laitimes

Review and reflection on the construction of label system (middle), management and application of labels

author:Everybody is a product manager
Everyone who leaves traces on the Internet will leave some of their own labels, so how should enterprises manage and apply labels? The author combined his own experience, reviewed and thought about the construction of the label system, and summarized the management and application of labels, hoping to inspire you.
Review and reflection on the construction of label system (middle), management and application of labels

1. Review of the foregoing

In the previous article, we reviewed what is a label, what is a labeling system, and the necessity, core steps and principles of establishing a labeling system: https://www.woshipm.com/pd/5958816.html.

Here's a closer look:

Labeling: The abstract classification and generalization of a certain characteristic of a particular group or object [1]. Labels should be explainable, with specific or potential business application scenarios, such as all kinds of products on the shelves of supermarkets, books of different categories on library shelves, and the specifications and attribute labels of the goods themselves, all of which can better guide consumers to find the items they need;

Label system: A collection of labels, and this collection needs to be scientific and reasonable, manageable, maintainable, and iterative.

Preconditions for the construction of a labeling system:

  • The enterprise has sufficient content, services, or goods;
  • Enterprises have a variety of complex customers;
  • The company's business has entered a stable period, and the customer base has reached a certain scale.

Before the Internet era, enterprises/suppliers manufactured goods, services, or products, put them on the shelves after production, provided classification and query capabilities, and waited for consumers to buy them.

However, in the Internet era, consumers' attention has been taken away by various apps and websites, and enterprises that originally only did offline business have to "combine online + offline" in order to survive and develop, so as to carve up the traffic of the Internet, attract consumers' attention, and buy their products or services. As a result, the original business model of "waiting for people to buy" has gradually changed to a business model of "actively selling products/services to the target group". It's cruel, every one of us who leaves a mark on the Internet will be the business object to be analyzed in this era.

In the field of e-commerce, or content recommendation-oriented products (such as Xiaohongshu and various information flow products), its user (consumer) level is complex and diverse, its products or content are rich enough, and the enterprise business has also entered a stable period; ——Readers can think about how to build and manage the labeling system for these two businesses?

2. Urban governance business | Tag management and application

In the urban governance business, the objects of government governance are: people, places, things, things, organizations, and intelligence, but the vast majority of the goals of governance are still implemented on specific people or organizations, analyzing people's basic attributes, behavioral Xi, psychological characteristics, etc., predicting people's behavior trends, and taking corresponding decision-making plans and disposal measures.

Government departments usually have two governance paths to choose from:

1) The government department provides its own resources to actively and passively search for clues/cases within the scope of its authority and responsibility from offline/online (that is, the path (1) in the figure below);

2) The government uses third-party companies to help them achieve the corresponding business goals or improve their efficiency in achieving the business goals (i.e., path (2) in the figure below);

Review and reflection on the construction of label system (middle), management and application of labels

For model (2), some enterprises will set up corresponding departments to specialize in TOG business, and provide solutions and products and services for government departments, such as [national security], [public security], [cyberspace administration], [housing and construction], [urban management], [market supervision bureau], [financial supervision and administration bureau] and other departments, and use various intelligent systems or modules to provide some risk clues and intelligence clues to improve the work efficiency of government departments.

This requires these enterprises that do model (2) to collect and do governance analysis (and even refined governance analysis) of various entity elements scattered in "real space" or "cyberspace" or "reality + cyberspace" and their basic attributes and behavioral Xi, so as to form the final "clues"/"cases"/"intelligence" available to TO government departments. (This is the case with the two companies I've worked with, and the same is true for the same type of companies I've been dealing with for government business.)

——Whether it is facing "real space" or "cyberspace", it is complex enough, so in order to refine the governance and analysis of various physical elements on the Internet or in the real world, the establishment of a label system is naturally the only way.

2.1 Classification and management of tags

1. Why do I need to manage tags?

After the label or label system is built, the process from 0-1 is completed, and the entire life cycle of the label needs to be managed and iterated to better meet the business development needs of different stages.

2. Liaison management?

First of all, the label requirements generally come from the business side, and in general, there are mainly the following types of label requirements:

From the perspective of whether the label value changes:

1) Static tag: that is, the tag value of the tag is generally unchanged or the possibility of change is very low, such as the person's date of birth, name, etc.——Generally also called "attribute tag".

2) Dynamic tags: The tag value of the tag will change. For example, the age of the person, the weather is cloudy and warm, and the daily and monthly activity of the App.

Among the dynamic tags, there are relatively simple statistical tags, such as App daily active and website monthly active tags, and there are also tag requirements that need to be calculated by algorithm models (also known as "fitting labels" and "prediction labels"), such as judging whether there is "pornography" based on an input image, and some simple rule-matching tags, such as 3 months, if there is no operation, the user is considered to be a "lost user".

From the perspective of whether the tag value is trustworthy or not:

1) Fact tag: refers to the fact that the tag value is credible, such as manually filled in by the user and imported by external purchases;

2) Prediction label: that is, the depth Xi or machine Xi model research and judgment, there is a certain degree of credibility;

Therefore, we see that no matter how simple the label requirements are, such as the basic attribute tags of people, there are only three dimensions: name, gender, and age, and it is necessary to support the business to select target users according to gender and age, and it also needs to be managed by R&D.

