Product matching, or back and forth with millions of products
You've definitely already done matching
We are all familiar with the good old marketplaces – online markets that sell absolutely everything: from pins to cars. Each marketplace has its own gentleman’s set: a unique name (preferably memorable), a website (where you can order goods), a mobile application (to make ordering even easier and more convenient), a logistics service (to bring the order to the buyer as soon as possible), then separated by commas .
When you go to a regular market, you hardly buy something the first time, right? You walk, bargain, try, choose the best. It's no different with online markets. Let's say you urgently need to buy a washing machine. You open the Megamarket application, enter a query into the search bar and see, for example, Indesit EWSD 51031.
As an experienced buyer, you want to see what other sellers and markets have to offer. You go to the second marketplace, enter the coveted “Indesit EWSD 51031” in the search bar.
How are things on any other marketplace?
Then, rolling up your sleeves, you get down to business. You open the characteristics of the Indesit EWSD 51031 and start comparing them with all the proposed washing machines that appeared in the search on all marketplaces. After a certain amount of time, success will inevitably await you: you will select the final proposals, compare the price, name, image, attributes and, finally, make the long-awaited order – the best characteristics for the optimal cost and on the most convenient terms for you.
Congratulations! You have just solved the problem product matching.
Product matching: what kind of beast is it?
Our today main character is the process of comparing products based on various criteria. This seems to be a case where the definition sounds deceptively simple. Let's figure it out.
Grocery – (from English “product”) – product/product/product. All that What we compare.
Matching – (from English “to match”) – compare/compare. This is what we are we do.
So, in product matching we compare products, but how do we do this?
This is the most important part – comparison/contrast is always done based on some criteria. Any product on the marketplace is a set of characteristics that define it. For example: photo, name, price, product parameters, reviews and more.
Now we put everything together and get product matching. Solving the problem of product matching means comparing the characteristics of products with each other and finding the most similar, and preferably identical, products.
Why does the marketplace need product matching?
Let's look at the example of Megamarket. There are millions of products on display here. Using the example of a washing machine, we saw that the same product is sold on different marketplaces. If we talk about Megamarket, then for its assortment there are about 100 million products on competitors’ sites that you want to find and compare in order to solve the problem dynamic pricing.
Let’s imagine that exactly the same washing machine on another marketplace will cost half as much as on Megamarket. A pragmatic buyer will go there, not to us. And we don't want to lose a client. So it makes sense to keep an eye on prices. everyone similar products on other marketplaces, and to do this they need to be found and compared.
If product matching were used only to solve the problem of dynamic pricing, it would already be critical for the marketplace. However, there are other reasons.
Millions of new sellers visit Megamarket every month. They send descriptions of new products, and it is important for the marketplace to understand: how new and unique are these products, is it necessary to create a new product card, or do we already have exactly the same product in our assortment? You can find out by comparing the new product with the current assortment. Matching helps to decide here the task of adding new product cards.
Ok, let's say we have a lot of new products. They all have some common properties, according to which they could be combined into groups (using matching). After all, when we come to the market, tomatoes of different varieties lie with tomatoes, and cucumbers with cucumbers, because similar products should be kept together. This is much more convenient for both the buyer and the seller. Finally, comparing products is useful even within your own assortment, for example, to find and remove duplicates.
Thus, we can highlight the main areas where product matching is extremely useful and can significantly increase the profitability of the marketplace:
dynamic pricing;
adding products from new sellers;
combining products into groups.
How difficult is it to match and what is needed for this?
The first thing that comes to mind is to simply take and compare each product with each. In mathematics this operation is called Cartesian product. Let's say we want to compare our current range with our competitors. We take the Megamarket assortment and make a Cartesian product of 50 million products by 100 million products.
And here we start having problems: calculating such a huge matrix will take an impressive amount of time and resources. It seems more reasonable to find a specific small number of candidates for each product and analyze only those. After all, comparing a washing machine and cucumbers doesn’t make much sense. These are too different products (although…). Comparing each product with each is too expensive and unnecessary.
Let's now see how difficult it is in principle to compare product attributes? There are two product cards in front of you, what do you think: are these the same product?
At first glance, yes. After all, the pictures are absolutely identical. But an attentive reader may notice that the products have a different color, as written in the product card. From which we can immediately conclude that you cannot rely only on images.
The attributes are almost identical, as is the name. But we see a big difference in prices. From which we can conclude that one of the goods is a replica of the other, which means that these are different goods. Of course, there are no replicas on Megamarket.
The prices are very close, the attributes are identical. But look carefully at the photo: on the product on the bottom left of the packaging it is indicated that the beans are fully roasted, and on the product on the right it is medium roasted. Yes, these are also different products. The devil is in the details. Now imagine that there are tens of millions of such products and you need to find something similar for each one.
So, let's summarize. Solving the problem of product matching is extremely important for any marketplace. The task of product matching itself is complex and multi-level. In order to competently compare products, a detailed analysis of both textual information and images is required, which leads us to the idea that we must be able to work effectively in two modalities at once. For effective matching, it is necessary to find a certain number of candidates for each product.
Let's assume that this will be enough introductory concepts for us to further delve into the technical details.
In the next part we will tell you how we solve the problem of product matching for Megamarket and circumvent the problems identified above. Don't switch!
