How to build a user portrait system? Just read this one article!

Only by deeply understanding the characteristics and needs of users can we create products that satisfy users and improve user experience. Drawing a user portrait is the most basic step, which can help understand the user's basic information. So, how to build a user portrait system? This article will analyze from three dimensions, I hope it will be helpful to you.

Chopper: I am a data product manager who changed my career halfway. Big data is very popular now and my company wants to build a portrait system, but I have no idea about user portraits. I have no idea what the architecture of the portrait system is, what common functions it has, etc. It really stumps me!

Zoro smiled knowingly and said: Product managers exist to solve problems, and problems must be solved step by step. If you want to master the portrait system, you can first understand the basic content such as the label system and the OneID system.

Then let’s take a closer look at how the portrait systems of various major companies are built, such as the portrait systems of giant companies such as Weibo, Baidu Maps, JD Digits, Tencent, and Alibaba. As the saying goes, “He who seeks the best will get the middle, and he who seeks the middle will get the worst.” The right way is to take the above and combine it with reality.

Of course, this sorting is not easy and requires a lot of energy. I spent 2 months organizing it and wrote a 60-page PPT. It was a painstaking effort. I believe you will get twice the result with half the effort.

Chopper: Cool, I'll pay attention and learn from you first~

Zoro: The following structure will be used:

V1.0: Primary version of the portrait system, which is the prototype of the portrait system. It selected the early portrait versions from 2015 to 2017. At that time, the portrait systems of some companies were just starting out and various functions had not been explored clearly, such as the Baidu map portrait system and the Weibo portrait system;

V2.0: Standard version of portrait, with complete basic functions and certain analytical capabilities, such as Sensors portrait system and JD Digital portrait system;

V3.0: Portrait marketing version, which has evolved into a marketing tool to help business growth, such as Tencent Guangdiantong DMP and Alibaba Damopan DMP;

01 V1.0: Primary version of portrait

The primary version of the portrait was based on the early version of 2015, when the portrait systems of some companies were just starting out and various functions had not yet been fully explored.

For example, the overall design of Baidu Map's channel portrait system is to add some basic portrait elements, such as gender, age, industry, city, and other label data, based on the data analysis/reporting system. During this phase, efforts were made to conduct data analysis on various businesses, such as building dashboards, custom reports, and subscriptions, similar to BI platforms. However, there was a lack of label system construction and insight modules, and the functions of each portrait were relatively shallow.

As for the Weibo portrait system, the basic form of portrait has been formed, from data collection, data cleaning, label generation, to single user portrait, grouped user portrait, API interface provision, etc. Its functional modules are relatively complete, but its insight capabilities and label system construction capabilities still need to be further enhanced.

1. Baidu Map Portrait System

According to relevant reports, in 2017, Baidu Maps' market share was between 45% and 50%, failing to cover all users and its access to user data and information was restricted. Therefore, Baidu Maps has carried out in-depth cooperation with top mobile phone manufacturers such as Huawei, OPPO, vivo, Xiaomi, and Meizu, allowing their mobile phone systems to call the LBS basic functions of Baidu Maps, thereby exchanging regional data sharing resources. The user portrait system was born in this context.

This user portrait system provides strong data support in budget assessment, delivery effect analysis, cost savings, user implementation monitoring, etc.

1) System architecture

The overall architecture of the system includes:

Data layer: data collection, data cleaning, database construction, data synthesis

Management: Management modules and personal center

• Management module: role management, user rights management, report addition, deletion, modification and query management
• Personal center module: login settings, browsing settings, dashboard customization settings

Presentation layer: navigation module and browsing module.

