A patent filed by Google discloses use of machine learning to deliver better search results to the YouTube users.
“I’m on a roller coaster that only goes up, my friend”, says the protagonist to the girl. Unfortunately, this is the only information I have of the video I watched at my friend’s place yesterday.
And now, I want to watch that video on YouTube, but unfortunately I can’t.
I can also imagine you in a similar situation, when you forget the title of a video, and only remember some other related information only; like a small portion of the lyrics.
Then, when we insert that bit of information into the search box of YouTube, and hit the enter button, the probability of getting the spot on result always use to be less than 1.
But, why does this happen, you ask? Because YouTube doesn’t bring its search results based on what the contents of a video are.
Presently, YouTube indexes its videos by using the textual Meta data, like names and description provided by the uploader. Sometimes this method doesn’t perform well while searching for a scene that lies somewhere in the middle.
Relevant Read : How Google Detects Fake Names
The other problem is with the relevancy of search results for when you want to watch something in the middle of a long video clip. In such cases, even if a result seems familiar, you still have to skip a few parts of video to watch the portion you’re interested in, which again is not a guaranteed solution.
This problem becomes even more intricate when YouTube hurls hundreds of results for an entered query.
Allow us to break the good news. Google wants to kill this pain by adding an extra intelligence to the YouTube’s search engine, say a recent patent application.
YouTube, as per the patent, will use machine learning to identify the content of a video and to assign relevant search keywords. Feature-keyword is the name given to this machine learning model.
After rolling this machine learning algorithm, YouTube will start indexing the videos, using their audio and visual content besides their titles and descriptions. This will let YouTube recognize each video with its actual content rather than just the textual data.
Not only this, the machine learning model will determine and display thumbnail images representing the content of the video. This means you don’t have to open every video to check whether the content that you were searching for exists in the video or not. The relevant thumbnail will help you determine this.
This will increase the search results of YouTube in a drastic way. So, in the future, if you search for a particular term like “Sit Josephine in my flying machine”, YouTube search will no longer be bounded to just description and title.
The machine leaning algorithm of YouTube will match your term with the Meta data created by audio and video content. Now, wherever in between any video, the line or voice comes “Sit Josephine in my flying machine”, it will be indexed in the search results.
Moreover, to make the search results easy as pie, the thumbnails of videos in the results will also resemble the scene matching with your entered keyword. This will help you to recognize the correct result without actually playing the video.