1 thought on “Reveal the portrait portrait of Alibaba”
Ramiro
Alibaba has been exploring the new e -commerce model of Class B in the future, and has begun to build three new systems for “new supply, new links, and new marketing” since 2019. Buyers are the core of the three new systems, and the digital business system that lacks buyers’ dimensions is incomplete. The difference between the platform scene target group and the scene between the scenes is not clear. The customer base matrix is the algorithm research theme that is specially set up to solve this business pain point in the scene control and improve the distribution effectiveness of the market. At the same time, the customer base matrix is also the core data of user growth and algorithm. Given that the customer base matrix is so important and has many applications, its construction is imminent.
Alibaba is intended to make the customer base matrix a vane of the platform, so that the business is target, layered, differentiated, efficiently selected products, and operates scene operations and business operations. Optimization of algorithm models provides motivation and provides a basis for digital operations. We are mainly constructed around the four dimensions of people, goods, fields, and commercials.
The customer base matrix is superimposed. While building the target users of the scene and measuring scene differences, it can also improve the scene effect, effectively guide the target traffic, and then model the algorithm of various business scenarios. Provide underlying data foundation.
B buyers do not have basic coordinate dimensions such as category C buyers. Class B users are mostly enterprises or wholesalers. It is very important for class B e -commerce, and it is also a question that Class B e -commerce “Primary Second” has been thinking.
Since the user group B is mainly enterprises and wholesalers, how to accurately describe the customer base matrix? Procurement power is a prominent representation. The purchasing power includes the purchase amount and the purchase frequency. From the purchasing power, it can be seen that the user’s operating scale and consumption ability. Therefore, we use the purchasing power as the basic coordinate dimension and provide accurate differentiated services in layers.
The purchase amount is mainly the amount of user procurement within a certain cycle. In order to avoid the layered interference caused by the price differences of different categories, first of all, the category sets the procurement amount file, and then according to the amount of the amount, the most proportion amount is the purchase amount of the purchase amount of this user.
The purchase frequency is the purchase frequency of users within a certain period. Sort the user according to the purchase time, and then calculate the frequency of the user purchased within a certain period of time. All users are divided into high, medium, and low -grade in accordance with the proportion of Gaussian distribution, as a layered grade for the purchase frequency.
The stages including new installations, new users, low -active, medium -sized, medium and high living, high living, sleeping, loss, etc. This life cycle is mainly divided according to the user’s activity on the e -commerce platform. And some of them also incorporate some business knowledge. For example, the new installation user refers to the user who has just installed the machine, and the new user refers to the user who is within 2 orders. The low work refers to users with a monthly visit to the number of days within 2 days.
The analysis of the user’s life cycle from the transaction cycle, as shown in Figure 2, including the new installation activation user, login user, first single user, active buyer (high purchasing power buyer, potential buyer), potential buyer), potential buyer), potential buyer), potential buyer), potential buyer) At the stage of dive sleep buyers, deep sleep buyers, the conversion relationship between various life cycles is also presented intuitively in the figure. Accurate user operations adjust the goals according to the different life cycle of the buyer, and the strategy adopted will be adjusted accordingly.
Is to understand the life cycle of the user, you can do a targeted user new, promote, and reserve to improve user stickiness: for new installations and new users, mainly to improve their user experience , Cultivate users ‘consumption habits and reserve transformation; for low -living users, mainly promoting lives and retention; for medium and high living users, mainly to maintain users’ habits and strengthen stickiness; for sleeping and loss users, mainly through red envelopes Rights and interests promote work. The maintenance of the user’s life cycle plays a vital role in the continuous user growth of e -commerce.
CBU, as a typical representative of the B2B e -commerce platform, has always been committed to serving the world’s hundred million B.C. users. Users verify the identity and main category (such as imported mother and baby store owners, boutique women’s clothing shop owners, part -time jobs, small supermarket owners, etc.) as one of the core attributes of Class B user portraits, which not only represents the user’s offline entity Identity also directly affects users’ behavioral preferences, procurement cycles, and demands for merchant service capabilities on the e -commerce platform. Therefore, it has always been one of the core user portrait attributes of class B e -commerce platforms that deepen and operate.