• Navigation module: users can select their own demand report
• Browsing module: users select viewing dimensions and customize dashboard display

Extension layer:

• Favorites module: Users can collect frequently visited pages by clicking Favorites, making it easier to quickly find the reports they are interested in among many reports.
• Email settings module: Users select the core data they are interested in, and the selected data is sent to the designated mailbox in the form of reports on a regular basis every day.
• Abnormal warning module: mainly for administrators, set thresholds for reports and values, and notify administrators by email when the value exceeds the threshold, so as to detect data abnormalities in time
• Export module: Excel raw data export, chart export

2) Product features

As shown in the figure below, the channel portrait platform can be divided into business modules such as pre-installation channels, after-installation stores, manufacturer brands, and pre-installation and post-installation comparisons based on the needs and characteristics of the user groups it targets. For each channel, we can further analyze channel quality, basic portraits, and user behavior.

The basic portrait contains personal attribute data, such as gender distribution, age distribution, education level distribution, consumption level distribution, etc.

Through the city user distribution heat map, you can intuitively understand the distribution of the number of users in each city.

2. Weibo portrait system

The Weibo portrait system is based on Weibo's big data. It profiles all Weibo users and can perform data analysis on each type of user.

Weibo's portrait system mainly includes data crawling module, single portrait module, batch portrait module and query interface module.

1) Data crawling module

The data crawling module is mainly responsible for data collection and data cleaning. It collects information filled out by users on the developer platform, restricts the categories of interface calls, keeps the latest user data, cleans the data, and provides interfaces for basic user information and user relationship chains to facilitate calls from various systems.

2) Single user portrait module

The single portrait module is mainly divided into label generation, user behavior analysis, and relationship chain analysis.

Tag generation:

As the name suggests, the main function of the label generation module is to label users through data analysis, that is, user portraits. Labels are mainly divided into three categories: security labels, clustering labels, and statistical labels.

Security label: Describes whether the account is abnormal and analyzes the abnormal overview based on the pre-established security policy. The first is to analyze the entire process of the black industry chain and the characteristics of user accounts after they are stolen. The second is to analyze the user's historical behavior to determine the possibility of abnormal current user behavior.

Cluster label: The conclusion drawn after analyzing and interpreting the results of the clustering algorithm. The main algorithm used is the K-mean clustering algorithm.

Statistical labels: After statistical analysis of each indicator, statistical labels are obtained based on the distribution of users.

Zoro: It can be seen that this is an early labeling system, but from the perspective of label classification, it lacks clear classification logic.

User behavior analysis:

By observing changes in user behavior over a period of time, we can judge user status and predict future behavior. Such as user login duration, number of followers, number of fans, number of Weibo posts, number of favorites, and other indicators.

Relationship chain analysis:

A user’s relationship chain can describe a person very well, so relationship chain analysis is also the focus of portraiture. The main indicators analyzed here are the age, city, and number of followers of the user's friends.

3) Batch user portraits

Batch user portraits are mainly divided into file upload, result statistics and display modules.

File upload:

The file upload module supports users to write the user ID that needs to be analyzed in a txt file and transmit it to the backend for analysis through the front-end page. Each user ID in the file is separated by a newline separator. In principle, the size of each file does not exceed 10M.

Data statistics and display:

The main difference between batch user portraits and single portraits is that single portraits only need to describe a single user, while batch portraits require statistical analysis of multiple user information. The data statistics and display module is mainly implemented using highcharts, using bar charts, pie charts, and scatter charts for data visualization.

Independent analysis indicators: gender, number of followers, number of fans, number of Weibo posts, number of favorites…

Joint analysis indicators: number of fans, number of Weibo posts, number of favorites; number of followers, number of Weibo posts, number of favorites; number of followers, number of fans, number of favorites; number of followers, number of fans, number of Weibo posts;

Query interface:

The interface module is also very important in the portrait system. The clusters created in the portrait system can be called by various business systems in the form of interfaces.

3. Summary

The Baidu channel portrait system in this case is still relatively traditional, following the BI approach. Relatively speaking, the Weibo portrait system has gone a step further and has standard portrait modules such as single user portrait, group portrait, tag management, and interface. However, the tag system and user insights are not in-depth enough. Next, let’s take a look at the evolution of Portrait V2.0.

