E-Commerce Commodity Center category system

Taobao currently has more than 1 billion online products, how to accurately help users find the goods he wants. After years of exploration, Taobao through the establishment of a complete class of attribute system, and finally better solve the problem, today with you to talk about the category of Taobao property system. ▍ A little history and architecture

2003 Taobao just on-line, the volume of goods is very small, no classification. Later on the volume of Goods hundred, began to have a single-level classification of commodities, a bit similar to the current level of industry category.

When the goods tens of thousands of times, the single-level classification of goods can not meet the demand, began to have a multi-level classification, is a kind of tree. Attributes have been introduced since 06, and the merchant fills in attributes according to the property template, which allows the user to filter the items by attributes.

By the year 08, the beginning of the front and back table categories, users according to the front-end category to screen products, merchants will hang the merchandise to the backstage category, before and after the table to establish a good mapping between the tree.

Today's Taobao category attribute system is mainly composed of backstage category tree, foreground category tree, hanging in the later leaf category of the product attribute template and management before and after the mapping relationship between the category management platform, the overall structure is as follows:

As can be seen from the figure, Taobao category attribute system is a very basic data service, on the Product publishing page, the Merchant chooses backstage category to upload the product information, the details page to the user to display the product to the customer in the way of bread crumbs, in the search results page to allow users to filter products according to the foreground category. Operation students can manage the mapping relationship between the front and back categories and the Background Class object property template through a management background. ▍ Backstage Category

Background category for businesses, mainly for the classification of goods and property management. When a merchant uploads a product, it is the backstage category, as shown below:

Backstage category has the following characteristics:

Backstage category tree is the most important is the leaf category, that is, the category tree can not be divided into categories, any goods must be attached to the background leaf category.

Leaf category attached to the property template, the merchant released the product after the selection of the background category will be based on the property template, to supplement the required product attribute information, before the successful uploading of goods.

Background category is relatively stable, can not be arbitrarily deleted, leaf category can not be repeated. ▍ Front desk category

Front desk classification for users, user-friendly screening to find products, most of the time users see the category is the foreground category. The following image is the Taobao search results page of the Taobao front-end category tree:

The front desk category has the following features:

The category you see when you buy is the foreground category.

The foreground category is indirectly linked to the commodity through a background map.

The foreground category consists of the background leaf category + attributes, a foreground category can contain multiple background categories, a background class can also be mapped to multiple foreground categories.

Front-end class does not directly mount the property template, the foreground class corresponding to the property from the background of the public properties of the leaf class.

Front-desk category is flexible, can overlap, can be deleted. Why should ▍ separate the front and rear table categories?

After the introduction of Taobao's Class attribute system, perhaps you will ask why the front-end category and background categories to separate it. Directly with a set of class system is not OK. Here I would like to explain why, to 08 years, Taobao merchandise is increasing, category level is also more and more deep, buyers more and more difficult to find the goods. Taobao's small two began to constantly adjust the class attributes, the category of trees to shallow, so that the goods more easily be found by buyers. Only a set of class system in the case of small two each adjustment, the seller must follow the adjustment. If it is changed only once, the seller can reluctantly accept, but this is cyclical. For example, in the summer clothing this category may be dresses, shirts and other summer clothes, to the winter will become leggings, down jacket and other winter clothing, and sellers are aware of these adjustments reasonable, but the adjustment of the whole year, it means that he was at the end of the years in the editing category, may also have to specially raise a person there, all the time , virtually the cost of the seller's operation has increased.

In 08, a Taobao product manager from the offline retail giant Wal-Mart was inspired, he found that the Wal-Mart warehouse in the category and shelf area of the category division is separate, the warehouse inside the category is relatively stable, few changes, and the shelves of the class area will be adjusted according to the activity and season. Similarly, Taobao can also be the original set of categories of the system into the background classification and front-end classification, background classification for businesses, used to mount goods and property templates, relatively stable little change; front desk classification for users, the main convenience for users to find goods, very flexible, can be adjusted frequently. Backstage category and front-end category through mapping, a background class can be mapped to a number of front-end categories, a front-end category can also contain a number of background categories.

From a technical point of view, the foreground category is to set up a virtual category based on the background class purpose. Application of ▍ attribute system in Search

Let's take a look at how the Class attribute system can help improve the accuracy of your search, and the following diagram illustrates the whole process of user query to product search results:

The front desk category has the following features:

The category you see when you buy is the foreground category.

The foreground category is indirectly linked to the commodity through a background map.

The foreground category consists of the background leaf category + attributes, a foreground category can contain multiple background categories, a background class can also be mapped to multiple foreground categories.

Front-end class does not directly mount the property template, the foreground class corresponding to the property from the background of the public properties of the leaf class.

Front-desk category is flexible, can overlap, can be deleted. Application of ▍ attribute system in Search

Let's take a look at how the Class attribute system can help improve the accuracy of your search, and the following diagram illustrates the whole process of user query to product search results:

User input Query:iphone, Taobao category property system forecast to the most relevant to this query is the front desk category is mobile phone, and then the front-end category through the mapping to find the backstage category phone, finally found in this category "IPhone 5s local tyrants Gold" this product.

With the class attribute system, the user searches the IPhone, Taobao will not give the phone sets such accessories, because it will only go to the mobile phone category to find products.

With the class-based attribute system, users can filter products by category and attributes directly on the search results page, and find the desired product more efficiently. Taobao search can record the user search query after the class attribute filtering behavior, accurate prediction of each query's class and attribute intent, when the user searches again, you can based on the query's Class property intent, based on the query text relevance to calculate the category and attribute relevance of the product , the product containing query, which is most relevant to the user's class and attribute intention, is presented to the user first. This is also Baidu such a general search engine can not do, because it does not have the class attribute system.

With the class-based attribute system, it is also possible to optimize the search algorithm vertically. That is, in each of the fine classification, you can use different business weights and brand weights, and even can have a different sort of formula. For example, when users choose mobile phones, more consideration of the parameters and evaluation of the article, this time the corresponding ranking weight factor can be strengthened, and the choice of clothing, a beautiful main map is extremely important, the ranking factor inside the main image quality weight needs to be improved. With the class attribute system, we can also do personalized search based on class and attribute. Search engine through the user's class attributes to filter the log, you can tap some characteristics of the user, such as gender, intention price range, preference brand, preference category, etc., so that users can personalize the display. ▍ class attribute system everywhere

In addition to e-commerce, in fact, in each need to help quickly find information in the field can be seen in the category of the system of attributes. For example: Curriculum classification in the education industry, diseases in the medical Industry and hospital classification. When we look at the information that the user is looking for, it is a very common method to classify the entity to help locate the object quickly. From this point of view, Google's knowledge Graph can be seen as the e-commerce website inside the product category of the property system to the whole world of everything, it tried to use a unified network of class attributes to cover all entities, perhaps this is the generic search engine class attribute system it.