If modern enterprises want to achieve refined operation, they must lay a good foundation for user portraits, which are an important tool for us to insight into users and enable business. Building a perfect user profile requires not only solid professional knowledge, but also a comprehensive understanding and deep thinking of the business.
Understanding user profiles
When it comes to user portraits, in terms of form, we can easily think of a summary statement describing user attributes; Perfect user portraits are not achieved overnight. They need to be accumulated and precipitated gradually with the deepening of user cognition.
one point one User portrait is the mathematical modeling of users in the real world
User profiles are derived from real data, but they are abstract mathematical modeling.
1.1.1 User portrait is derived from a large amount of data of real users
User portraits describe the data of real users, including online behavior data, offline behavior data, natural attribute data... In a specific business scenario, these data are abstracted in a data-driven way, and formally described to get a user portrait.
The premise of constructing user profiles is to analyze and mine as much data information as possible. When mining user related data, only by increasing the search scope, depth and breadth as much as possible can we build a more perfect user profile.
1.1.2 User portrait cannot be separated from the actual business scenario
Some enterprises make the mistake of putting the cart before the horse when doing user portrait projects: they have labeled hundreds of user tags on existing user data, but they do not know what appropriate business scenarios to apply these user tags to; Finally, at the end of the project, these user tags did not give full play to their value and did not have the expected positive effect on the business.
Therefore, we generally recommend that when building a user tag system, we first start from business requirements. Based on specific business scenarios, decide which tags to use to drive business progress, such as increasing revenue, optimizing service experience, and then design the overall user tag system from top to bottom, gradually refining the rules and caliber of each specific tag.
one point two A user profile is a collection of user tags
User profiles can be represented by a set of tags, and tags are symbolic representations of the characteristics of a certain class of users.
The label specifies the specific perspective of observing, recognizing and describing users
In user profiles, the smallest elements are user tags, which are symbolic representations of a series of user characteristics. Specifically, on the implementation level, a label is corresponding to a set of rules and algorithms.
There are hierarchical relationships between tags, such as first level tags, second level tags, and third level tags. Different levels of tags correspond to different business scenarios.
User tags are dynamic and associated
As time changes, human behavior and business scenarios will change, and most user tags will also change.
The change frequency of different labels is different. For example, the gender label is fixed; The age tag changes once a year; Labels such as the recent visit duration and the interval segment of the amount of goods recently purchased change more frequently, and may change once every few days.
When building the user tag system, we need to accurately identify the update frequency of each tag, and then design the overall calculation scheduling frequency of the tag background. At the same time, it should be noted that if a label changes, the associated labels should be updated accordingly.
There are multiple classification methods for user tags
Labels are classified in various ways.
According to the metadata of tags, it can be divided into three categories: the actual use scenario of business, the source type of data, and the update frequency of tags.
It can also be divided into primary and secondary, internal and external, explicit and implicit.
It can also be divided according to the types of tags, such as natural attributes, shopping behavior, behavioral attributes, offline transaction attributes and other attributes.
Basic types of user profiles
In general, the categories of tags that make up user profiles are hierarchical, and they are subdivided from large categories to small categories. Each category has different usage scenarios. Take the user profile of an Internet enterprise as an example:
The user profile is divided into six categories: user basic attributes, user association, user interest preferences, user value information, user public opinion information, and user marketing information. Here we focus on the first three categories.
The demographic labels are usually natural attributes, such as name, age, mobile phone number and operator. The labels of life information include some life related labels, such as water and electricity consumption; There are also location information classes, custom information classes, and so on.
Including family relations, social relations, corporate relations, credit information, etc., which can be obtained through formal channels in the market.
User interests and preferences
One type of tags related to user interests and preferences is the behavior itself, such as what content the user has recently browsed; The other is directly related to the actual business, such as the actual order or payment.
There will be multiple secondary classifications under the labels of these six categories, such as demography, life information, etc. The third level category corresponding to the second level label is the label used to group and layer users in practice.
How to generate perfect user profiles
3.1 Systematic and standardized production and management labels
Systematic production label
The first party data and the third party data of an enterprise together constitute the data source for generating user tags.
