**The Power of Product Recommendations at Amazon**

Amazon's product recommendations are a key factor in their massive revenue streams. It is estimated that nearly 35% of Amazon's revenue comes from product recommendations, with this number even higher when considering 2016 revenues of over $40 billion. This staggering figure highlights the importance of personalized product suggestions for online shoppers.

**How Amazon Uses Data to Make Recommendations**

Amazon uses two primary sources of data to make product recommendations: content-based recommendation and collaborative filtering based techniques. Content-based recommendation involves analyzing text descriptions, image descriptions, brand names, and other features associated with products to recommend similar items. For instance, if a user views a polka dot shirt, Amazon may also suggest other shirts that share similar characteristics, such as another polka dot shirt or a different design with similar patterns.

**Collaborative Filtering Based Techniques**

Collaborative filtering based techniques work by analyzing the behavior of past customers who have purchased certain items. If multiple users, say user A and user B, have both viewed and purchased item I1, item I2, and item I3, Amazon can infer that other users with similar purchasing patterns will also be interested in these items. This technique is often referred to as "user-based collaborative filtering." However, for this workshop, we will not be utilizing this data due to its restricted nature.

**Content-Based Recommendation Engine**

Given the limitations on accessing collaborative filtering based techniques, our focus will be on developing a content-based recommendation engine using text and image data readily available from Amazon's website. This approach involves analyzing the text description and images associated with products to recommend similar items. By leveraging these features, we aim to develop an effective system for recommending similar products to a given query product.

**The Task: Recommending Similar Products**

Our objective is to develop a solution that recommends similar products to a given query product. The query product can be any item on Amazon's website, and the goal is to recommend items that are similar in characteristics or features to the original query product. To achieve this, we will employ various techniques using text and images to identify patterns and relationships between different products.

In conclusion, Amazon's product recommendations are a crucial aspect of their business model, with nearly 35% of revenue coming from these suggestions. By leveraging data sources such as text descriptions, image descriptions, brand names, and other features associated with products, we can develop an effective content-based recommendation engine to recommend similar products to a given query product.