02 V2.0: Standard version of the image

The portrait system in the second phase has relatively complete functions and has initially been used to empower businesses, such as the Sensors Portrait System, JD Digital Portrait System, and Tencent Guangdiantong.

JD Digital's portrait system, a standard portrait system, provides relatively standard portrait functions, including label market, crowd portrait, crowd management, interface services, label management, etc., which can group users and serve other business systems.

The Sensors portrait system and analytical portrait system are based on Sensors' data analysis genes to carry out expansion related to user segmentation and portrait insights. It is essentially a data analysis tool, and its analytical capabilities can be used as a reference.

1. JD Digital’s portrait system

JD Digital Portrait System: provides unified data development and application services based on user tags and crowd portraits. Assist in the realization and service expansion of target scenarios such as marketing, analysis, identification, and value realization based on business entities.

Label Market:

All tags are displayed in the tag list. For a specified tag, you can view tag information, tag value distribution, and various tag indicator statistics.

Crowd portrait/insight:

It supports crowd quantity calculation, basic portrait analysis, custom portrait analysis, analysis of the intersection between two crowds, and analysis of the relationship between crowds and other labels.

Crowd Management:

Crowd management supports multiple ways to create crowds, and supports crowd information setting, rule modification, and crowd profiling and other operations after saving.

Interface services:

Administrators can manage users' system roles and tag usage permissions in "Permission Management"; manage system docking tags and permissions for interfaces during their term of office.

Tag Management:

Data developers and administrators have the authority to manage tags. After entering the page, they can view the status and information of the corresponding tags, and perform operations such as "testing, offline, modifying, and deleting" the tags.

2. Sensors Portrait System

The Sensors User Portrait System is a user data analysis platform launched for enterprise-level customers. It provides the ability to explore user characteristics and portraits, complete user identification, clustering and segmentation, and view the evolution of users throughout their life cycle through historical feature changes. Its main functions include the processing and production of feature labels, user characteristics and portrait analysis, and user group management.

1) User tag management

Create a tag user:

When creating tags, just create them in a way that suits the user's behavior habits. Fill in the tag name and set clear tag conditions. At the same time, you need to provide basic information of the tag when creating a tag, mainly including: tag display name, tag name, tag grouping, update method and remarks, etc.

Manage User Tags:

Supports enabling, stopping, deleting, recalculating, historical backtracking, viewing of tag calculation rules, and visualization of tag data.

2) User group management

To create a user group:

Supports creating groups based on rules or by uploading files.

Managing user groups:

3) User insight analysis

Single user portrait:

When viewing the portrait information of a single user, you can see all the attributes of the user, which is equivalent to a person's resume, right in front of you.

Group user portrait:

By selecting a control group, a comparative analysis of the profile characteristics of the two groups of users can be performed. The default setting for the control group is None. You can select user grouping or any version of user grouping for control analysis.

Check TGI, TGI (target user group index) = the proportion of the total number of users with a certain feature in the target user group to the total number of users / the proportion of the total number of users with this feature in the total number of users * standard number 100

And by selecting the user group portrait template, the portrait information pre-set by the administrator can be quickly loaded into the portrait analysis results, and the portrait information can be added or deleted based on it.

Similar people spread:

Through intelligent algorithms, we can find similar user groups with similar characteristics to the seed population, and implement targeted operation strategies for similar user groups.

The seed population can be divided into positive seed population, that is, the result of the predicted population is similar to that of the positive seed population; and negative seed population, that is, the result of the predicted population is contrary to that of the negative seed population.

4) System Settings

It includes functions such as single user portrait, group user portrait, similar group portrait template management, function permission and data permission configuration.

3. Summary

The portrait system functions of the second phase are close to perfection, the label system has begun to be built and managed independently, the analytical insight capabilities are also constantly strengthened, and it has the functions of the standard version of the portrait. However, the relationship with the business is not close enough at the current stage. If we want to give full play to the business value of the label, we need to further strengthen the relationship with the business side.

