for the sounds.

As with training images, the feature extraction module applies a tutorial to identify a subset of audio characteristics that make training audio clips more efficient.

For audio training data, the feature extraction module 220 can generate audio feature vectors by calculating Mel frequency cepstral coefficients (MFCC).

These coefficients represent the short-term power spectrum of a sound based on a linear cosine transform of a logarithmic power spectrum on a non-linear frequency scale.

At this point, the algorithm model can be understood to express the underlying physical relationship in terms of the keyword co-occurrences and the physical characteristics that the images/audio files represent (eg, color, texture, frequency information).

Long Videos

When YouTube receives a long video upload, as can happen a lot now with online game videos where the video can be several hours long, the search engine must be able to find relevant content within those long videos for specific queries.

Related: Ways To Use Animated Videos

The search engine must identify keywords within the video by joining image and audio files.

How The YouTube Video Recommendation System Works

The YouTube video recommendation system works by following a two-stage funnel through two neural networks.

The Recommended Video System On YouTube

The videos that appear recommended to us use these neural networks that classify them in a personalized way, for example, according to the similarity that a video has with those that we have already viewed previously.

YouTube is “learning” ( Machine Learning ) about the videos we view and is capable of detecting which videos we “pass by” and which we do not in order to continue learning about our tastes and improve the recommended videos.

Also, the YouTube algorithm will always recommend new or novel content. Therefore, the freshness of interesting published content works well as a ranking factor.

Guillaume Chaslot, an ex-Google worker, explains that:

We designed the AI YouTube to increase the time people spend online because it carries more ads … If, for example, we see that the theme ‘ the earth is flat ‘ keeps users more time online than ‘the earth is round ‘, the first theory will be favored by the recommendation algorithm”.

Related: Video Marketing Trends to Follow

Conclusion

The YouTube search engine goes beyond using only the metadata that the user can manipulate or enter when uploading a video to the platform.

Somehow YouTube needs to identify the real content that is inside a video and not just is guided by the tags or descriptions, since many times these are duplicated.

As we see through this patent, the YouTube search engine uses a more advanced system to be able to understand the content of the videos and classify them in its ranking.

Of course, metadata like title, tags, clicks received, views, comments, or meta description works, but keep in mind that they are only part of the score that YouTube gives to a video.

Perhaps insisting on having a clear audio file and speaking keywords during the video stream will help.

Finally, If you have any suggestions if you have any advice, and if you have your own success stories, kindly write them down in the comment section below. We would love to hear what you want to share with us. Keep the good work up.