"WEBVTTKind: captionsLanguage: enso let's understand what is the problem that we are trying to solve in this workshop so the problem is to recommend similar apparel items or products in e-commerce so for those of you who don't understand what it means let me walk you through amazon.com so this is a page on amazon.com this is called a product page this is called a product page because on this page if you notice you basically have a product which which is a polka dot women's shirt here and there's a world details about this shirt right about this about this Apple item all right so apparel basically means things that we wear right so I walk you through I walk you through each of the components of this page and what we will leverage right let me explain you with with what with what you see on amazon.com so that it's much easier so whenever you go to amazon.com this is your product image right this is your product image you have multiple images of a product many times so I mean you have the same product image from multiple angles right so and you can zoom in zoom out you can get multiple sizes and things like that right so for each of the product for each of these products this is the brand-name way over this is the brand name right and this is called the title this is called the title of the product okay which in a few words explains you what it is all about look look at this it says it gives you the brand name which is it's a casual onek onek short sleeve dot pleated chiffon cupcake top right I mean it has a lot of interesting words here if you notice right so it says it's it's it's a dotted shirt pleated chiffon cupcake top right so it gives you a lot of information so the title has lot of interesting and important information right now what you have here is the price right you have price here which we will use and along with all of them you have you have other important information for example if I scroll down a little I have lot of information here which is called the product scription which is called the product description so in the product description it says what is the quality of what is the material here it says 100% polyester it says on what occasions you can use and bunch of other things right whether it's a round neck or it says pleated fron sharp sleeves the bunch of interesting keywords but remember your product description is much much longer than your title right but it gives you some interesting information about the product itself right I'm just walking you through what a typical amazon.com page looks like now look at this so acid so this is a page for this product right for this a polka dot women stop on this page if you notice there are this says these are product sponsored products related to this item related to your polka dot item right now if you look at these there are some like this image right this is also a polka dot blouse right of course it's not short sleeves it's just long sleeves this is also a polka dot but slightly different color this is black in color right so these are items so these are this is what I meant by similar items that Amazon recommends you okay so all these are similar items now let's let's go let's go to the right end because it can give you many more okay here you get some more recommendations again one of the things that you'll notice quickly is it has polka dots long sleeve this is a different this is not a no neck here right this is some this is a slightly formal ish chiffon blouse right so if you look through each of these recommendations at Amazon capes of course there are some by the way there are some non polka dot images for the example this stop right this top is not at all about polka dots even this top is not about polka dots it's neither I mean probably it is recommending this because of the round neck probably right and there there are many many parameters which Amazon uses to recommend these items right so that this is one section this is called sponsored products related to this item there is one more section here which says customers who viewed this item also viewed okay this is also these are also nations now look at this this looks very much this looks very much like a like the top that that we originally saw right but it's probably by a different brand right here is here is another top so as you keep going as you keep seeing here that many of these many of these are called product recommendations right okay you get some more product recommendations here not all of them only need to be polka dots they could this is like a small flowery pattern in here right so what what is interesting oh this is this is a large polka dot here right so what is very interesting is all these so this these two these two horizontal bars of related items are called product recommendations that Amazon caves now you might say why are product recommendations important right if I care about product recommendations why care why care about protrek munitions why care about product recommendations it's a very important question that I too had long back but on amazon.com it is estimated it is estimated it is estimated that almost 35% of the revenue 35% of revenue that Amazon makes every year is because of product recommendations so which means when when somebody is on this page they not only look at this product but they also look at but they also look at all the related products and they probably end up buying some of these related items and that generates almost 35% of revenue for Amazon and that's if you look at 2016 revenues that's more than 40 billion dollars that's more than 40 billion dollars be for BOM be for Bangalore here okay this is 40 billion dollars of revenue massive amount of revenue because of product recommendations okay so so when you have a forty billion dollar incremental revenue because of product I think most companies will care about it right so Amazon what happens is internally internally internally Amazon uses two sources of data to do product recommendations one is called content one is called content-based recommendation so content-based recommendation is basically saying that if you have if you are looking at a polka dot if you are looking at a polka dot shirt there's a very high chance that you'll you'll also like another shirt which also is like a polka dot shirt the proper different shirt right so it's using the it's using the text description and the image description it's using text and image description of products to recommend okay so it's using the content itself it's using the text content and the image content itself to recommend products for you there is an other form of data that Amazon uses which is called collaborative collaborate of filtering based techniques so collaborative filtering based techniques works as follows if you have a user you one who has seen an item I 1 and I 2 and I 3 right of course people do a lot of window shopping on Amazon right so you user you one let's say zoom checks out product by one or an item i1 and then he or she goes and checks out product I 2 and then a 3 and so on so forth now a other user you - let's assume checks out item 1 and item 3 and item 4 right the next question is imagine if I have a new user you three who is checking out item one right since item since item 3 is being checked out both by item both by user 1 and user 2 after item 1 look look at this after item 1 both users u1 and u2 have checked out item 3 right so there is a very high chance user 3 also since user 3 is on is checking out product I won there is a very high chance that they will be interested to check out I 3 which means I can recommend a 3 so I can recommend I 3 to user 3 when they are on the high one page so I 1 is the item by the way okay that's because other users who came before you 3 2 mental note went and spend some time on the product page for item three right this is called collaborative filtering unfortunately for our problem we do not have this data this data is is very closely guarded by Amazon we do not have this in our workshop we will use the text and image data we will do content based recommendation so we will do content based record recommendation not collaborative filtering based organization in content based recommendation we use a text description and the images that we can easily get from amazon.com home page so we will use so we'll use things like we'll use things like brand the image will use the descriptor we'll use the title we will use the price and other features to recommend similar products so just just to be clear here we are not doing a collaborative filtering based solution in our case of course internally internally Amazon uses both like if every every major company uses both since we do not have access to this data this data is very closely colored of absent they would never give this data out very easily but this data is much more easy to obtain from Amazon itself right so we will build a solution which is a content-based recommendation so we do a content-based recommendation engine in this in this in this workshop so what is a task let's be very clear here imagine if somebody gives me a product right I want to recommend this user similar products I want to recommend this user similar products right all these products are different all these products are different right but they are similar so all these products are similar to my initial product so let us call this product a query product right and all these are similar products all these are similar products similar products are items I will keep using the words product and item interchangeably they mean the same thing okay there are similar products to the query product to the query okay and our objective is to try multiple techniques using text and images to recommend these similar products that's the problem that that we are trying to solve as part