什么是主数据？
主数据(Master Data)是指满足跨部门业务协同需要的、反映核心业务实体状态属性的企业（组织机构）基础信息。主数据有两个价值，价值一是建立企业基础数据共享“语言”，打破各系统信息交互壁垒，使数据能够在多个系统内充分共享、高度复用；价值二是通过制定主数据标准，为业务报表编制、数据统计分析提供基础条件；通过主数据建设，能为企业在数据应用与管理奠定基础。
Master data is the source of data, also known as the "golden data" of enterprises. It is the core of data asset management, the cornerstone of information system interconnection, and an important foundation for informatization and digitization. The importance of master data management is reflected in the fact that it can eliminate data redundancy, improve data processing efficiency, and improve the company's strategic synergy. Establishing an accurate, unique and authoritative data source and establishing a standard management system for enterprise master data is a key factor in improving the data quality and value of data assets of enterprises and institutions.
The first step in master data management is to identify master data. Generally speaking, master data has the characteristics of high value, entity independence, relative stability, high shareability, identification uniqueness, and long-term validity. The problems brought about by the method are also obvious:
►High threshold: rely on consultants and external experts;
►Slow results: long time period, at least one month to start;
►High cost: Consultants, business personnel, and IT personnel are required to complete manual consultation;
► Hysteresis: Business problems are not identified until they occur.
Facing these challenges, how should enterprises break through?
Solutions
Once the problem is clarified, a targeted solution can be formulated. In order to efficiently identify master data, the solution adopted needs to meet the following requirements: ►Shortly
shorten the project implementation cycle and cost;
►Be able to use technology or tools Support to quickly identify the status quo of master data distribution in various business systems of the enterprise;
►Be able to provide direction for the investigation of the status quo of the business and information of the enterprise, and realize visible value.
It is not difficult to achieve these requirements. By building a machine learning model and training corresponding data for master data management, the problem of master data identification can be perfectly solved, and the data situation of the enterprise can be quickly understood, so as to provide data for subsequent master data management. Base. However, this process involves multiple links such as data extraction, processing, feature engineering, and modeling. It is a big challenge for enterprises. Is there a faster and smarter way?
Learn more about Tempo master data management platform!
solution
The Tempo master data management platform is an enterprise-level master data management platform driven by business and intelligently assisted. It integrates the implementation methodology of master data into product capabilities, meets the master data management needs of different business perspectives, and can solve the problem of corporate governance in the traditional mode. The problem of high cost and slow effect brought by data breaks the single master data management idea and realizes the maximum value at the minimum cost.
△ Master data identification algorithm scheme frame diagram The
master data identification algorithm scheme of Tempo master data management platform starts from two aspects. First, information is extracted from various business systems of the enterprise, and then the machine model is constructed according to the built-in algorithm, and the master data is finally realized. automatic identification.
Step 1: Database information extraction
mainly revolves around the three basic elements of tables, fields, and field values, and sorts out the basic situation information of the database from the two dimensions of table information and field/value information. The sorted data can have an absolute high impact on the database. Interpretation, and then provide data support for the construction of subsequent correlation recognition algorithms.
At the same time, due to the large deviation in the data types of different databases, the Tempo master data management platform stipulates a unified data type standard. In the field feature extraction process, the maximum value, minimum value, and average value of the field value are extracted respectively. , so as to further understand the characteristics of each field value and increase the accuracy.
Step 2: The construction of master data table identification algorithm
mainly uses the data of database information extraction results as input data, adopts various methods to sort out and preliminarily summarize the characteristic features of master data table, and integrates key algorithms such as comprehensive evaluation model and machine learning algorithm, so as to realize Master data table identification.
Algorithm implementation process
Step1: Data input
The main source is the database information to extract the result data.
Step2: Data processing
The data processing process is mainly aimed at the extracted characteristic data of the database, through screening, merging, calculation, outlier value, missing value processing and other methods.
Step3: Feature engineering
sorts out the key indicators related to the target results for different tasks. The construction and determination of these indicators will be determined and selected from the perspectives of business and statistics.
Step4: Comprehensive evaluation/machine learning model
Based on the current data situation, comprehensively apply the comprehensive evaluation and machine learning model to realize the construction of the master data model.
Step5: Result output
Classify the master data identification results into three grades: high, medium and low, so as to realize the recommendation of the master data table.
In this process, the characteristics of table data are the key to describe the main information of the table. The Tempo master data management platform can not only recognize the characteristics of 14 table field values, but also realize the expression and description of event and organization characteristics based on the Bert model. identification of other diverse features.
program value
This scheme has been applied in a coal mine project, and through the verification of stratified + random sampling, and the review method of manual marking + verification, the accuracy and recall rate of master data and reference data have been tested: the master data is
accurate Accuracy: 65.1%; Recall rate: 100%
Reference data accuracy: 61.2%; Recall rate: 100%
The master data identification algorithm of the Tempo master data management platform can also be applied to the following scenarios:
1) Establish enterprise-level master data Unify the view, identify the master data of each business system, form unified data information, and clarify the correlation between the master data of the enterprise; 2) Assist the enterprise to
formulate unified master data standards and standardized management systems and processes, so as to ensure the generation of master data The accuracy of data storage and the controllability of data transfer and maintenance help enterprises to establish a set of complete, authoritative and high-quality master data; 3)
Analysis of factors such as update and change trends of master data can promote The continuous improvement of the management system and the continuous improvement of business development.
Master data identification through the Tempo master data management platform can help enterprises and organizations better manage and maintain master data, improve data quality and reliability, and also enable faster data analysis and decision-making, thereby improving business efficiency and the company The overall strategic synergy lays a solid data foundation for subsequent data sharing and cross-system business collaboration, ensuring the smooth progress of enterprise digital transformation.