Generally speaking, it is to label users. Labeling is a highly refined feature recognition through the analysis of user information. By labeling, users can be described with some highly generalized and easy-to-understand features, which makes it easier for people to understand users and facilitate computer processing.
There are many ways to classify tags, which can be classified according to the output mode of tags, actual business or the combination of the two.
According to the output mode:
1) fact statistics labels, such as the active duration in the last 7 days, the active times in the last 7 days, etc.
2) Active consumption: fact rule labels such as the number of transactions in the last 30 days >; =2
3) model class labels, such as RFM model and AARRR model.
4) Algorithm tagging, such as judging the shopping gender and preference of a commodity according to the commodities purchased by users.
According to the actual business:
1) User Attribute Label
2) User consumption label
3) User behavior tags
4) Risk control label
. . .
In the field of Internet and e-commerce, user portrait is often used as the basic work of precision marketing and recommendation system, and its functions generally include:
1) Precision marketing: According to the historical user characteristics, analyze the potential users of products and the potential needs of users, and use SMS, email and other means to improve marketing efficiency and marketing effect for specific groups.
2) User statistics: after users are classified according to their attributes and behavioral characteristics, the number and distribution of users under different characteristics are counted; Analyze the distribution characteristics of portrait groups of different users.
3) Data mining: Building recommendation system, search engine and advertising system based on user portraits to improve service accuracy.
4) Service products: Portrait the users of the products, analyze the audience of the products, understand the psychological motivation and behavior habits of users more thoroughly, improve product operation and improve service quality.
5) Industry report &; User research: through the analysis of user portraits, we can understand the industry trends, such as the analysis of people's consumption habits and preferences, and the analysis of consumption differences in different regions.
6)ABtest: used to create ABtest experiments and analyze the experimental results.
The user portrait must proceed from the actual business scene and solve the actual business problems. The reason of user portrait is to acquire new users, improve user experience, or save lost users, with clear business goals.
The data of data source is the lowest layer of label construction, which comes from the data of various business ends, mainly including offline and real-time data sources. The general big data architecture will have a separate link of stream batch processing and an architecture of stream batch integration, so data products can be omitted.
At the beginning of the data layer, data products will pay more attention. When designing labels for data products, we need to pay attention to the circulation caliber of label production in data warehouse. Especially when defining atomic labels, they need to understand the business deeply, and understand the source, status, order channel, online and offline, order status and so on of users.
Generally speaking, the label layer will build labels according to the actual business classification mentioned above. Generally, it is enough to build atomic tags, and the tag factory in the service layer can create new derivative tags separately.
The service layer mainly includes two parts, one is the application of portrait platform, and the other is the unified API service of portrait data, which provides label grouping data support for the marketing system and advertising system at the front desk.
The above is the basic concept of user portrait system. In the next section, we will understand one of the difficulties of portrait system: how to build oneid?