CN-121981746-A - Cross-system data integration and value layering method based on client numbers

CN121981746ACN 121981746 ACN121981746 ACN 121981746ACN-121981746-A

Abstract

The invention discloses a cross-system data integration and value layering method based on client numbers, and belongs to the field of digital marketing and financial science and technology. The method comprises the steps of cleaning and standardizing source data of an electronic bank, credit management and a client channel system by taking a client number as a unique association key to form a unified client view, utilizing a gradient lifting tree model to carry out nonlinear fitting on non-sensitive features based on the view to generate a client value ordering result, screening target clients according to the value ordering and outputting a desensitization list only containing the client number, constructing a full-flow automatic closed-loop system covering data integration, image generation, multi-channel touch and effect tracking, and realizing data interaction and task distribution among modules through an API (application program interface). The invention improves the accuracy and automation level of cross-system data integration efficiency and client value layering.

Inventors

QIAO GUANFENG
WU JING
ZHAO BOCHAO
LIU XUEHUA

Assignees

中国建设银行股份有限公司河北省分行

Dates

Publication Date: 20260505
Application Date: 20251217

Claims (10)

1. A method for integrating and layering value of cross-system data based on client numbers, which is characterized by comprising the following steps: S1, acquiring source data of an electronic bank, credit management and a customer channel system, and performing data cleaning and standardization processing by taking a customer number as a unique associated key to form a unified customer view; S2, based on the unified customer view, nonlinear fitting is carried out on the non-sensitive features by utilizing a gradient lifting tree model, and a customer value ordering result is generated; S3, screening target clients according to the value sorting result, and outputting a desensitized target client list only comprising client numbers; S4, constructing a full-flow automatic closed-loop system comprising data integration, image generation, multi-channel touch and effect tracking, and realizing data interaction and task distribution among the modules through an API (application program interface).
2. The method of claim 1, wherein the acquiring source data of the electronic banking, credit management and customer channel system, performing data cleansing and normalization processing by using the customer number as a unique association key, and forming a unified customer view further comprises: S11, performing distributed cleaning on source data by using Spark MLlib components, wherein the distributed cleaning comprises the steps of removing repeated client numbers, correcting field format errors and filling missing values; s12, normalizing the cleaned data according to preset fields through an ETL flow, wherein the preset fields comprise client numbers, customer group labels, daily AUM, transaction liveness and affiliated institution types.
3. The method of claim 1, wherein the non-linear fitting of the non-sensitive features using the gradient-lifted tree model, generating the customer value ranking result further comprises: s21, the non-sensitive characteristics comprise average AUM of the year, difference value of the number of quick payment transactions in the present year and the last year, and weight coefficient corresponding to the guest group type; S22, constructing a plurality of decision trees and weighting and accumulating the prediction results through iterative calculation residual errors.
4. The method of claim 1, wherein said screening target customers based on said value ranking results, outputting a desensitized target customer list containing only customer numbers, comprises: s31, taking the first 10 ten thousand clients as target clients according to the value ordering result of the clients; S32, the output result only comprises a client number list and is transmitted to the marketing automation system through the API interface encryption.
5. The method of claim 1, wherein constructing a full-flow automated closed-loop system including data integration, image generation, multi-channel reach and effect tracking, implementing data interaction and task distribution between modules through an API interface, comprises: S41, an API interface adopts a RESTful architecture, and a request body is in a JSON format and comprises a task_ id, channel, template _id array and a audience array; and S42, the client numbers and the portrait labels in the audience arrays are transmitted in an encrypted mode through an HTTPS protocol, and content rendering is carried out in the channel center.
6. The method as recited in claim 1, further comprising: S5, collecting client behavior data in real time through a buried point SDK, wherein the behavior data comprises task_id, uid, time stamp and event type; S6, triggering Spark SQL batch processing tasks every day, carrying out LEFT JOIN operation on the buried point behavior table and the transaction flow table, screening out client IDs with click records but without purchase records, forming a high-intention unconverted list, and distributing the high-intention unconverted list to manual, AI outbound and enterprise microchannels through a rule engine.
7. A cross-system data integration and value layering device based on customer numbering, comprising: the data acquisition and cleaning module is used for acquiring source data of the electronic bank, credit management and client channel system, and performing data cleaning and standardization processing by taking the client number as a unique association key to form a unified client view; The gradient lifting tree model processing module is used for carrying out nonlinear fitting on the non-sensitive features by utilizing the gradient lifting tree model based on the unified customer view to generate a customer value ordering result; the target client screening and desensitizing module is used for screening target clients according to the value sorting result and outputting a desensitized target client list only comprising client numbers; The automatic closed-loop system construction module is used for constructing a full-flow automatic closed-loop system comprising data integration, image generation, multi-channel touch and effect tracking, and data interaction and task distribution among the modules are realized through an API (application program interface).
8. The apparatus as recited in claim 7, further comprising: The embedded point SDK acquisition module is used for acquiring client behavior data in real time through the embedded point SDK, wherein the behavior data comprises task_id, uid, time stamp and event type; and the Spark SQL batch processing module is used for triggering Spark SQL batch processing tasks every day, carrying out LEFT JOIN operation on the buried point behavior table and the transaction flow table, screening out client IDs with click records but without purchase records, forming a high-intention unconverted list, and distributing the high-intention unconverted list to manual, AI outbound and enterprise micro-channels through a rule engine.
9. An electronic device, comprising: A processor; And when the processor executes the instructions, the method for integrating and layering the cross-system data based on the client numbers is realized according to any one of claims 1-6.
10. A computer readable storage medium storing a computer program which, when executed by a processor, implements a client numbering based cross-system data integration and value stratification method according to any one of claims 1-6.

