Customer data processing method, device, equipment and computer-readable storage medium

Abstract

The present application discloses a customer group data processing method, device, equipment and computer-readable storage medium, which receives a customer group data processing instruction, obtains customer group data to be amplified according to the customer group data processing instruction; amplifies the customer group data based on the customer group data to be amplified to obtain recalled customer group data; and sorts the recalled customer group data to obtain target customer group data. The present application first amplifies the customer group data to be processed in the customer group data processing instruction to obtain the recalled customer group data after the customer group amplification. The recalled customer group data can be produced in batches from the customer group data to be processed in a long-term and efficient manner. The recalled customer group data is then sorted to sort the customers in the customer group data, so that the customers produced in batches are high-quality customers, and the target customer group data is obtained. The target customer group data can be accurately promoted in a targeted manner to improve the efficiency of intelligent marketing.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a customer data processing method, device, equipment and computer-readable storage medium. Background Art

[0002] With the widespread adoption of machine learning and artificial intelligence methods, intelligent marketing technologies are widely used in customer targeting, prospect mining, and precision marketing. In addition to traditional manual screening, key methods for selecting marketing customer lists include tag-based collaborative filtering, tag diffusion, social relationship-based diffusion, cluster-based diffusion, and target population classification. These methods are based on real-time scenarios, using online real-time recommendation systems to make recommendations for different user profiles, while others rely on offline recommendation processes to accurately predict customers. However, these methods, as current intelligent solutions for customer list selection in the intelligent marketing field, are unable to efficiently and long-term produce large quantities of high-quality marketing customer lists, resulting in low efficiency in current intelligent marketing practices. Summary of the Invention

[0003] The main purpose of this application is to provide a customer data processing method, device, equipment and computer-readable storage medium, aiming to solve the technical problem of low efficiency of current intelligent marketing.

[0004] To achieve the above objectives, an embodiment of the present application provides a customer group data processing method, the customer group data processing method comprising:

[0005] receiving a customer group data processing instruction, and obtaining the customer group data to be amplified according to the customer group data processing instruction;

[0006] Amplify the customer group data based on the customer group data to be amplified to obtain recalled customer group data;

[0007] The recalled customer group data is sorted to obtain target customer group data.

[0008] Preferably, the step of amplifying customer group data based on the customer group data to be amplified to obtain recalled customer group data includes:

[0009] Filtering the customer group data based on the processing scenario in the customer group data processing instruction to obtain first customer group data;

[0010] performing customer group data sampling on the first customer group data based on the customer group data to be amplified to obtain sample customer group data, and filtering the sample customer group data from the first customer group data to obtain second customer group data;

[0011] A sample size assessment is performed on the second customer group data, and the customer group data is amplified based on the assessment result to obtain the recalled customer group data.

[0012] Preferably, the step of evaluating the sample size of the second customer group data, amplifying the customer group data based on the evaluation result, and obtaining the recalled customer group data includes:

[0013] Comparing the sample size of the second customer group data with a preset sample size threshold to obtain an evaluation result;

[0014] If the evaluation result is that the sample size of the second customer group data is greater than or equal to the preset sample size threshold, customer group data is recalled on the second customer group data based on the trained recall model to obtain recalled customer group data.

[0015] Preferably, the step of recalling the second customer group data based on the trained recall model to obtain the recalled customer group data includes:

[0016] Clustering the sample customer group data to obtain a seed user center;

[0017] Calculating the distance between the second customer group data and the seed user center based on the trained recall model to obtain customer group distance information;

[0018] The second customer group data is recalled according to the customer group distance information to obtain recalled customer group data.

[0019] Preferably, after the step of comparing the sample size of the second customer group data with a preset sample size threshold to obtain an evaluation result, the method further includes:

[0020] If the evaluation result is that the sample size of the second customer group data is smaller than the preset sample size threshold, the second customer group data is determined as recalled customer group data.

