The internal mechanism of Tmall navigation reveals

Online shopping has become our habits. When we want to buy the same item, you often go to the online shopping website to enter Query to search. The page returned to the page is generally divided into two parts, the top is the navigation area, the SRP below (Search Result Page) Search results page. Two cases:

1. If the user knows the details of the goods you want to buy, he can enter a more powerful Query, find the result you want directly from the SRP page;

2. If the user doesn’t know how it wants to buy, he can only enter a simpler Query. The SRP page returned at this time is likely to not buy it, so he may click some of the recommended navigation area. Category and attributes perform the next choice to find the goods you want. Our navigation often plays a role in the second case.

Before I pick up the work of Tmall, I always thought that the category and attributes of the navigation area were stored in a table of a database. It is fixed, and the keywords according to the Query are determined which category and properties are displayed. If the Query contains the keywords, the category and properties are the same. In fact, Tmall’s navigation shows is in a smart way, and each Query display is related to its category and attribute, and sorted in accordance with certain algorithms. This article uses this as a starting point, focusing on the internal architecture and sorting algorithm of the entire Tmall navigation, suitable for newcomers who want to understand the principles of Tmall navigation.

Tmall navigation from function into category navigation and attribute navigation two types, from the way of displaying, can be divided into QP recommendations and engines recommend two categories. Before sorted, the QP recommended will be recommended by the engine, the engine is recommended, and if there is no QP recommended or QP recommended, it will be displayed at the front end, otherwise it is placed in “more”. middle. The engine makes it easy to sort the number of items on the line. The specific point is to load the baby through B2C Merger, and build an index in the Search service, and store detailed baby information in the Detail service. Sort Query’s baby number in Ranker . And our algorithm can generate a part of the QP recommended part.

The algorithm will enter the “Query, Search Category), the recommended ID list] of the recommended ID list. After loading the QP navigation module, the service is provided online, and each Query is input at the front end, by The front end calls the QP service to get the Query and the following category ID or attribute ID recommendation list, then call Forest to get the category name or attribute name corresponding to each ID, so that the front end is displayed. Specific working principle Figure 1 shows:

The way of displaying QP category navigation is divided into:


2. Direct

3. Father and child profile

4. Properties push the first class destination tile display

Specifically correlate with the type of interest in the recommended category list. The process recommended by the specific category is shown in Figure 2.

QP’s attributes are also similar, and the specific underlying data processing and output logic follow-by-category is similar, and is sorted by linear weighting in the click, transaction, and number of goods (Query, Search Category). Attribute recommendation is divided into:

1. Root type attribute

2. Property attributes

3. Three kinds of property recommendations under the leaves

If there is only one attribute value in the attribute list or the score of the attribute value exceeds the threshold, the attribute is preselected, and it is also known as the attribute. Specifically, as shown in Figure 3.

Now the entire Tmall navigation area shows PV to about 30 million per day, Tmall intelligent navigation carries the entire Tmall Search’s shopping guide. On the basis of comprehensive understanding of the current intelligent navigation, we will play in-depth navigation optimization.

Tip: This article is simultaneously published in micro signal {alibabatech}, please pay attention.