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

This article will talk about Jingdong and Taobao On site search From front end to back end logic, from operate Methodology to sorting algorithm. enjoy~

Search in e-commerce sites is designed to enable consumers to quickly locate the products they want and help websites improve their sales channels. Especially under the regular sales rhythm, the importance of search is self-evident.

For now, the more e-commerce websites with massive sku, the more they rely on search. The contribution rate of search gmv of major domestic e-commerce companies such as JD, Tmall&Taobao has exceeded 40% (iResearch&Ebon). At the same time, accurate and intelligent search on the site is also the way for e-commerce to start payment Marketing One of the important channels of.

In the next section, we will focus on all aspects of in station search, from front-end presentation to back-end logic, from operation methodology to sorting algorithm. I hope I can have a more straightforward impression on you (in most cases, this chapter will directly quote JD or Taobao app as the presentation material. After all, both of them are highly representative in China, and they are also the two best interactive experiences, of course, only in China).

Front end rendering

The front-end presentation is currently divided into the following sections:

Search Box&Default Word
associational word
Hot Words&History Search
Search Results Page

(1) Search box default words

That is, the keywords automatically given to the user in the search box after the user opens the app and enters the home page. The user only needs to click the search key on the right side of the search box to directly reach the specified search results page/activity page, as shown in the figure. The default words of the search box can be involved in personalized algorithms and can be manually maintained.

If you need to place activities, you only need to link specific keywords to activities. If the personalization algorithm is involved, the user's behavior data can be obtained to generate a thesaurus of category preferences, or the category keywords that the user belongs to after the last browsing of goods can be presented. This module is a good entrance for in station promotion.

(2) Associative words

That is, when the user starts to input keywords in the search box, the corresponding suggestions and associative words will be given below the search box to facilitate the user to click and reduce the input time and cost, as shown in the figure. This module has the function of guiding users to search. It uses the principle of text left matching to enable users to find the product keywords they want as soon as possible and reduce the cost of user input.

In the figure, JD associative words are displayed in the form of keywords+attribute tags, which is more convenient for users to click and select. Of course, associative words can also be subject to necessary manual intervention as required, in the case that the algorithm is not so perfect.

For example, Huawei Mate30 suddenly made its debut, but when searching Huawei, there was no word associated with Huawei Mate30. For marketing purposes, Huawei mate30 can be placed at the first place below the search box by means of manual intervention to facilitate users to click on it, thus achieving the purpose of diversion

(3) Hot Words&History Search

Historical search is the keyword information that users have searched in the app before. Search discovery is the keyword that e-commerce websites display to users to click according to their own needs. It can be either ordinary keywords or activity words that jump directly to the activity page. This module is also one of the important locations for guiding.

In general, e-commerce websites will add some personalized algorithms with thousands of people and thousands of faces in search discovery to improve click through rate and transaction conversion. Generally, they use the category thesaurus generated from the behavior data that users have visited before to display.

(4) Search Results Page

The search results page is the most important part of the front-end presentation optimization, and its burden is the important traffic receiving and indicator conversion area of the entire search module, including the important sub module, the commodity display area; Secondary sorting area, filtering area and top rendering area. The top display area is located between the secondary sorting and the search box. Generally, e-commerce stores will put store entrance maps of brand flagship stores or advertising maps of some categories of large-scale promotions here for drainage.

The above are the front-end rendering parts that we can see. These parts are the faces that determine whether the user will generate the next action. But behind the face is the presentation logic and algorithm that we can't see or touch. We are accustomed to making every effort to search. In fact, there are many process logic and steps in the background of e-commerce search. Sometimes it is as complex as the last math final question in the college entrance examination. Even a top eye expert must instantly "surrender" in front of this logic But the front end presentation is just a blink of an eye.

Next, we will take these invisible logic, factors and algorithms one by one and give you a direct and in-depth introduction with words that Xiaobai can understand.

Back end intervention logic and elements

When explaining the back-end logic, we still need to build on the impact on the front-end presentation. Let's start with what needs human intervention.

Back end manual intervention modules mainly include the following ten or more:

Synonym/phrase maintenance
Superior/Inferior Word Maintenance
Word Chain Maintenance
Hot search word maintenance background,
Key words: commodity manual intervention module
Image classification maintenance at the top of search results page
Keyword manual override maintenance
Keyword error correction maintenance
Associative word maintenance
Keyword Search Result Page Filter Item Maintenance
Category top intervention
Keyword ranking weight intervention
Maintenance of inactive/sensitive words
Maintenance of top graph chain of search result page
Maintenance of recommended keywords for no less results on the search results page
Display and maintenance of product waterfall parameters on the search results page
Paid search system

Let's talk about their specific functions one by one. After that, you should have an overall product framework for search.