The portrait attributes of most C user can be modeling directly based on the user’s historical behavior on the website, but the portrait of the Class B user is different. Because to verify the identity of the user’s nucleus and the requirements for accuracy of the main category, the general B e -commerce platform is mainly determined by the user’s self -filling form. The accuracy of this user’s self -filling method is high, but the position of the position, the lengthy link, and the guidance of the no benefit point not only have a low rate of user filling, but also insufficient combination with the scene. rn rn 为解决原表单式核身用户操作成本高的问题，阿里巴巴CBU电商平台通过用户核身组件借力算法模型对用户核身进行预测，依据置信度排序，为Users have launched TOP K options for users to click. The overall algorithm solution is as follows.
The on -site behavior is the first feedback base for user needs and preferences. It is the data source that the algorithm needs to be excavated. Relative to other preference portrait attributes, the user nucleus is a relatively stable and long -term user attribute. Therefore, in the algorithm application, we have selected users in the past six months as the underlying data. There are two main considerations for the selection of long -term window selection for half a year: First, the current websites are rich and high -quality, the search and recommendation algorithm is increasingly improved, and the cost of users to browse various products is lower. Maintaining focus, the needs of category B/C category B/C are mixed, the data is dirty, and the longer time window is conducive to filtering interference, capturing users with longer and stable needs; second, user behavior data, especially procurement behavior It is relatively sparse. However, the purchasing behavior of Class B users is one of the core features that reflect the identity of the user’s core, and the user procurement behavior has a certain periodic. Therefore, the long -term window can help the algorithm understand the user more comprehensively.
. Different from many preference user portrait attributes, the user’s nucleus can have a real mapping relationship with the user’s identity in reality, such as the owner of the milk tea shop — the owner of the tea shop and the bakery shop owner -Baodao Baodao The owner of Jin Dian, the owner of the boutique women’s clothing store, the owner of Taobao women’s clothing shop. Therefore, the identity mapping relationship of the upper and lower reaches of the user station can assist us to further improve the prediction of the user’s core identity and improve the coverage and accuracy.
In view of the mixed behavior of category B/C on the website, there are more noise, and the category B user’s core preference is susceptible to the interference of the popular category and commodities, so we also introduced a large number of industry knowledge As a guidance to help complete the prediction of the category B user’s nucleus, and settle the category data of the nuclear preferences based on this.
Ip using the above user station behavior, upstream and downstream identity and industry knowledge data, the algorithm can be used to achieve the prediction work of the user’s core identity through the following steps. The prediction process is shown in Figure 3.
FIG. 3 User nuclear body prediction flow chart
The seed users are mainly defined as users who have the core information of the nuclear body in the station and the mapping relationship between the upper and lower reaches of the station.
We based on the in -site behavior data of seed users in recent periods, we can dig and identify significant characteristics, provide it to operating colleagues, and make another round of seed users. Users to eliminate and optimize the selection of seed users.
The industry preference category as the threshold, screening out the products purchased by seed users in the last six months as seed products.
The I2I table based on the team precipitated, using seed products as Trigger to trigger key, expand the seed products, expand the preferences of seed products equal to the similar score of the product I2I and the Trigger seed preferences. product.
For a user’s nucleus prediction, we choose its recent six -year behavior data for modeling. Then calculate the confidence of the user’s preference for each possible nuclear identity based on the user behavior of the user’s behavior, and use it to distinguish the user’s personal procurement behavior and Class B procurement behavior to reduce the impact of the user’s personal procurement behavior on the prediction results. Increase the weight of the user B’s procurement behavior.
This is excerpted in “Alibaba B2B E -commerce Algorithm Activity” and was authorized by the publisher.
This is the summary of the 15 -year experience of B2B e -commerce in the Alibaba CBU Technology Department (1688). Alibaba B2B has experienced the upgrade iteration of information platforms, trading platforms and marketing platforms in a strategic form. This book focuses on the algorithm and technical capabilities behind the business form of the marketing platform. Business, combined with Alibaba Group’s precipitation in infrastructure and algorithm innovation to create a smart B2B business operating system.
Recommended reading
“User Portrait: Method and Engineering Solution”
Explain how to build a user portrait system from 0 to 1, and at the same time, it also gives solutions for how to use the user portrait system to drive the revenue of enterprises.