03 V3.0: Portrait Marketing Edition

With Portrait 3.0, the portrait system is more like a marketing tool. In addition to having complete basic portrait capabilities, it can also provide marketing insights that are closer to the business. Personally, I think Alibaba's Dharma Plate is more in line with the marketing version.

Tencent Guangdiantong DMP, an advertising portrait system, is a data management platform launched by Tencent. It is mainly used for intelligent advertising delivery and provides capabilities such as label market, user segmentation, and segmentation insights.

Alibaba Dharmapan DMP, a marketing portrait system, is a data management platform launched by Alimama, which provides capabilities such as label market, user segmentation, segmentation insights, and marketing academy.

1. Tencent DMP

Tencent's DMP data management platform is an important tool for advertisers to realize the added value of their own data. It can help advertisers manage their own audience in the Tencent system and more flexibly use first-party data for advertising, as well as remarketing, dynamic product advertising, conversion statistics and other applications.

Advertisers can also use data tags such as keywords and LBS to create their own target audiences. DMP's Lookalike crowd expansion function can also help advertisers find potential new users in Tencent's user system.

1) Data access

Users will perform certain actions on websites, applications, and other scenarios, such as browsing products, collecting content, and completing levels. Access behavioral data to help understand the actions users take on their website or app, and then use this data for advertising marketing campaigns.

Population extraction: Advertisers can extract user populations that meet specific conditions based on the user behavior data they upload for use in advertising targeting, etc. If an advertiser reports a behavioral data item "a user purchased women's clothing worth 300 yuan in a certain app at a certain time", then the advertiser can create a behavioral group in the DMP that "purchases have occurred in the app in the past 7 days, the product type is women's clothing, and the product price is > 200 yuan".

Conversion statistics: Advertisers can associate the user behavior data they upload with their advertisements to count the conversion effects of the advertisements.

2) Crowd management

Create a crowd:

According to user needs, the crowd is extracted or divided in different ways to create crowds of different natures.

Crowd classification can be divided into three categories: custom crowd, expanded crowd, and combined crowd. The following is divided into several subcategories.

For example, a client file refers to a file in which a client uploads the user IDs that they want to use for advertising.

Creation requirements: A crowd can only be created when the customer has the target user's QQ number, mobile phone number, IDFA, cookie, MAC address and other information.

Application scenario: You can upload the user ID and other information you want to manage to the DMP in the form of a TXT file or a compressed package. Once created successfully, these users can be expanded, combined, analyzed, and advertised on the DMP.

For example, a beauty brand uploaded a group of registered members and targeted this group with advertisements for the brand’s anniversary discount event, which greatly increased sales and member repurchase rates. When used for intelligent expansion, the high-quality first-hand data package and the created seed group can be expanded to obtain a large number of accurately positioned potential users.

Supported data types:

Crowd Management:

Users can perform operations such as screening, editing, authorization, and deletion on the crowd management page.

Operation: Click the drop-down menu to perform intelligent expansion, insight analysis, authorization, deletion, and other operations on the group.

Crowd Editing: Click on the crowd name to enter the details page, where you can edit the crowd name and description, and view the crowd's specific information, creation time, source, and operation records.

Search: Search for people by name or ID, and the group will be automatically filtered as you enter it.

Delete: Delete unnecessary crowds to release crowd quotas. Regularly deleting crowds also makes crowd management easier. For example: For some test populations, it can be deleted.

3) Insight Analysis

Each population is created under certain conditions. There are groups of people who frequently interact with certain advertisements, groups whose locations change, or groups created based on the first data provided by the customer. Insight analysis can help customers gain a more detailed and comprehensive understanding of the distribution of characteristics such as the attributes, interest categories, focus points, and regional characteristics of the population. These characteristics can be used to optimize advertising creativity, guide marketing strategies, and provide a reference for further formulation of advertising campaigns. The clearer the customer's insight into the crowd, the more accurately they can convey the brand's product information and improve delivery efficiency.