(1) Maintenance of synonyms and phrases

That is, in our language, two words with totally different shapes and sounds may express a meaning, such as: women and women, because the information on the input end and the matching end of the search is not equal; For example, due to the limited number of words maintained in our product title, it is impossible to enter all relevant keyword information. If we maintain the title as women's high heels *******, then users can't search for the corresponding products by using the exact matching rules when entering "women's high heels", so synonyms are used to expand the search results, It can provide users with more corresponding products.

At present, there are very perfect dictionaries for Chinese and English, but with the continuous addition of network catchwords, even the rich dictionaries can not meet the demand. It is necessary to retain the manual maintenance module of synonyms to solve some unexpected situations.

(2) Superior/Inferior Word Maintenance

Its function is similar to that of synonyms, but it is more hierarchical than synonyms and is also used to expand search results. For example, when users input Glory mobile phones, Glory is actually a sub brand of Huawei. Once there are only a few Glory products on the website, in addition, users are presented with Glory mobile phone accessories. Is this conducive to transformation.

Therefore, it is necessary to maintain the superior and subordinate words of hierarchical relationship. Huawei will be maintained as the superior word of glory. Once the glory mobile phone is presented after the search for glory, Huawei will complement the rest, which enriches the search results and is more conducive to the transformation when there are fewer products. Similarly, shoes>men's shoes or women's shoes or children's shoes. The extended relationship of keyword hierarchy can also increase the query range of search

(3) Word Chain Maintenance

Correspond the word to the link. As long as the user searches for this keyword, the keyword will jump to the specified link, which is generally used to maintain the activity keyword and to channel the promotion of the activity.

(4) Hot word maintenance

More clearly, many e-commerce platforms call this module Search Discovery. This module has a typical role of drainage. Generally, active words with links and high-quality keywords with high conversion often used by users will be added to the module for users to click.

Keyword commodity manual intervention module: generally speaking, this kind of forced intervention sorting function is despised by algorithms and products, but in e-commerce companies, you can't challenge the authority of the boss with your own ideas. Generally, the rules of search operations have to bow to the face of money, and bear the humiliation and burden.

When some new products are released, or when participating in activities or commodities at its grand promotion nodes, the company will require increased exposure, including search. In search, some algorithms may not be able to achieve. At this time, the function of rudely interfering with the sorting of goods on the search results page seems like the existence of fire fighting heroes. You only need the ID sequence of the product to force these specific products to the top in specific keywords. Of course, the logic of forcing the top should conform to the most basic search rules.

(5) Image classification maintenance at the top of search results page

This function is actually a part of the filter items on the search results page, but it is only displayed in the form of pictures. In fact, this function can be automatically brought out by the algorithm, but the accuracy rate is not 100%.

Therefore, the entrance of manual intervention is a bit necessary, as shown in the figure: when a user enters a broad word, the meaning of the word includes a wide range of commodity categories, such as notebooks, which include notebooks and paper notebooks for students, while there are many sub categories in notebook computers. At this time, if the top image classification is added, the user can quickly locate the category he wants. Clicking the above image classification means searching for "notebooks" under this category, and the rendered results are also the rendered results of searching for notebooks under this category.

The function of this module is to quickly help locate words with a wide range of "model words", such as "shoes". We don't know what kind of shoes the user wants to search for. Therefore, if users are presented with image classifications at the top, they can quickly locate them in the desired category, which is more conducive to users' click and conversion.

(6) Keyword manual override maintenance

This function is very powerful and rude. I call it the tyrant function. This function can handle many emergency cases and some problems with no less results, that is, one keyword is forcibly transferred to the results page of another keyword. For example, if you search for women's sexy T-shirts and use the tyrant function to directly rewrite them as "women's T-shirts", then women's sexy T-shirts will directly display the results page of women's T-shirts.

Of course, the example I gave is not appropriate. The search still needs to restore the user's original search intention. There is really no way to use the tyrant function. This is the next best thing!

(7) Keyword error correction maintenance

This function is designed to correct the keyword input by the user and then search the module, usually in the way of algorithm and dictionary. Some keywords are not 100% correct by the algorithm, so the artificial keyword correction thesaurus is the best supplement.

