<Required for risk control> How to build user portraits from scratch?

Consumer Finance Risk Control Alliance Knowledge Planet data summary (updated)-20191120

Consumer Finance Risk Control Alliance Introduction & Business Cooperation

Learning materials receive in-depth analysis for free Control the application in 8 major scenarios

1. What is a user portrait

User portrait, that is, data information labeling, is through the collection and analysis of users’ social attributes, living habits, and consumption The main information data, such as behavior, is an abstract labeled target user model.

User portrait is a target user model based on a deep understanding of real users and a summary of high-precision related data. It is a virtual task image of real users in an abstract world. It is usually a tagged user model abstracted based on information such as user demographic characteristics, web browsing content, online social activities and consumption behavior.

2. What does the user portrait include? The

tag is a highly refined user identification that describes the attributes of the user, such as age, gender, region, user preference, etc. The three-dimensional "portrait" of the user can be outlined by integrating all the tag information of the user. Now the usual user profile data mainly include: demographic attributes, social attributes, behavior habits and so on.

Give an example

1. Demographic attributes: including age, gender, height, region, education, income and education
2. Social attributes: including social position, marital status, housing and vehicles, social relations, etc. Wait
3. Behavior habits: including sports, leisure travel, hotel accommodation, diet and daily life, etc.

3. How to build user portraits?

1, clear purpose
The target groups of different credit services are different, so the corresponding user portraits are also different. Therefore, it is first necessary to clarify the meaning of user portraits, construction goals and effect expectations, and then carry out targeted implementation work.

2, data collection

According to the purpose of constructing the user profile, mine and collect the required user data.

Only based on objective and true data, the generated portrait is valid. When collecting data, multiple dimensions need to be considered, such as industry data, user attribute data, user behavior data, user debt data, user repayment data, and so on.

3, data cleaning

Students who have done data modeling should know that the raw data collected and mined may contain non-target data, invalid data and false data, so the raw data needs to be filtered.

4. Data modeling (tagging)

According to the business situation, combine a variety of user characteristics together. The choice of label directly affects the richness and accuracy of the final portrait. In risk modeling, data sources are often huge, and we need to use big data analysis methods. In addition, in the process of modeling and constructing user portraits, data accuracy is very important, and the definition and scope of bad samples are needed. Strictly defined, this will have a great impact on the accuracy and recall rate through the rules. Therefore, the verification phase of the user portrait needs to verify the result data that is labeled by analysis. For fact-based (true bad samples) and unfacto-based (suspected malicious samples), a complete verification mechanism is required.

5. Generate portrait

At this step, we need to start building user portraits, using a unique iconic KEY (usually user ID) as the "user" identification to label it with various labels. Here are the user profile tags in the risk control field: whether they have been stolen, commonly used equipment, commonly used locations, consumer preferences, company profile, whether there are dangerous partners, mobile phone profile, IP profile, social work account, etc. In the above tags, it is not difficult to find that like mobile phone portraits, IP portraits may not be directly obtained from data collection. This kind of information is often collected through other technical means, such as whether the mobile phone is a cat pool number, whether Is a small number and a blank number, such as whether the IP is VPN, proxy, or yesIs it a crawler? After the data runs in the model, the final generated portrait can be displayed in visual forms such as the following figure. However, the user portrait is not static, so the model needs to have a certain degree of flexibility, which can be modified and adjusted according to the user's dynamic behavior.

Remarks: Some of the definition pictures are from the Internet. If there is any infringement, please manage by private message

Free online open class at 8 o’clock every Friday night

2019-12-06 20 :00-20:30 <风控讲堂>Wind control model optimization and suggestions, please private message

for registration: Consumer Finance Risk Control Union