[0021] Preferably, before the step of sorting the recalled customer group data to obtain target customer group data, the step further includes:

[0022] Obtaining a customer group data processing mode in the customer group data processing instruction;

[0023] The step of sorting the recalled customer group data to obtain target customer group data includes:

[0024] If the customer group data processing mode is the historical strategy mode, the recalled customer group data is predicted using the trained sorting model, and the recalled customer group data is sorted and filtered according to the prediction result to obtain the target customer group data.

[0025] Preferably, after the step of obtaining the customer group data processing mode in the customer group data processing instruction, the method further includes:

[0026] If the customer group data processing mode is the seed list mode, the recalled customer group data is filtered to obtain target customer group data.

[0027] To achieve the above objectives, the present application further provides a customer group data processing device, the customer group data processing device comprising:

[0028] an acquisition module, configured to receive a customer group data processing instruction and acquire the customer group data to be amplified according to the customer group data processing instruction;

[0029] an amplification module, configured to amplify customer group data based on the customer group data to be amplified to obtain recalled customer group data;

[0030] The sorting module is used to sort the recalled customer group data to obtain target customer group data.

[0031] Furthermore, to achieve the above-mentioned purpose, the present application also provides a customer data processing device, which includes a memory, a processor, and a customer data processing program stored on the memory and executable on the processor. When the customer data processing program is executed by the processor, the steps of the above-mentioned customer data processing method are implemented.

[0032] Furthermore, to achieve the above-mentioned purpose, the present application also provides a medium, which is a computer-readable storage medium, and a customer data processing program is stored on the computer-readable storage medium. When the customer data processing program is executed by the processor, the steps of the above-mentioned customer data processing method are implemented.

[0033] Furthermore, to achieve the above-mentioned purpose, the present application also provides a computer program product, including a computer program, which implements the steps of the above-mentioned customer data processing method when executed by a processor.

[0034] The embodiments of the present application provide a customer group data processing method, apparatus, device, and computer-readable storage medium, which receive a customer group data processing instruction, obtain customer group data to be amplified according to the customer group data processing instruction; amplify the customer group data based on the customer group data to be amplified to obtain recalled customer group data; and sort the recalled customer group data to obtain target customer group data. The present application first amplifies the customer group data to be processed in the customer group data processing instruction to obtain the recalled customer group data after the customer group amplification. The recalled customer group data can be efficiently and long-term batch-produced from the customer group data to be processed, and then the recalled customer group data is sorted to sort the customers in the customer group data, so that the customers produced in batches are high-quality customers, and the target customer group data is obtained. The target customer group data can be accurately promoted in a targeted manner to improve the efficiency of intelligent marketing. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 This is a schematic diagram of the structure of the hardware operating environment involved in the embodiment of the customer group data processing method of this application;

[0036] Figure 2 This is a flow chart of the first embodiment of the customer data processing method of this application;

[0037] Figure 3 This is a flow chart of the second embodiment of the customer data processing method of this application;

[0038] Figure 4 This is a functional module diagram of a preferred embodiment of the customer data processing device of this application.

[0039] The realization of the objectives, functional features and advantages of this application will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0040] It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0041] The embodiments of the present application provide a customer group data processing method, apparatus, device, and computer-readable storage medium, which receive a customer group data processing instruction, obtain customer group data to be amplified according to the customer group data processing instruction; amplify the customer group data based on the customer group data to be amplified to obtain recalled customer group data; and sort the recalled customer group data to obtain target customer group data. The present application first amplifies the customer group data to be processed in the customer group data processing instruction to obtain the recalled customer group data after the customer group amplification. The recalled customer group data can be efficiently and long-term batch-produced from the customer group data to be processed, and then the recalled customer group data is sorted to sort the customers in the customer group data, so that the customers produced in batches are high-quality customers, and the target customer group data is obtained. The target customer group data can be accurately promoted in a targeted manner to improve the efficiency of intelligent marketing.

