Shulou(Shulou.com)06/03 Report--

In the era of mobile Internet, fine operation has gradually become an important competitiveness of enterprise development, and the concept of "user profile" also arises at the historic moment. User portrait means that in the era of big data, enterprises abstract the data into tags by cleaning, clustering and analyzing the massive data information, and then use these tags to specify the user image. The establishment of user portraits can help enterprises to better provide targeted services for users.

Accordingly, more and more third-party big data companies also begin to rely on their own data accumulation to provide customers with user profile services. For example, a user profile product under Twitter can conduct big data analysis of users' online and offline behavior, helping APP developers and operators to build a comprehensive, accurate and multi-dimensional user profile system. The following will take a user profile product as an example to explain in detail the technical characteristics and use value of "user profile".

The formation of user profile needs to go through four processes: data accumulation, data cleaning, data modeling and analysis, and data output. Among them, data cleaning and data modeling are collectively called data processing. After data processing, a push produces unique cold, hot and temperature data dimensions, and analyzes users' online interest preferences and offline behavior scenarios to form user portraits.

1. What technologies are used in user portraits?

In the data processing stage, the big data computing architecture of user portrait products adopts Kafka distributed publish and subscribe message system, which has the characteristics of high throughput and high stability. Data cleaning can use HADOOP and SPARK to identify the uniqueness of equipment, clean behavior data, and remove redundant data. This process supports interactive computing and a variety of complex algorithms, as well as real-time / offline data computing.

Cdn.xitu.io/2019/4/2/169dce991e7a357c?w=483&h=274&f=png&s=59750 ">

In the process of data modeling, the user profile product mainly uses the clustering (unsupervised learning) and deep learning technology in machine learning to let the model actively learn the user behavior data and judge the behavior, thus producing user tags.

After data cleaning and modeling, the user portrait will form four kinds of portraits: cold data portrait, temperature data backtracking, hot data scene and customized label.

Cold data portrait refers to the user attributes based on big data's analysis, which are relatively stable, such as the user's age, gender, permanent residence, etc. "warm data" can trace back to the recently active online and offline scenes of users, with a certain degree of timeliness. "Hot data" refers to the user's current scene and real-time user characteristics to help APP operators seize fleeting marketing opportunities. Customized tags are the combination of push data and third-party data to jointly model to get valuable feature tags.

Second, how to build user portraits?

The construction of "user profile" requires the participation of technical and business personnel to avoid formal user portraits. The process for building a user profile with a push is as follows:

(1) the design of label system. Developers need to know their own data and determine the form of tags that need to be designed.

(2) data fusion of multiple data sources. When building user profiles, a tweet will integrate its own huge amount of data as well as the APP's own data.

(3) realize the unified identification of users. In most cases, many users of APP are distributed in different account systems, and a tweet will uniformly identify them to help APP open accounts and share information quickly.

(4) the construction of user portrait feature layer. Each data is characterized.

(5) Portrait label rule + algorithm modeling. Both are indispensable, in practical applications, the algorithm is difficult to solve the problem, the use of simple rules can also achieve good results.

(6) use the algorithm to label all users.

(7) Portrait quality control. In practical applications, user portraits will have certain fluctuations. In order to solve this problem, a corresponding monitoring system has been built to monitor the quality of the portraits.

In short, the overall process of building a user profile can be summarized into three parts:

First, basic data processing. The basic data includes user equipment information, users' online APP preferences and offline scene data.

Second, the middle data processing of the portrait. The processing results include online APP preference features and offline scene features.

Third, the portrait information table. There should be four kinds of information in the table: basic attributes of the device; basic user portraits, including the user's gender, age, and relevant consumption level; and user interest profile, that is, the direction in which the user is more interested, such as whether the user prefers group shopping APP or other portraits of Haitao APP; users.

In the process of building a user profile, machine learning occupies a more important position. Machine learning is mainly used in the process of massive equipment data finishing, data cleaning and data storage.

Third, what can user portraits do?

The role of user portraits on e-commerce, news and information APP is self-evident, which can help APP to create an accurate recommendation system to achieve the operation of thousands of people.

Personalized recommendation based on user characteristics

APP operators can display different content to users through a user profile, such as gender, age, interests and other tags, in order to achieve the goal of accurate operation.

Content recommendation based on user characteristics guidance

Recommendation based on user characteristics guidance content refers to finding a user group that is similar to the target group, and making use of the behavior characteristics of the similar user group to recommend the content to the target user, as shown in the following figure:

In the process of implementing this content recommendation, similarity modeling technology plays an important role. Similarity modeling can be compared with clustering modeling, and it is a kind of unsupervised learning. It can find the features in the data, gather the data with the same characteristics in a group, and give the clustered data the same feature label. According to these feature tags, find users with these features and push the same content to them.

The advantage of this recommendation method is that it is characterized by long-term accumulation of APP, finer granularity, stronger applicability, more comprehensive understanding of users, and continuous improvement of the effect. And it can also tailor matching algorithms to APP's industry and its own needs to make recommendations more accurate.

In addition, the user profile can be customized with third-party data, and significant value and feature tags can be obtained through joint modeling. This way of tag addition can not only ensure that the content of the push is more accurate, but also greatly enhance the value of traffic.

4. How can developers access it?

There are two main ways to access the profile SDK of a Twitter user:

SDK integration: the client integrates a push user profile SDK. After initializing the SDK, it returns an ID (unique identity) to the customer. The ID needs to be submitted by the client to the client server, and then the server passes GIUID through the API API to query the user profile tag data.

API API call: the customer returns APP ID and other related information after the application name, package name and server export IP are provided. Customers can call API interface to query portrait information after integration testing according to "user profile data service interface document" and "user profile coding table".

For specific integration documentation, see the following link:

Android: http://docs.getui.com/gexiang/start/android/

IOS: http://docs.getui.com/gexiang/start/ios/

Server: http://docs.getui.com/gexiang/start/server/

To know users is to better serve users. It is the thirst of APP developers and operators for user awareness that gives rise to user profiles. APP developers only put the needs of users in the most important position, in order to better optimize the user experience and retain users. Accessing a push like SDK can not only help developers improve the efficiency of development decisions, but also help APP operators to carry out fine operations, thus improving the marketing efficiency and market competitiveness of enterprises.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.