Data Label Classification, Design & Implementation: From Attribute to Algorithmic Tags
This article explains data label concepts, three-tier classification (attribute, statistical, algorithmic), Huawei's fact/rule/model framework, design steps (object selection, relationship mapping, category design), and implementation via vertical vs horizontal fusion tables for scalable data platforms.
Data labels sit between the data warehouse and data marts, serving as preparation work that makes subsequent data analysis deeper, more convenient, and faster to deliver.
What Is a Data Label
A label consists of a tag and a tag value attached to a target object. The practice originated in the internet domain for precise marketing, targeted push, and differentiated user experience, then expanded to industry domains for strategic tiering, intelligent search, operational optimization, precise marketing, service optimization, and smart operations.
Classification of Data Labels
Three-Layer Classification (Shallow to Deep)
Attribute Labels : Direct depiction of business entity attributes — e.g., enterprise type, industry, business scope, region; user gender, age group, occupation, height, weight; invoice category, proxy-issued invoice, voided invoice, abnormal invoice. These come directly from fields or via simple conditional judgment.
Statistical Labels : Aggregated measures of a business entity from a certain dimension — e.g., enterprise monthly operating performance, monthly growth, quarterly growth, transaction volume of top-N customers or suppliers. They reflect the true operating condition.
Algorithmic Labels : Characteristics inferred through algorithms. Simpler ones include enterprise industry status, transaction success rate, customer acquisition ability, customer loyalty, enterprise growth score. More advanced ones apply data mining algorithms — e.g., inferring user gender, occupation, interests, shopping habits, pregnancy, or children from recent purchases for precise marketing and recommendation.
Huawei's Three-Type Framework
Fact Labels : Objective, static entity attributes — e.g., whether a part is purchased or not, whether an employee is male or female. Sourced directly from entity attributes.
Rule Labels : Results of attribute-measure combination via judgment rules — e.g., whether goods are overweight, whether a product is a hot-seller. Relatively objective and static.
Model Labels : Business-value-oriented evaluation and prediction — e.g., consumer phone-replacement potential (high/medium/low). Generated from attributes plus algorithms; subjective and dynamic.
Data Label Design Process
Determine Label Objects : Objects can be people (individuals/groups), things, or relationships — users, enterprises, orders, invoices, invoicing behavior, supply-sell relationships. Choice is driven by analysis interest points: precise marketing needs user shopping preferences; anti-fake-invoicing risk control needs enterprise invoicing behavior.
Connect Object Relationships : Many labels, especially algorithmic ones, require joining entity attributes via key fields. Example: link users to purchased products via orders, then link products to baby-care category to infer whether a user has a child; link enterprise industry and region, aggregate industry/regional averages to infer the enterprise's economic position.
Label Category Design : Partition labels by business aspects, then enumerate specific labels under each category.
Data Label Implementation
Core of implementation is the label fusion table — how labels are stored in the database. Two forms:
Vertical Fusion Table
Each label of each object is a separate record. Flexible: variable number of labels per object, easy to add new labels. Drawback: larger data volume because every label becomes a row.
Horizontal Fusion Table
Multiple labels of one object placed as columns in a single row. Greatly reduces row count. Drawback: adding a new label requires schema change (ALTER TABLE), plus updates to generation and query programs — higher maintenance cost. Therefore used for relatively fixed attribute or statistical labels.
Huawei Data Label Management System
1. Label System Construction
Select target objects per business needs, referencing the company's information architecture business objects.
Design label hierarchy by complexity.
Detailed label and label-value design: definition, applicable scope, generation logic.
Fact labels must match business object attributes/values — no additions or modifications allowed.
Rule labels designed per business department rules.
Model labels generated from algorithmic models.
2. Labeling (Tagging)
Storage Structure : Tagging establishes label-value-to-instance relationships. Can tag a business object, a logical data entity, a physical table, or a single record. A user table can be added for "user"-centric lookup, association, and consumption — "user" broadly means person, organization, department, project, etc.
Implementation Methods :
Fact Labels : System auto-tags based on label-value-to-allowed-attribute-value mapping.
Rule Labels : Design tagging logic; system auto-tags.
Model Labels : Design tagging algorithm/model; system auto-tags.
Signed-in readers can open the original source through BestHub's protected redirect.
This article has been distilled and summarized from source material, then republished for learning and reference. If you believe it infringes your rights, please contactand we will review it promptly.
How this landed with the community
Was this worth your time?
0 Comments
Thoughtful readers leave field notes, pushback, and hard-won operational detail here.