(8) Associative word maintenance

As mentioned above, I will not repeat it here. However, it should be emphasized that associative words are an important part of the search method, accounting for almost 30% of the search term traffic, and some apps account for more. Therefore, with so much traffic, associative words must also be added to the entrance of manual maintenance. After all, the algorithm is not omnipotent.

(9) Category top intervention

This function is to force the top keyword search on some products, such as "notebook", but the search results page will appear paper notebook and computer notebook products, which is not conducive to user selection, so you can manually top the notebook category of products, which solves the problem. Of course, this function can also be used as a part of manual category prediction on start-ups whose category sorting algorithm is not particularly good.

(10) Keyword ranking weight intervention

It refers to that on the basis of a unified sorting algorithm, operators can customize the sorting weights of different dimensions according to different keywords to achieve a part of refined operation. However, at present, since many algorithms are deployed based on the personalized nature of thousands of people and thousands of faces to maximize the conversion of search to gmv, the function of manual intervention sorting weight has been used less, but it is still necessary to maintain the entrance of intervention to avoid the waste of human resources in secondary development.

(11) Maintenance of inactive/sensitive words

This function is intended to regulate a series of forbidden words that are not allowed to appear in businesses and advertising laws, and keywords that are explicitly prohibited in some Arab countries, such as pornography, violence and blood. In some phonetic characters, such as English, Arabic and French, prepositions, For with at by and so on. When users have keywords with these prepositions and stop words, the word segmentation system will automatically filter them out of the keywords recalled by the index. Avoid affecting the display of search results.

(12) Maintenance of top graph chain of search result page

The location is shown in the figure, and its styles can be varied. At present, mainstream e-commerce stores usually place picture links of brand flagship stores here, and marketing map chains can also be placed in some category words to channel corresponding activities. Therefore, this part is also realized by the operation background of manual intervention.

(13) Maintenance of recommended keywords for no less results on the search results page

This function aims to display the display interface presented to the user when there is no corresponding product or there are less than 4 or 8 corresponding products after the user enters the keyword to search. The presentation style can be varied, which can be keyword display after automatic segmentation of algorithm, or similar product display (less results); It can also be similar to the meaning of the keywords entered by the user or related keyword recommendations for the user to search again.

The following figure is a specific example of JD: the keywords entered are the no results page after JD search, and the full word matching has no results. After the algorithm subtracts words, two recommendations are identified for consumers, and one of the keyword search results page is selected by default to display to users, while the other is reserved for users to choose. This is more conducive to the conversion of no less results and prevents users from jumping directly.

(14) Display and maintenance of product waterfall parameters on the search results page

As shown in the figure below, when the user enters the search results page and browses the products on the page, the products slide upward with each other, indicating that the user has not seen the products he is satisfied with. At this time, it is necessary to give users more accurate positioning to ensure that users will not directly miss after a long time of browsing. Therefore, the core parameters of commodities are interspersed between waterfalls, which is conducive to more accurate filtering after users click, and increase turnover.

Of course, this module is not only in the form of parameters, but also can be filled with activities or channel banners as part of the drainage. Parameters are generally automatically generated by the algorithm, and can also be maintained manually.

(15) Payment system

As the system involves the search marketing background, it is equivalent to Taobao's through car or drill show. The system is designed for e-commerce suppliers and open platform merchants, aiming to let them pay for advertising space on the system, which can also be regarded as the pure profit of e-commerce.

Generally, the system is set with administrator account and merchant platform account. Merchants can properly recharge and purchase advertising space according to their own marketing rhythm. Administrators manage accounts and track data statistics according to platform operation rules. I will give a detailed introduction and interpretation in the following article.

Well, having said so much, you may need to digest it carefully, but none of the above arguments is very accurate and should be carefully adopted. The following chapters will continue to be updated within one week, and the specific contents of the next chapter will be previewed:

Basic composition of search logic in the site: how to be searched and how to rank first?
Basic materials searched in the station: catalog and commodity description
Logic and data indicators of in station search operation
On site search background word segmentation and index
How to sort logic from 0 to 1 to 100
Specific composition and product construction of payment system

I always wanted to finish writing all the points related to the search on the site, but I didn't have too much experience to put on the codeword because of my hard work. I haven't made a summary on this platform for a long time. Let's start with these. At the end of Buddhism, typing is painful.

#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.