转载/码工小熊
Hello everyone, I'm the little bear girl who loves to learn.
Today, I will share with you a skill that is as important as the indicator system: constructing labels (commonly known as: labeling). The ability to label is an important ability to distinguish real data analysts from sqlboy
1. What is a label?
A label is a high-level description of something. Just like the product classification and main raw materials will be written on the product label. Although a product has many attributes, we can lock the product we want with only a few limited tags, which is the role of the tag.
2. What are the labels?
In terms of complexity, there are four types of tags:
1. Factual labels. Such as the color of the product, the gender of the person. These are factual descriptions and can be used directly.
2. Regular labels. For example, "consumption of more than 1,000 yuan" is defined as a high-consumption group. Rule class labels are often based on a data indicator and then classified according to specific rules.
3. Composite labels. For example, "Gao Fu Shuai" is a typical compound label, which is based on N indicators, performs comprehensive calculation, and finally obtains a label result.
4. Predictive labels. Note that the above three types of labels are calculated using the data that has already occurred. Predictive labels are estimates of future conditions. Predictions can be made algorithmically or manually. For example, classify users and label them "expected to lose users", which means that the user will be lost in the next XX time.
The complexity of these four types of labels is different: predictive type > composite type > regular type > factual type. The corresponding difficulty of labeling is also different.
3. How to label?
Labeling is a colloquial term that refers to the process of producing labels. If it is a simple factual label, just drag it over and use it. The other three categories have to go through the action of labeling. The more complex the label, the more troublesome it is to produce.
There are four standard steps to labeling.
明确打标签的对象
明确标签的用途
明确标签规则
明确标签的名称
To give the simplest example: Little Bear has no boyfriend yet, so I want to find a rich and handsome man.
Gao Fushuai is just a compound label, which follows four steps:
1. Tag objects: men, living
2. Clear purpose: choose a potential blue pot friend
3. Clarify the rules: This is the most complicated step, because there are three dimensions to being rich and handsome. First, get one dimension and one dimension, explain it separately, and then find a way to synthesize it.
Here, height is relatively easy, just look at the height and give a standard.
Handsome, and relatively easy to see, I think it's ok when I see the photos, I can just tag it by hand.
Rich, very confused. Some people earn more, but also spend more. Not only income, but also debt.
In this way, the three dimensions are first labeled, and then integrated (as shown in the figure below).
For a comprehensive method, you can use priority sorting, or do comprehensive scoring (as shown in the figure below):
it seems that comprehensive scoring seems to be more scientific, but it is not in actual decision-making! For example, I can't walk when I see a handsome guy, no matter whether he is rich or not. The comprehensive score may select mediocre individuals, so pay attention to this.
4. Specify the name of the label: Gao Fushuai, not Gao Fushuai
Get it done!
It can be seen that in the process of producing labels, clarifying the rules is the most troublesome step. But in reality, the trouble is not limited to that.
Fourth, labeling, where is the difficulty?
After reading the above small example, many people will think: "Tagging is very simple, I can type 1,000 in a morning". It's really easy if you just add 1000 new fields to the database.
However, these 1000 fields:
有几个能被业务部门用起来？
用完以后能提升业务表现？
还有多少业务想要的标签，没有在其中？
That's the real problem.
Anyway, I’ve seen them foolishly with hundreds of labels, and as a result, apart from the online reporting ppt, the business department didn’t even look at them, let alone use them.
A good label must be:
业务高频使用
指向明确动作
产生明显效果
It's like when Little Bear girl hears someone introduce "Gao Fu Shuai", she will get up and put on makeup for two hours to go out with the dark circles under her eyes from staying up late on Friday. This is a label with high frequency use, driving force, and obvious effect!
Some friends must ask if there is such a label in the business, of course there is, such as my personal favorite: promotion-sensitive users (yes/no), this label. Special distinction: users who do not buy if there is no promotion, and users who have a high probability of buying in promotion. It is easy to use to explain the conversion rate of daily consumption, predict the effect of the activity in advance, and review the achievement after the event (sell it first, and then share this specifically).
Real business problems are often complex, and it is difficult to describe the situation with a single label, so it needs to revolve around a business scenario. Construct several labels to form a label system to drive business work.