Description

Cross-system data integration and value layering method based on client numbers Technical Field The invention relates to the field of digital marketing and financial science and technology, in particular to a method, a device, equipment and a storage medium for integrating cross-system data and layering value based on client numbers. Background With the evolution and popularization of mobile communication networks, the online and convenient implementation of payment behaviors has become an irreversible social development trend. Among them, the fast payment has become the mainstream payment mode because of its easy and simple to handle, trade high-efficient and risk controllable characteristic, has accumulated massive user and trade data. Under the background, the construction of the refined operation and marketing model based on the quick payment guest group plays a vital role in improving service efficiency and mining customer value in banking industry. However, in the current industry, there is a critical transition from solving the underlying "availability" problem to overcoming the deep "availability" and "value" problems. The existing technical scheme and development direction show that the bottleneck is difficult to break through by simply relying on a complex algorithm, and the real breaking point is that how to break through the data barriers in enterprises on the premise of strictly adhering to the data compliance and privacy protection requirements, how to formulate a highly personalized strategy according to dynamic and accurate client insight, and how to trade-off and optimize the short-term benefits of marketing activities and the long-term client life cycle value in real time, so that the phenomenon of falling into inefficient marketing inner rolls is avoided. Despite the continuous progress of technology, there are several key technical drawbacks and serious challenges to be overcome at the level of actual business ground: First, there is an inherent barrier to the data fusion layer. The key behavior and attribute data of users, such as payment transaction data, financial management data and credit business data, are generally stored in different business systems in a scattered manner and managed by independent departments to form stubborn 'data islands'. This state makes it difficult to effectively correlate and clean multi-source heterogeneous data by a safe and compliant means, so that a real-time, unified 360-degree customer view cannot be constructed. Meanwhile, the protection requirement on the sensitive information of the user in the cross-system data circulation process is extremely high, and the traditional data aggregation mode has leakage risk, so that the complexity of technology integration and the compliance cost are further increased. Secondly, customer portraits and policy generation accuracy is not sufficient. Because of the reliance on one-sided and lagged data, the user portraits generated by the existing models are often not accurate and comprehensive enough. This results in marketing strategies that tend to be rough and uniform and fail to meet the customer's personalized needs. By adopting the fixed marketing technique and the fixed marketing scheme for pushing, the response rate of the marketing activity is low, the conversion effect is not in the best, and customer complaints can be caused by disturbing users or recommending irrelevant products, so that the user experience and the bank reputation are damaged. Finally, the marketing effectiveness assessment lags behind the feedback mechanism. Marketing campaigns typically have significant time delays from execution to producing a measurable business effect, and existing technical architectures lack the ability to track and analyze the real-time or near real-time effects of marketing actions. Due to the lack of the feedback closed loop, an operator cannot adjust the strategy in time, so that the effect analysis of a marketing activity is insufficient, and the allocation of marketing resources is difficult to dynamically optimize, so that the improvement of the overall input-output ratio and the maintenance of the long-term customer value are affected. In summary, in the prior art, when realizing accurate, efficient and compliant marketing based on fast payment customer groups, obvious short plates exist in key links such as data integration, image accuracy, real-time feedback optimization and the like. Therefore, an innovative solution is urgently needed to systematically solve the above-mentioned problems. Disclosure of Invention The present invention aims to solve at least one of the technical problems in the related art to some extent. The invention provides a cross-system data integration and value layering method based on client numbers, which is characterized in that a unified client view is formed by cleaning and standardizing multi-system source data by taking the client numbers as unique associated keys, client