I. Introduction
A friend asked me, what is an indicator, what is a label, and what is the most essential difference? How to identify? Don't ask me, I think I am relatively clear, it seems to be very clear, but when someone asks, I think I am not clear again, so I study it again, and share the learning notes with everyone, hoping to help and inspire everyone. .
Second, the understanding of indicators
1. Indicators are concepts that describe the overall comprehensive quantitative characteristics. All indicators can be represented by numerical values. A complete statistical indicator must talk about time, place, and scope (Baidu);
2. The evaluation of indicators is easier to quantify, and there are usually certain standards and scales;
3. Indicators are productive thinking and dismantling thinking, focusing on breaking things into pieces, breaking things down for multi-angle descriptions, and getting a lot of indicators;
4. The best applications of indicators are monitoring, analysis, evaluation and modeling;
5. Indicators are business management-oriented and need to be planned in advance. There are many application scenarios, such as strategic goals, market positioning, business monitoring, performance assessment, task decomposition, data analysis, data modeling, and BI applications.
Third, label understanding
1. Labels are attributes of objects, and "labels" from granularity to field level refer to data resources that are cleaned and processed from raw data and can be used by businesses and generate value. Generally, they need to be structured to field granularity to ensure service-oriented use. (label category system)
2. Labels are synthetic thinking and aggregated thinking, focusing on turning parts into wholes, comprehensively processing multiple scattered indicators according to certain principles, and obtaining general results;
3. Tags are often also called attributes, features, indicators, parameters, etc.;
4. The indicator is a semi-finished product, the label is the finished product, and the label is the result of further productization of the indicator;
5. The label is oriented to the data application side, and answers the questions of "how to use data" and "what is the value of data";
6. A tag is a resource, an asset, a data product that can be priced, sold, and traded;
8. Labels are application-oriented, changing with business needs and increasing at any time;
9. The applications that labels are best at are labeling, characterization, classification and feature extraction;
10. Tags are mainly used in customer grouping, portraits, customer contact, customer acquisition, sticky customer, customer renewal, data modeling, data visualization, etc.;
11. The evaluation of tags is generally strongly related to the user's feelings and application results. Different people and different application scenarios may have different effects of tags.
Four, label layering
1. Understand the difference and connection between the root directory, label category, label, and label value, and the label system will be clearer. The following is the thinking of the insurance asset level, which can correspond to the design thinking of the data middle platform.
2. The root directory points to the object to which the label belongs : The root directory is often a vague, broad, and simple noun or gerund, such as users, house buyers, hotels, browsing (records), transactions (records), repairs (records) . According to data thinking, everything in the world can be classified into three types of objects: people, things and relationships. Therefore, a word used to refer to an object (a noun refers to people, objects, and a gerund refers to relationships) should not be a label. Label root directory. At the physical level of data, it is often mapped to the primary key in a large-width table. The information in this large-width table is a detailed description and data record of the primary key object: the columns of the large-width table are mapped to labels, and the rows of the large-width table are mapped to labels. The record corresponds to the specific attribute value record of the specific object on each tag attribute.
3. Category is the classification of tags : customer tags can be classified into basic information, geographic location, social relationship, etc. These category names are also category names. Categories are often made up of nouns. A category and its classified labels can correspond to a specific table at the physical level of data. For example, under the [Basic Information] category of the "customer" object, there are multiple labels such as "gender", "age", and "hometown". , which generally corresponds to a basic customer information table in the customer database, which has multiple fields such as "gender", "age", and "hometown".
4. The label is the attribute of the object, and the granularity is at the field level : "Customer name", "customer phone number", "customer residential address" and other field granular attributes are the label of the "customer" object. Labels are often composed of two nouns before and after, and the former noun is used as an object attribute to modify the latter noun. Labels generally correspond to a field in a data table in a database.
5. The label value is the specific value of the object attribute : for example, [Xiao Ming] [Xiaohong] is the label value of the "Customer Name" label, and [Male] [Female] is the label value of the "Gender" label. Label values are often adjectives, nouns, or numbers, and generally correspond to the value of a field in a data table in the database. The value type of the tag value can be numeric, text, date, or key-value, but it is mainly numeric. Numerical types are divided into enumerable discrete values and non-enumerable continuous values.
Five, label classification
The classification of tags is for applications, and you can add them as needed.
1. According to the variability of tags, it is divided into static tags and dynamic tags;
2. According to the label's reference and evaluation index, it can be divided into qualitative label and quantitative label;
3. According to the hierarchical method of label assets, it can be divided into first-level labels, second-level labels, third-level labels, etc. Each level of label is equivalent to a slice of a business dimension, which conforms to the MECE principle.
5. According to the degree of complexity, it is divided into: fact label, rule label and model label. Fact labels are usually realistic and have a high degree of coincidence with indicators.
6. Such as gender, age, etc.; rule labels are generally controlled by some simple rules, and corresponding labels are generated when certain rules are met; model labels generally need to be generated by some machine learning algorithms.
6. Conclusion
Labels are attributes of objects, generally down to field granularity, facing the data application side, are resources, are assets, a data product that can be priced, sold, and tradable, including attributes, features, indicators, parameters, etc.;
Indicators are quantifiable, fields represented by numerical values, business management-oriented, and need to be planned in advance. The applications they are good at are monitoring, analysis, evaluation and modeling;