How do internet recommendations and web searches work?
Netflix Recommendations
Statistics report that around 80% of the shows watched on Netflix are discovered through recommendations provided by the streaming service provider. So how does it work?
Netflix uses multiple machine learning algorithms in order to tell it's users what to watch next. Instead of relying on broad genres Netflix uses the subtle characteristics within its content.
To help understand, consider a three-legged stool. "The three legs of this stool would be Netflix members; taggers who understand everything about the content; and our machine learning algorithms that take all of the data and put things together," says Todd Yellin, Netflix’s vice president of product innovation. They look at what the user has watched recently, what they watch before and after, what they watched a year ago. The company has employees who watch every single episode of every TV show and every minute of all the movies and tag them. These tags are used and compared with the usage pattern of the user.
The data that is fed in the algorithm is of two types: implicit and explicit. Implicit is the one the user directly gives like a thumbs up, whereas explicit is the one such as binge-watching the show in two nights.
To give an example of how this data works and how Netflix uses it to improve and show users what they should watch next I give you Marvel's "The Defenders" a series based on the data of Luke Cage, Jessica Jones, Daredevil and Iron Fist. The Netflix algorithm observed that those four shows are like by a lot of users so they put to gather a brand new show and also gave special suggestions to the users who had watched even one episode of any of the above shows.
Google's Instant Autocomplete
Whenever you go to the Google search bar and start typing you will notice that Google automatically displays possible searches you might make. They are actually not suggestions but predictions (at least that's what Google says). So how does that work and why do you see the picture of Donald Trump when you google the word idiot?
What most people think (well the senators in the United States Congress do) that the employees at Google can manually modify and change certain search results (they can but only if they are paid ads).
Turns out google actually has a complex algorithm running in the background when any search takes place on the Google home page or any websites that use Google search for that matter. Google has a database of the searches that are made by actual people using the service and when you start typing the algorithm starts looking for the words you type or even words and sentences that start with the alphabets you entered. But then how does usually have the correct prediction in the list when there can be millions of possible word combinations people have used? Then algorithm then sorts the results based on your location what searches have you have made in the recent past what might be trending on the internet at that particular period of time etc.
Taking the example of Donald Trump which came up in Sundar Pichai's congress testimony where he explained it but if I try to explain it in little more layman terms(which might also give the Congress a better idea), it is basically influenced by social media. There are meme pages that make fake google results displaying pictures of Donald Trump when searched for the word idiot. That leads to people making relevant searches and many memes and small "media houses" that only have websites display that kind of information on their webpages. This is basically how things get viral. So now there are many places on the internet that have a link to Trump with the word idiot. Now to quote Pichai "We have gone out and crawled billions of webpages in our index. We take the keyword and match it against webpages and rank them based on over 200 signals - things like relevance, freshness, popularity, how other people are using it". So now that it is viral and many people are looking for that particular result the google algorithm using it AI technology figure it out and display that particular result.
Statistics report that around 80% of the shows watched on Netflix are discovered through recommendations provided by the streaming service provider. So how does it work?
Netflix uses multiple machine learning algorithms in order to tell it's users what to watch next. Instead of relying on broad genres Netflix uses the subtle characteristics within its content.
To help understand, consider a three-legged stool. "The three legs of this stool would be Netflix members; taggers who understand everything about the content; and our machine learning algorithms that take all of the data and put things together," says Todd Yellin, Netflix’s vice president of product innovation. They look at what the user has watched recently, what they watch before and after, what they watched a year ago. The company has employees who watch every single episode of every TV show and every minute of all the movies and tag them. These tags are used and compared with the usage pattern of the user.
The data that is fed in the algorithm is of two types: implicit and explicit. Implicit is the one the user directly gives like a thumbs up, whereas explicit is the one such as binge-watching the show in two nights.
To give an example of how this data works and how Netflix uses it to improve and show users what they should watch next I give you Marvel's "The Defenders" a series based on the data of Luke Cage, Jessica Jones, Daredevil and Iron Fist. The Netflix algorithm observed that those four shows are like by a lot of users so they put to gather a brand new show and also gave special suggestions to the users who had watched even one episode of any of the above shows.
Google's Instant Autocomplete
Whenever you go to the Google search bar and start typing you will notice that Google automatically displays possible searches you might make. They are actually not suggestions but predictions (at least that's what Google says). So how does that work and why do you see the picture of Donald Trump when you google the word idiot?
What most people think (well the senators in the United States Congress do) that the employees at Google can manually modify and change certain search results (they can but only if they are paid ads).
Turns out google actually has a complex algorithm running in the background when any search takes place on the Google home page or any websites that use Google search for that matter. Google has a database of the searches that are made by actual people using the service and when you start typing the algorithm starts looking for the words you type or even words and sentences that start with the alphabets you entered. But then how does usually have the correct prediction in the list when there can be millions of possible word combinations people have used? Then algorithm then sorts the results based on your location what searches have you have made in the recent past what might be trending on the internet at that particular period of time etc.
Taking the example of Donald Trump which came up in Sundar Pichai's congress testimony where he explained it but if I try to explain it in little more layman terms(which might also give the Congress a better idea), it is basically influenced by social media. There are meme pages that make fake google results displaying pictures of Donald Trump when searched for the word idiot. That leads to people making relevant searches and many memes and small "media houses" that only have websites display that kind of information on their webpages. This is basically how things get viral. So now there are many places on the internet that have a link to Trump with the word idiot. Now to quote Pichai "We have gone out and crawled billions of webpages in our index. We take the keyword and match it against webpages and rank them based on over 200 signals - things like relevance, freshness, popularity, how other people are using it". So now that it is viral and many people are looking for that particular result the google algorithm using it AI technology figure it out and display that particular result.
Comments
Post a Comment