The background in the background of “Real -time Streaming Calculation” reply “0814” participation in the lottery
will send two books at 12:00 noon on August 15th to welcome everyone to participate ~
Alibaba has been exploring the new e -commerce model of Class B in the future, and has begun to build three new systems for “new supply, new links, and new marketing” since 2019. Buyers are the core of the three new systems, and the digital business system that lacks buyers’ dimensions is incomplete. The difference between the platform scene target group and the scene between the scenes is not clear. The customer base matrix is the algorithm research theme that is specially set up to solve this business pain point in the scene control and improve the distribution effectiveness of the market. At the same time, the customer base matrix is also the core data of user growth and algorithm. Given that the customer base matrix is so important and has many applications, its construction is imminent.
Alibaba is intended to make the customer base matrix a vane of the platform, so that the business is target, layered, differentiated, efficiently selected products, and operates scene operations and business operations. Optimization of algorithm models provides motivation and provides a basis for digital operations. We are mainly constructed around the four dimensions of people, goods, fields, and commercials.
The customer base matrix is superimposed. While building the target users of the scene and measuring scene differences, it can also improve the scene effect, effectively guide the target traffic, and then model the algorithm of various business scenarios. Provide underlying data foundation.
B buyers do not have basic coordinate dimensions such as category C buyers. Class B users are mostly enterprises or wholesalers. It is very important for class B e -commerce, and it is also a question that Class B e -commerce “Primary Second” has been thinking.
Since the user group B is mainly enterprises and wholesalers, how to accurately describe the customer base matrix? Procurement power is a prominent representation. The purchasing power includes the purchase amount and the purchase frequency. From the purchasing power, it can be seen that the user’s operating scale and consumption ability. Therefore, we use the purchasing power as the basic coordinate dimension and provide accurate differentiated services in layers.
The purchase amount is mainly the amount of user procurement within a certain cycle. In order to avoid the layered interference caused by the price differences of different categories, first of all, the category sets the procurement amount file, and then according to the amount of the amount, the most proportion amount is the purchase amount of the purchase amount of this user.
The purchase frequency is the purchase frequency of users within a certain period. Sort the user according to the purchase time, and then calculate the frequency of the user purchased within a certain period of time. All users are divided into high, medium, and low -grade in accordance with the proportion of Gaussian distribution, as a layered grade for the purchase frequency.
The stages including new installations, new users, low -active, medium -sized, medium and high living, high living, sleeping, loss, etc. This life cycle is mainly divided according to the user’s activity on the e -commerce platform. And some of them also incorporate some business knowledge. For example, the new installation user refers to the user who has just installed the machine, and the new user refers to the user who is within 2 orders. The low work refers to users with a monthly visit to the number of days within 2 days.
The analysis of the user’s life cycle from the transaction cycle, as shown in Figure 2, including the new installation activation user, login user, first single user, active buyer (high purchasing power buyer, potential buyer), potential buyer), potential buyer), potential buyer), potential buyer), potential buyer) At the stage of dive sleep buyers, deep sleep buyers, the conversion relationship between various life cycles is also presented intuitively in the figure. Accurate user operations adjust the goals according to the different life cycle of the buyer, and the strategy adopted will be adjusted accordingly.
Is to understand the life cycle of the user, you can do a targeted user new, promote, and reserve to improve user stickiness: for new installations and new users, mainly to improve their user experience , Cultivate users ‘consumption habits and reserve transformation; for low -living users, mainly promoting lives and retention; for medium and high living users, mainly to maintain users’ habits and strengthen stickiness; for sleeping and loss users, mainly through red envelopes Rights and interests promote work. The maintenance of the user’s life cycle plays a vital role in the continuous user growth of e -commerce.
CBU, as a typical representative of the B2B e -commerce platform, has always been committed to serving the world’s hundred million B.C. users. Users verify the identity and main category (such as imported mother and baby store owners, boutique women’s clothing shop owners, part -time jobs, small supermarket owners, etc.) as one of the core attributes of Class B user portraits, which not only represents the user’s offline entity Identity also directly affects users’ behavioral preferences, procurement cycles, and demands for merchant service capabilities on the e -commerce platform. Therefore, it has always been one of the core user portrait attributes of class B e -commerce platforms that deepen and operate.