For example: Through insight analysis, users find that men account for a relatively high proportion of the population, and most of their interests and hobbies are in the fields of sports and health. Based on this information, they can modify the advertising creative to better suit user characteristics. When placing ads, they can filter out ads in categories such as women's products that may result in low click-through conversions, making the advertising more targeted. And can export and generate reports.

Drill down to see deeper insights:

4) Label Square

The Tag Square contains two modules: Tag Area and Hot Groups. The label area includes label maps, exclusive recommendations and industry recommendations.

Label map: Label index area, showing all directional labels in a tree structure. Basic demographic attributes include several major categories of labels, such as consumption status, lifestyle, and work status. Click a label to go to the next page to view the specific label structure and learn more about the label information.

To further improve operational efficiency, there is also a favorites function for storing tags that users are interested in, making it easy to quickly find the target tag. When you don’t know which tag to choose, you can check the exclusive recommendation & industry recommendation module.

As Double Eleven is approaching, you can use hot groups. Hot groups are a special area for theme tag combinations launched according to marketing scenarios. They will be combined around a series of marketing scenarios or current affairs topics in combination with industry characteristics or holiday themes.

5) Docking

By selecting the audience in the DMP, you can connect to the self-service advertising delivery system to deliver ads.

2. Alibaba Dharmapan DMP

Dharmapan, also known as DMP, is a data management platform built by Alimama based on commercial scenarios. Merchants can use Dharmapan to gain insights and analyze various groups of people and tap into potential customers; they can quickly identify target groups through the tag market and establish personalized user segmentation and precision marketing.

I have to say, Dharma Pan is very complex. It took Straw Hat Boy almost two months to study the label system, system background, and gameplay of Dharma Pan. The more I dig, the more I find. Now, Straw Hat Boy will show you the overall system structure of Dharma Pan. In the future, there will be special articles on each module such as Dharmapan label system, insights, and grouping. Friends who are interested can first follow the official account: A Data Person’s Private Land and join the group to communicate.

1) Home page

The homepage of Dharmapan mainly includes store consumption assets, where you can view the trends of all consumers, potential customers, new customers, and old customers; recommended tag usage, old customer value segmentation, store consumer turnover, etc.

2) Insights

The insight module mainly includes crowd portrait analysis, single product intelligent insight and circle of people, and store super user insight and circle of people. For example, when conducting population portrait insights, you need to select comparison groups and compare the characteristics of different groups, such as gender, age, industry, consumption level and other data.

3) Tags

As more and more merchants have personalized crowd operation demands, Dharmapan is positioned around the refined crowd-oriented middle platform. Based on deep mining and analysis of big data, it enables precise marketing for merchants at different levels through refined crowds, meets merchants' crowd demands in different marketing scenarios, and improves merchant-consumer operation efficiency.

4) Crowd

The crowd module mainly includes the crowds I created, the partner crowds, and third-party data uploads. It can manage the crowds, support the creation of groups by labels and crowd combinations, and support the creation of groups by various combinations of intersection and difference.

5) Reports

The system provides an overview of the overall Dharmapan population delivery effect. You can customize historical events to view the overall delivery effect. At the same time, you can check the average delivery data of peers at the same level to understand the comparison with peers. Check the effect of crowd delivery under different channels and view the delivery trend. At the same time, it provides crowd flow analysis based on three dimensions: advertising exposure, Dharma disk exposure, and all store consumers to help merchants comprehensively evaluate the crowd composition.

3. Summary

Through the above comparison, it can be found that the functions of the portrait system are similar, and the core lies in its labeling system, insights and marketing application capabilities. During construction, it can also be built from bottom to top according to the actual needs of the company.

Since the Dharma disk has many functions, this article does not expand on them. The next article will analyze the Dharma disk, so stay tuned~

Author: Straw Hat Boy

Source: A data person’s private land