Comprehensive Guide to On site Search of E-commerce Operation (II)

In the last chapter, the author gave you a preliminary explanation, On site search Some logic and page presentation of front-end and back-end manual intervention. Then, we will follow this narrative trunk to lead you to have a deeper understanding of the following: miscellaneous and trunk needs to be understood in the station search.

Basic composition of search logic in the site: how to be searched and how to rank first?

How to describe it?

For businesses, how to find out whether there is a problem, and how to rank the goods in the first place is seen by consumers to solve the problem of food and clothing. Let's talk about the two elements of search through these two questions: Index and sort.

Indexes

Search actually relies on crawlers to extract information.

What is a reptile? To be simple, it is like a little brother who helps you move information to the warehouse. The crawler is only responsible for collecting information for you, and will store it in the warehouse for you. It is arranged neatly by category. Then, you, the boss, just need to check the information.

Then, compared with the wide area search of the Internet, such as Baidu And Google. The search of e-commerce websites is much simpler and more blocked. All goods are stored in a database. Your crawler doesn't even need to work all the time every day. It only needs to run once a day. Therefore, whether a product can be captured by search is a necessary condition for showing it to users.

The index refers to the commodity database captured and sorted by the crawler.

Indexes are divided into Full Index and Incremental index Two parts.

The full index is an update of the goods in the whole station. All the changes of the goods, including on and off the shelf, and the changes of the inventory information of the name attribute tag, are updated through the full index; The incremental index is updated in real time (not necessarily in real time, but in accordance with the hourly rule). The updated content is mainly part of the attributes of goods.

The following figure shows the PPT made by a colleague of mine for the purpose of propaganda, which was not originally created by him, and the screenshot was directly reused.

What information about the goods will be included in the index?

——Product name, sales region, inventory, product attributes (size, color, style, etc.), name of the category to which the product is attached, product number, label information and price of the product, etc.

In addition to the above inherent information, commodity data information, such as sales volume, evaluation, exposure and click, will also be included in the index, which will be used to intervene in the next sorting steps and calculate the sorting score. Let's take a simple example, for example: I launched an Apple headset this morning. However, my product title and attributes do not describe the English AirPods of this headset.

Then, when users search for AirPods, will they find this product. It is obviously impossible to search, because the search is aimed at the exact matching of text information. At this time, there are two solutions.

It is a synonym for AirPods and Apple earphones.
Add AirPods keyword to the item title attribute or label. In this way, the modified product will be searched.

That is, when the title, attribute, category, label, description and other text information within the index range of the product do not include the keywords entered by the user and the synonyms in the system, the product cannot be searched by the search engine, and certainly cannot be presented to the user.

So, do you have an intuitive understanding of the concept of index? No, it doesn't matter. I won't say it again.

sort

The question of whether it can be found has been solved. The next question that businesses are most concerned about is how to make their products visible to more users in the search?

This involves the sorting algorithm of search. Each e-commerce company's search sorting is based on its own algorithm logic. The sorting algorithm logic also has its own merits and characteristics. The sorting algorithm should be the core secret of each major e-commerce in the site search.

However, the sorting factors are not so secret, and basically you want to get something like this. However, the weight of each factor is different, and the calculation method is also different.

Let's first list these factors: sales volume (the number of transactions on Taobao), click, collection, additional purchase, praise rate, product quality score (the perfection and accuracy of product information), user behavior preference (to personalize thousands of people and faces), store score and other information.

I have some sorting algorithm formulas here, but I still have a headache. You only need to know the factors that affect the sorting algorithm and which factors play a decisive role.

The following figure shows the ranking factors of my company in the initial version. What conclusions can I draw? Now many Taobao merchants have sharpened their heads to explore the sorting algorithm. In fact, there is no need to search for tens of millions of e-commerce businesses. For products and operate For example, it is enough to know what can directly promote sales and transformation.

In fact, sorting is divided into two main stages, one is the sorting of a wide range (category sorting), and the other is the sorting of the range of goods.

What do you mean? First of all, after entering a keyword, the user is likely to use an item word or brand word. When using a search on an e-commerce website, the user has a 95% probability of finding the product he wants to buy. At this point, we need to give them a collection of these goods. We call it list or category or classification.

You will generally see this module at the bottom of the home page of e-commerce apps. I will give you a more in-depth explanation of this project in the following chapters. As shown in the following figure (the part in the red box is the classification): that is to say, you should first locate the classification of the goods you want to find. Only by finding the collection category of this commodity, and then sorting the commodities in this category, is the two-step strategy of sorting - the first step is category prediction; The second step is to sort the goods.

At present, category prediction is mainly supplemented by two ways: manual and machine algorithms.

Manual means the manual background intervention mentioned in the previous chapter. At present, the main methods of machine category prediction are: category text+commodity volume clustering algorithm; Bayesian prediction algorithm; Head word category prediction algorithm. Each of these algorithms has its advantages and disadvantages. I will introduce them in detail in the next chapter on search algorithms.

Well, to summarize this chapter, the first step is to solve the problem, and the second step is to solve the problem of food and clothing. It needs to be searched before it can be seen by more users. The above content does not mean absolute accuracy. There is a personal point of view in it, so be careful!

Preview of the next chapter:

Basic materials searched in the station: catalog and commodity description
Logic and data indicators of in station search operation

#Columnist#

Author: Wang Huan, WeChat: wanghuan314400, a small piece of operation ash.

Operation article of last year and today

This article is reproduced in Wang Huan , this article does not represent the love operation position, please contact the original source for reprinting. If there are any copyright problems with the content and pictures, please contact Love Operation.