The portrait attributes of most C user can be modeling directly based on the user’s historical behavior on the website, but the portrait of the Class B user is different. Because to verify the identity of the user’s nucleus and the requirements for accuracy of the main category, the general B e -commerce platform is mainly determined by the user’s self -filling form. The accuracy of this user’s self -filling method is high, but the position of the position, the lengthy link, and the guidance of the no benefit point not only have a low rate of user filling, but also insufficient combination with the scene. rn rn 为解决原表单式核身用户操作成本高的问题，阿里巴巴CBU电商平台通过用户核身组件借力算法模型对用户核身进行预测，依据置信度排序，为Users have launched TOP K options for users to click. The overall algorithm solution is as follows.
The on -site behavior is the first feedback base for user needs and preferences. It is the data source that the algorithm needs to be excavated. Relative to other preference portrait attributes, the user nucleus is a relatively stable and long -term user attribute. Therefore, in the algorithm application, we have selected users in the past six months as the underlying data. There are two main considerations for the selection of long -term window selection for half a year: First, the current websites are rich and high -quality, the search and recommendation algorithm is increasingly improved, and the cost of users to browse various products is lower. Maintaining focus, the needs of category B/C category B/C are mixed, the data is dirty, and the longer time window is conducive to filtering interference, capturing users with longer and stable needs; second, user behavior data, especially procurement behavior It is relatively sparse. However, the purchasing behavior of Class B users is one of the core features that reflect the identity of the user’s core, and the user procurement behavior has a certain periodic. Therefore, the long -term window can help the algorithm understand the user more comprehensively.
. Different from many preference user portrait attributes, the user’s nucleus can have a real mapping relationship with the user’s identity in reality, such as the owner of the milk tea shop — the owner of the tea shop and the bakery shop owner -Baodao Baodao The owner of Jin Dian, the owner of the boutique women’s clothing store, the owner of Taobao women’s clothing shop. Therefore, the identity mapping relationship of the upper and lower reaches of the user station can assist us to further improve the prediction of the user’s core identity and improve the coverage and accuracy.
In view of the mixed behavior of category B/C on the website, there are more noise, and the category B user’s core preference is susceptible to the interference of the popular category and commodities, so we also introduced a large number of industry knowledge As a guidance to help complete the prediction of the category B user’s nucleus, and settle the category data of the nuclear preferences based on this.
Ip using the above user station behavior, upstream and downstream identity and industry knowledge data, the algorithm can be used to achieve the prediction work of the user’s core identity through the following steps. The prediction process is shown in Figure 3.
FIG. 3 User nuclear body prediction flow chart
The seed users are mainly defined as users who have the core information of the nuclear body in the station and the mapping relationship between the upper and lower reaches of the station.
We based on the in -site behavior data of seed users in recent periods, we can dig and identify significant characteristics, provide it to operating colleagues, and make another round of seed users. Users to eliminate and optimize the selection of seed users.
The industry preference category as the threshold, screening out the products purchased by seed users in the last six months as seed products.
The I2I table based on the team precipitated, using seed products as Trigger to trigger key, expand the seed products, expand the preferences of seed products equal to the similar score of the product I2I and the Trigger seed preferences. product.
For a user’s nucleus prediction, we choose its recent six -year behavior data for modeling. Then calculate the confidence of the user’s preference for each possible nuclear identity based on the user behavior of the user’s behavior, and use it to distinguish the user’s personal procurement behavior and Class B procurement behavior to reduce the impact of the user’s personal procurement behavior on the prediction results. Increase the weight of the user B’s procurement behavior.
This is excerpted in “Alibaba B2B E -commerce Algorithm Activity” and was authorized by the publisher.
This is the summary of the 15 -year experience of B2B e -commerce in the Alibaba CBU Technology Department (1688). Alibaba B2B has experienced the upgrade iteration of information platforms, trading platforms and marketing platforms in a strategic form. This book focuses on the algorithm and technical capabilities behind the business form of the marketing platform. Business, combined with Alibaba Group’s precipitation in infrastructure and algorithm innovation to create a smart B2B business operating system.
Recommended reading
“User Portrait: Method and Engineering Solution”
Explain how to build a user portrait system from 0 to 1, and at the same time, it also gives solutions for how to use the user portrait system to drive the revenue of enterprises.
The background in the background of “Real -time Streaming Calculation” reply “0814” participation in the lottery
will send two books at 12:00 noon on August 15th to welcome everyone to participate ~