[0042] like Figure 1 As shown, Figure 1 It is a structural diagram of the customer data processing device of the hardware operating environment involved in the embodiment of the present application.

[0043] In the subsequent description, suffixes such as "module," "component," or "unit" used to represent elements are only used to facilitate the description of the present application and have no specific meaning. Therefore, "module," "component," or "unit" may be used interchangeably.

[0044] The customer data processing device in the embodiment of the present application may be a PC, or a portable terminal device such as a tablet computer or a portable computer.

[0045] like Figure 1 As shown, the customer data processing device may include: a processor 1001, such as a CPU, a network interface 1004, a user interface 1003, a memory 1005, and a communication bus 1002. Among them, the communication bus 1002 is used to realize the connection and communication between these components. The user interface 1003 may include a display screen (Display), an input unit such as a keyboard (Keyboard), and the user interface 1003 may also include a standard wired interface and a wireless interface. The network interface 1004 may optionally include a standard wired interface and a wireless interface (such as a WI-FI interface). The memory 1005 may be a high-speed RAM memory, or a stable memory (non-volatile memory), such as a disk memory. The memory 1005 may also be a storage device independent of the aforementioned processor 1001.

[0046] Those skilled in the art will understand that Figure 1 The structure of the customer group data processing device shown in the figure does not constitute a limitation to the customer group data processing device, and may include more or fewer components than shown in the figure, or combine certain components, or arrange the components differently.

[0047] like Figure 1 As shown, the memory 1005 as a storage medium may include an operating system, a network communication module, a user interface module, and a customer group data processing program.

[0048] exist Figure 1 In the device shown, the network interface 1004 is mainly used to connect to the backend server and communicate data with the backend server; the user interface 1003 is mainly used to connect to the client (user end) and communicate data with the client; and the processor 1001 can be used to call the customer data processing program stored in the memory 1005 and perform the following operations:

[0049] receiving a customer group data processing instruction, and obtaining the customer group data to be amplified according to the customer group data processing instruction;

[0050] Amplify the customer group data based on the customer group data to be amplified to obtain recalled customer group data;

[0051] The recalled customer group data is sorted to obtain target customer group data.

[0052] Furthermore, the step of amplifying customer group data based on the customer group data to be amplified to obtain recalled customer group data includes:

[0053] Filtering the customer group data based on the processing scenario in the customer group data processing instruction to obtain first customer group data;

[0054] performing customer group data sampling on the first customer group data based on the customer group data to be amplified to obtain sample customer group data, and filtering the sample customer group data from the first customer group data to obtain second customer group data;

[0055] A sample size assessment is performed on the second customer group data, and the customer group data is amplified based on the assessment result to obtain the recalled customer group data.

[0056] Furthermore, the steps of performing sample size evaluation on the second customer group data, amplifying the customer group data based on the evaluation result, and obtaining the recalled customer group data include:

[0057] Comparing the sample size of the second customer group data with a preset sample size threshold to obtain an evaluation result;

[0058] If the evaluation result is that the sample size of the second customer group data is greater than or equal to the preset sample size threshold, customer group data is recalled on the second customer group data based on the trained recall model to obtain recalled customer group data.

[0059] Furthermore, the step of recalling the second customer group data based on the trained recall model to obtain the recalled customer group data includes:

[0060] Clustering the sample customer group data to obtain a seed user center;

[0061] Calculating the distance between the second customer group data and the seed user center based on the trained recall model to obtain customer group distance information;

[0062] The second customer group data is recalled according to the customer group distance information to obtain recalled customer group data.

[0063] Furthermore, after the step of comparing the sample size of the second customer group data with a preset sample size threshold to obtain an evaluation result, the processor 1001 may be configured to call a customer group data processing program stored in the memory 1005 and perform the following operations:

[0064] If the evaluation result is that the sample size of the second customer group data is smaller than the preset sample size threshold, the second customer group data is determined as recalled customer group data.