The first party data refers to the data that can be directly obtained by the enterprise itself. The third party data enterprises cannot directly obtain, such as the user's credit card consumption, mobile phone charges, mobile phone operators, personal travel and other data, which can be supplemented by external procurement.
After input, data sources need to be classified first, and then produce corresponding calculation rules and algorithms according to different classifications. These data sources can only be developed into labels.
The sorting process of production labels needs to go through approval, management, development, release, supplement, analysis and other links (as shown in the figure below) before it can be truly applied to business practice.
In the process of label production, because some labels are relatively abstract, Yiguan Ark will set an execution range of 0~1 to measure the execution of each label. The closer the execution degree is to 1, the more accurate the label's description of the user is; On the contrary, the closer the value is to 0, the lower the accuracy of the label description.
Standardized management label
The specification of label management is often ignored. In the process of Yiguan Ark serving customers, we found the following three types of problems:
● Some enterprises basically have no label management;
● In the process of label management, some enterprises basically only focus on the input of data sources (the red part in the figure above), which leads to the worse performance of the subsequent links;
● Some enterprises have a lot of original data and labels, but when other departments need to apply labels, they cannot find suitable use scenarios.
Labels need to be managed and maintained systematically. In this process, enterprises need to focus on the following issues:
● Which business departments are suitable for the use of labels?
● Does the label processing process meet the company's internal approval requirements?
● Does the operation of labels from development, testing to release follow scientific management methods?
In addition, data related issues in label management also need enterprises to focus on, such as:
● What is the status of the identity cycle of tag data?
● When is the validity period of label data?
● Is the label currently in a failure state?
From development to testing to release, the label system needs a set of overall management methods to operate, otherwise the company will have inconsistent original data and label corresponding caliber.
Only through standardized label management can enterprises understand the life cycle, utility period and maintenance of labels.
three point two The basis for building accurate labels: ID Mapping
In the process of producing labels, enterprises often generate a series of labels from the data of system A, and then generate a series of labels from the data of system B. However, it is impossible to connect user IDs between A and B. That is, the labels of A and B systems cannot be integrated together to describe the same user.
This kind of tag fragmentation is caused by the failure to complete the overall ID mapping.
ID Mapping is the basis for building an accurate user tag system, and it is also a challenge, because many data may have been deposited for several years, or even more than ten years, and the ID rules described by users may also have undergone some qualitative changes, and it is difficult to get through between rules.
To solve the above problems, Yiguan Ark provides the following two ideas:
Idea 1: Accurate matching and fuzzy matching
● Exact match
Exact matching can identify the unique information of the user, such as ID card number, mobile phone number, bank card number, User ID. When multiple systems have the same ID, accurate connection can be achieved.
● Fuzzy matching
Fuzzy matching involves a certain algorithm, and the basic logic is that we should associate the user's related information with the degree of approximation.
For example, fuzzy matching often uses location information related data, such as IP, longitude and latitude, WeChat, Weibo User ID, and so on. We give these data different weights. Then, according to the mutual matching relationship between the weights, the approximate values from different systems may be drawn fuzzily.
If the IP, longitude and latitude of a user's WeChat and Weibo are similar, we can infer that these two IDs are the same person by comprehensively considering multiple similar relationships.
Of course, the accuracy of fuzzy matching is not as high as that of exact matching. However, on the premise that we can't do exact matching through the unique ID, we can get through the data of some users through fuzzy matching, and then there will be a foundation for the construction of the label system.
In some common user memory, we will first collect data and information about users' APP, applets, landing pages, etc., and then test whether they can match by exact matching; If not, then perform fuzzy matching again.
Idea 2: Operation means assist in obtaining user information
When the exact matching and fuzzy matching can not achieve the desired effect, we can obtain the user's unique identity through some means on the operation side or some means on the product side, so as to serve the overall process of ID mapping.
For example, we can organize the activity of issuing coupons. Users need to fill in the mobile phone number and receive the verification code to obtain coupons, so as to obtain the unique ID of users, and finally achieve the overall data connection.
Author: AnalysysData of Yiguan Ark
This article is authorized to be released and does not represent the position of APP dry goods stores. If it is reproduced, please contact the original author.
The original link is as follows: https://mp.weixin.qq.com/s/Oyq-au777Rz-C4end15Gqw