Taking the above business needs as an example, R&D students need to design the database table according to the business needs, whether to use [vertical table] (vertical table is convenient to calculate, but not convenient to query) or [horizontal table] (convenient to query, but not convenient to calculate), and need to consider the calculation logic of these tags: where does the tag value come from?—— is it based on the user's form submission, or is it based on the algorithm based on the model prediction, what is the update frequency of each tag value, under what conditions is the update triggered, and so on.

With the development of business, label management tends to evolve in the following stages: 1) R&D management->2) production and research provide label tools for operators to independently obtain data >3) manufacturers provide tools, and customers can independently obtain data to achieve custom analysis;

Review and reflection on the construction of label system (middle), management and application of labels

3. What do you manage tags?

Manage labels around the whole life cycle of labels:

1) Adding, deleting, altering and checking labels;

2) Management of label calculation strategy;

3) Label application frequency management;

Review and reflection on the construction of label system (middle), management and application of labels

4. What do you use to manage tags?

Excel Online Document System?

——For R&D students, it is necessary to use databases and various monitoring logs to achieve tag management;

——For the product operation team, if the label requirements do not change frequently in the early stage, some online excel can be used to manage them to assist in recording the label change records of each student.

However, as the requirements for labels become more and more complex and complex, the operation team expects that the production and research institute can provide a [label operation management tool] to realize the addition, deletion, modification, and query of tags, the management of tag calculation strategies, and the management of tag application frequency.

2.2 Design of Tag Management Tools

This article uses the Tag Management Tool for internal operations teams as an example to design and practice products.

Business scenario: In the urban governance business/public opinion intelligence monitoring and analysis scenario, the vendor provides "clues" for government customers, mainly risk information, including sensitive information, harmful information, negative public opinion events, "violated" accounts, "violated" characters, and "violated" gangs).

Whether it is a single message (there may be only text, there may be pictures + text, there may be text + video), or public opinion events obtained through message clustering, or accounts, groups, people, and gangs, the "risk label" system used is a set. In addition, each business object has a label for other dimensions in addition to the "Risk Dimension". For example, the "urgency" of public opinion events, the risk level of "people", etc.

The first goal of the tool's product is to manage the risk tags of each business object. Risk labels, which need to be calculated by the risk model.

1) Adding, deleting, modifying, and checking risk tags, and managing operation logs;——P0

2) Development and management of risk label calculation strategy;——P0

3) Application frequency management of risk tags;——In this period, monitoring and statistics can be realized in the background;

Review and reflection on the construction of label system (middle), management and application of labels
Review and reflection on the construction of label system (middle), management and application of labels

The prototype interface is designed as follows:

Review and reflection on the construction of label system (middle), management and application of labels

The tag tree is just a form of presentation, which mainly realizes the entry, query, editing, and deletion of tags.

Review and reflection on the construction of label system (middle), management and application of labels

Due to the risk label, which needs to be calculated by the algorithmic model, the corpus needs to be trained. Operations can set keywords and corpora for each tag. (i.e., the configuration of the label calculation policy).

In the future, the following functional modules can be added to the tag management tool to make the tag tool "productized":

1) Display of each tag calculation policy, configuration of tag execution cycle, and configuration of tag task running (manual trigger/disable, scheduled trigger/shutdown);

2) Expand the type of labels, currently only algorithm model labels. Consider supporting Rule Class tags, such as the Liveness tab and Importance labels. It enables operators/algorithms to select the data of "important & active &xx risk tags" based on this to achieve targeted push;

3) Statistical analysis of the business application of each type of label and each label. - BI module;

4) changes in labels, considering whether there is a need to increase the review process;

5) Provide some commonly used label system templates in the industry (including the label system and the calculation rules of each label are preset in advance), such as the RMF user grouping system;

6) Sold together with the algorithm model self-training platform (machine Xi platform, machine Xi platform presets commonly used models such as NLP classification, text classification, NLP element extraction, NLP emotion recognition, etc.);

2.3 Application of Labels

  • Search scenario: Provide users with classified query navigation, so that users can quickly find the goods, items, books, and music they need.
  • Recommendation scenario: Combined with user tags and product tags, recommend content and products that users may be concerned about.

3. Summary of the full text

This paper first reviews what is a label and the label system, and the value of the label system, then introduces what kind of business scenarios need to establish a label system (e-commerce, content recommendation, urban governance business, etc.), and then introduces why to manage labels, who manages labels, and what to manage labels.

Appendix: References for this article

[1] Technical architecture diagram of the label system Label system management: https://blog.51cto.com/u_16213704/8832284?articleABtest=0

[2] Data analysis | Precise data management based on smart tags: https://blog.51cto.com/cicadasmile/2499769?articleABtest=0

[3] Data Label Management Architecture Storage and management of data tags: https://blog.51cto.com/u_13250/8162925

Write at the end:

When I interviewed for the product manager post and the byte content tag operation post in Shopee, I was asked why I wanted to build a label system, where I used it, how did you build it, and so on. At that time, I didn't systematically review and think deeply, so I regretted that I passed by my favorite position. Take advantage of the recent time, so hereby learn Xi review and sort it out, I hope it will also help you~

This article was originally published by @南方碟道 on Everyone is a Product Manager. Reproduction without permission is prohibited

The title image is from Unsplash and is licensed under CC0

The views in this article only represent the author's own, everyone is a product manager, and the platform only provides information storage space services.

Read on