[0065] Furthermore, before the step of sorting the recalled customer group data to obtain target customer group data, the processor 1001 may be configured to call a customer group data processing program stored in the memory 1005 and perform the following operations:

[0066] Obtaining a customer group data processing mode in the customer group data processing instruction;

[0067] The step of sorting the recalled customer group data to obtain target customer group data includes:

[0068] If the customer group data processing mode is the historical strategy mode, the recalled customer group data is predicted using the trained sorting model, and the recalled customer group data is sorted and filtered according to the prediction result to obtain the target customer group data.

[0069] Furthermore, after the step of obtaining the customer group data processing mode in the customer group data processing instruction, the processor 1001 may be configured to call the customer group data processing program stored in the memory 1005 and perform the following operations:

[0070] If the customer group data processing mode is the seed list mode, the recalled customer group data is filtered to obtain target customer group data.

[0071] To better understand the above technical solutions, exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments described herein. Instead, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0072] In order to better understand the above technical solution, the above technical solution will be described in detail below with reference to the accompanying drawings and specific implementation methods.

[0073] Reference Figure 2 The first embodiment of the present application provides a flow chart of a customer group data processing method. In this embodiment, the customer group data processing method includes the following steps:

[0074] Step S10, receiving a customer group data processing instruction, and obtaining the customer group data to be amplified according to the customer group data processing instruction;

[0075] Understandably, with the widespread adoption of machine learning and artificial intelligence methods, intelligent marketing technologies are widely used in customer targeting, prospect mining, and precision marketing. For customer list screening, in addition to traditional manual screening, key methods include tag-based collaborative filtering, tag diffusion, social relationship-based diffusion, cluster-based diffusion, and target audience classification. These methods are based on real-time scenarios, using online recommendation systems to make recommendations for different user profiles, while others rely on offline recommendation processes to accurately predict customers. These methods, as current intelligent solutions for customer list screening in the intelligent marketing field, suffer from the following main issues: they are limited to specific scenarios, requiring high costs to change scenarios; model evaluation results differ from actual A / B testing results when reusing solutions, resulting in poor reuse of prediction strategies; and some solutions have high feature requirements or inconsistent performance across specific scenarios. These issues hinder the efficient and long-term production of high-quality, large-scale marketing customer lists, impacting the effectiveness and efficiency of intelligent marketing. Moreover, there is no complete methodology for fan amplification at present, and the existing customer group amplification method cannot perform adaptive model selection based on sample distribution and scenario characteristics; traditional technology has a high dependence on the size of training samples and the source of complex features; traditional technology processes are complex, cannot directly complete end-to-end tasks, and do not have high availability and reusability for multiple scenarios. On this basis, this embodiment provides a customer group data processing method. In this embodiment, the customer group data processing method is applied to a customer group data processing system. The customer group data processing system may include a front-end amplification platform and an algorithm back-end. The front-end amplification platform is used to interact with users (such as marketers) to generate and send corresponding instructions to the algorithm back-end. The algorithm back-end is used to receive the instructions sent by the front-end amplification platform and perform corresponding data processing according to the instructions combined with its internal algorithm to meet the user's data processing needs. Among them, the data processing in this embodiment can be customer group data processing, specifically amplification processing of customer group data, and efficiently generating batch customer group data based on a small amount of customer group data extracted according to user input or historical marketing strategies, so that users can conduct targeted promotion on the generated batch customer group data, thereby improving the efficiency of intelligent marketing.