of this workshopso let's understand what is the problem that we are trying to solve in this workshop so the problem is to recommend similar apparel items or products in e-commerce so for those of you who don't understand what it means let me walk you through amazon.com so this is a page on amazon.com this is called a product page this is called a product page because on this page if you notice you basically have a product which which is a polka dot women's shirt here and there's a world details about this shirt right about this about this Apple item all right so apparel basically means things that we wear right so I walk you through I walk you through each of the components of this page and what we will leverage right let me explain you with with what with what you see on amazon.com so that it's much easier so whenever you go to amazon.com this is your product image right this is your product image you have multiple images of a product many times so I mean you have the same product image from multiple angles right so and you can zoom in zoom out you can get multiple sizes and things like that right so for each of the product for each of these products this is the brand-name way over this is the brand name right and this is called the title this is called the title of the product okay which in a few words explains you what it is all about look look at this it says it gives you the brand name which is it's a casual onek onek short sleeve dot pleated chiffon cupcake top right I mean it has a lot of interesting words here if you notice right so it says it's it's it's a dotted shirt pleated chiffon cupcake top right so it gives you a lot of information so the title has lot of interesting and important information right now what you have here is the price right you have price here which we will use and along with all of them you have you have other important information for example if I scroll down a little I have lot of information here which is called the product scription which is called the product description so in the product description it says what is the quality of what is the material here it says 100% polyester it says on what occasions you can use and bunch of other things right whether it's a round neck or it says pleated fron sharp sleeves the bunch of interesting keywords but remember your product description is much much longer than your title right but it gives you some interesting information about the product itself right I'm just walking you through what a typical amazon.com page looks like now look at this so acid so this is a page for this product right for this a polka dot women stop on this page if you notice there are this says these are product sponsored products related to this item related to your polka dot item right now if you look at these there are some like this image right this is also a polka dot blouse right of course it's not short sleeves it's just long sleeves this is also a polka dot but slightly different color this is black in color right so these are items so these are this is what I meant by similar items that Amazon recommends you okay so all these are similar items now let's let's go let's go to the right end because it can give you many more okay here you get some more recommendations again one of the things that you'll notice quickly is it has polka dots long sleeve this is a different this is not a no neck here right this is some this is a slightly formal ish chiffon blouse right so if you look through each of these recommendations at Amazon capes of course there are some by the way there are some non polka dot images for the example this stop right this top is not at all about polka dots even this top is not about polka dots it's neither I mean probably it is recommending this because of the round neck probably right and there there are many many parameters which Amazon uses to recommend these items right so that this is one section this is called sponsored products related to this item there is one more section here which says customers who viewed this item also viewed okay this is also these are also nations now look at this this looks very much this looks very much like a like the top that that we originally saw right but it's probably by a different brand right here is here is another top so as you keep going as you keep seeing here that many of these many of these are called product recommendations right okay you get some more product recommendations here not all of them only need to be polka dots they could this is like a small flowery pattern in here right so what what is interesting oh this is this is a large polka dot here right so what is very interesting is all these so this these two these two horizontal bars of related items are called product recommendations that Amazon caves now you might say why are product recommendations important right if I care about product recommendations why care why care about protrek munitions why care about product recommendations it's a very important question that I too had long back but on amazon.com it is estimated it is estimated it is estimated that almost 35% of the revenue 35% of revenue that Amazon makes every year is because of product recommendations so which means when when somebody is on this page they not only look at this product but they also look at but they also look at all the related products and they probably end up buying some of these related items and that generates almost 35% of revenue for Amazon and that's if you look at 2016 revenues that's more than 40 billion dollars that's more than 40 billion dollars be for BOM be for Bangalore here okay this is 40 billion dollars of revenue massive amount of revenue because of product recommendations okay so so when you have a forty billion dollar incremental revenue because of product I think most companies will care about it right so Amazon what happens is internally internally internally Amazon uses two sources of data to do product recommendations one is called content one is called content-based recommendation so content-based recommendation is basically saying that if you have if you are looking at a polka dot if you are looking at a polka dot shirt there's a very high chance that you'll you'll also like another shirt which also is like a polka dot shirt the proper different shirt right so it's using the it's using the text description and the image description it's using text and image description of products to recommend okay so it's using the content itself it's using the text content and the image content itself to recommend products for you there is an other form of data that Amazon uses which is called collaborative collaborate of filtering based techniques so collaborative filtering based techniques works as follows if you have a user you one who has seen an item I 1 and I 2 and I 3 right of course people do a lot of window shopping on Amazon right so you user you one let's say zoom checks out product by one or an item i1 and then he or she goes and checks out product I 2 and then a 3 and so on so forth now a other user you - let's assume checks out item 1 and item 3 and item 4 right the next question is imagine if I have a new user you three who is checking out item one right since item since item 3 is being checked out both by item both by user 1 and user 2 after item 1 look look at this after item 1 both users u1 and u2 have checked out item 3 right so there is a very high chance user 3 also since user 3 is on is checking out product I won there is a very high chance that they will be interested to check out I 3 which means I can recommend a 3 so I can recommend I 3 to user 3 when they are on the high one page so I 1 is the item by the way okay that's because other users who came before you 3 2 mental note went and spend some time on the product page for item three right this is called collaborative filtering unfortunately for our problem we do not have this data this data is is very closely guarded by Amazon we do not have this in our workshop we will use the text and image data we will do content based recommendation so we will do content based record recommendation not collaborative filtering based organization in content based recommendation we use a text description and the images that we can easily get from amazon.com home page so we will use so we'll use things like we'll use things like brand the image will use the descriptor we'll use the title we will use the price and other features to recommend similar products so just just to be clear here we are not doing a collaborative filtering based solution in our case of course internally internally Amazon uses both like if every every major company uses both since we do not have access to this data this data is very closely colored of absent they would never give this data out very easily but this data is much more easy to obtain from Amazon itself right so we will build a solution which is a content-based recommendation so we do a content-based recommendation engine in this in this in this workshop so what is a task let's be very clear here imagine if somebody gives me a product right I want to recommend this user similar products I want to recommend this user similar products right all these products are different all these products are different right but they are similar so all these products are similar to my initial product so let us call this product a query product right and all these are similar products all these are similar products similar products are items I will keep using the words product and item interchangeably they mean the same thing okay there are similar products to the query product to the query okay and our objective is to try multiple techniques using text and images to recommend these similar products that's the problem that that we are trying to solve as part of this workshop\n"