[0076] It should be noted that the front-end amplification platform in this embodiment is mainly composed of two parts: an amplification task creation module and a task management module. The amplification task creation module is used by users to create amplification tasks, and the task management module is used by users to manage the created amplification tasks. Specifically, when users (such as marketing personnel of banks and shopping malls) need to obtain batches of marketing customer groups for precise promotion due to marketing needs, the amplification task creation module is used to establish the customer group name for this amplification task and select a specific processing scenario (i.e., marketing scenario). Each processing scenario corresponds to an adapted amplification algorithm, so that the subsequent algorithm backend automatically selects the appropriate amplification strategy for the processing scenario; further, the user provides a small number of high-quality marketing customer groups as a seed list, so that the algorithm backend can mine marketing customer groups based on the seed list to obtain batches of marketing customer groups, where the seed list can be manually uploaded by marketing personnel or extracted from successful high-quality customers in historical marketing strategies. Furthermore, the user needs to provide the algorithm with amplification parameters, specifically by directly entering the number of people on the desired amplification list and the marketing channel. After completing the task parameter setting, the task is created. The front-end amplification platform generates customer data processing instructions based on the input task parameters and sends them to the algorithm backend, which automatically matches the appropriate algorithm and generates a list of the corresponding number of people and a customer list that meets the desired marketing channel preferences. Furthermore, in this embodiment, the task management module can display submitted amplification tasks to the user and display the current status of the amplification tasks, allowing users to conveniently manage the amplification tasks visually based on the displayed content.

[0077] Furthermore, the algorithm backend receives customer data processing instructions sent by users based on the front-end amplification platform. Since the request process of the algorithm backend is: build the server with Python's flask module, specify the open local port, fan amplification service address and POST method, Flask is a lightweight Web application framework written in Python; the service function uses the first error handling, receives the parameters requested by the service interface, and performs name and type checks on the specified task parameters. After checking, it builds the task parameter dictionary. If there is an error, it returns 400 and an error message; using the second error handling, it builds the shell command parameters for starting the fan amplification task, uses the Popen function method of the python package subprocess to submit the shell command, and specifies the output log address. If there is an error, it returns 500 and an error message; and returns the startup information of the service port model to the service call interface. Therefore, after receiving the customer data processing instruction, the task parameters in the customer data processing instruction are first obtained, and the Spark environment and related basic parameter settings are initialized. The first layer of error handling is then entered to check the relevant task parameters. The task parameters are then used to initialize the customer amplification modeling class instance. If this process reports an error, exception No. 3 and an error message are thrown. If the first layer does not report an error, the second layer of error handling is entered to run the startup function of the customer amplification modeling instance. If the modeling process reports an error, exception No. 4 and related error messages are thrown. Furthermore, the customer data processing mode selected by the user is determined. In this embodiment, the customer data processing mode can include a seed list mode and a historical strategy mode. The seed list mode is where the user directly provides a seed list, while the historical strategy mode extracts a seed list from high-quality customers that have been successful in historical marketing strategies. Therefore, if the customer group data processing mode is the seed list mode, the seed user is read from ECS (Elastic Cloud Storage, a distributed cloud storage framework) according to the ecs_key in the task parameters and passed to the get list function, thereby obtaining the seed list as the customer group data to be amplified; if the customer group data processing mode is the historical strategy mode, the activity and strategy parameters in the task parameters are passed to the get list function to obtain the seed list as the customer group data to be amplified, where the get list function is a kind of encapsulation used to generate seed users according to the seed list mode to obtain the seed list. The customer group data to be amplified is obtained according to the customer group data processing instructions, which facilitates the subsequent customer group data amplification based on the customer group data to obtain the recalled customer group data.

[0078] Step S20, amplifying the customer group data based on the customer group data to be amplified to obtain recalled customer group data;

[0079] After obtaining the customer group data to be amplified according to the customer group data processing instruction, the processing scenario is obtained from the task parameters of the customer group data processing instruction, and the users corresponding to the processing scenario are filtered out as the first customer group data to obtain larger customer group data. However, in order to ensure that accurate marketing can be carried out based on the customer group data finally generated, it is also necessary to perform customer group data sampling on the first customer group data based on the customer group data to be amplified, wherein the sampling may include positive sampling and / or negative sampling to obtain sample customer group data, and filter out the sample customer group data from the first customer group data to obtain the second customer group data as the remaining customer group data of the processing scenario. Furthermore, the sample size of the second customer group data is evaluated, specifically, the sample size of the second customer group data can be compared with the preset sample size threshold to obtain an evaluation result; and the customer group data is amplified according to the evaluation result obtained by the sample size evaluation to obtain the recalled customer group data, wherein the preset sample size threshold is a sample size value set according to actual needs. Specifically, if the evaluation result shows that the sample size of the second customer group data is greater than or equal to the preset sample size threshold, the second customer group data is recalled based on the trained recall model to obtain recalled customer group data. The recall model is an unsupervised learning model used to reduce the sample size to reduce the amount of calculation. In this embodiment, the recall model is pre-built and trained using unlabeled data. On the contrary, if the evaluation result shows that the sample size of the second customer group data is less than the preset sample size threshold, the second customer group data is determined to be recalled customer group data. The recalled customer group data obtained after the customer group data is amplified can be efficiently and long-term batch-produced from the customer group data to be processed, making it easier to sort the recalled customer group data, so as to sort the customers in the customer group data, so that the customers produced in batches are high-quality customers, and the target customer group data is obtained. It can be accurately promoted to the target customer group data, thereby improving the efficiency of intelligent marketing.

[0080] It is understandable that after obtaining the second customer group data and the sample customer group data, it is necessary to perform data preprocessing on the second customer group data and the sample customer group data to reduce the amount of calculation and improve data processing efficiency, wherein the data preprocessing in this embodiment may include feature classification, feature cleaning, feature conversion and feature selection. Specifically, the second customer group data and the sample customer group data are first subjected to feature classification, more specifically, the second customer group data and the sample customer group data are subjected to feature dimension classification. After analyzing the processing scenarios, the various features required in each scenario are summarized and classified into the contents of each basic dimension feature group, including user portrait dimension features, scenario context features, and scenario behavior features. When modeling, for a specific scenario, the names of various feature groups related to it can be input and combined. The processing scenario setting and feature selection work can be decoupled in advance. Moreover, different data can be initially divided into quantitative and qualitative based on their nature, and further divided into categorical, ordinal, interval, and ratio levels based on meaning, format, data type, etc. Each level will have different controls and mathematics to operate. In actual use, different feature engineering methods will be used to convert the data into features that can better reflect the differences between samples. Furthermore, the data after feature classification is cleaned. The specific data cleaning work includes: evaluating each feature, replacing outliers according to features, discarding according to feature missing rates, and supplementing missing values. Data cleaning is completed at two levels. The first is to complete a part of targeted data cleaning when extracting feature groups; the second is to perform general cleaning by the function module in pyspark (PySpark is the API provided by Spark for Python developers) during feature processing. Furthermore, it is necessary to perform feature conversion on the customer data. The purpose of feature conversion is to optimize the distribution of data in the features and convert the data into digital features that can be used for machine learning modeling according to its type and format. Finally, feature selection is performed on the customer data. The main purpose of feature selection is to select an effective feature subset to reduce the number of features used in modeling. This improves the prediction accuracy of the machine learning pipeline, accelerates training, reduces the risk of overfitting, and increases the interpretability of the model. Considering product efficiency, filtering is mainly used to reduce computing power and reduce coupling with the model.

[0081] It can also be understood that after completing the data preprocessing, it is necessary to obtain the feature groups of the second customer group data and the sample customer group data respectively, and associate the feature groups with the customers. Specifically, in this embodiment, the relevant user primary key features of a time slice are selected in advance according to the scenario requirements, and the preliminary feature quality cleaning is completed to obtain a wide table of user features; the features are divided into four categories, and a pyspark feature engineering pipeline is constructed. The four categories of features are processed and assembled into a single sparse vectorized feature column; the pipeline model is trained during the first run, and then the pipeline model is directly used for conversion to obtain the full user feature group after feature engineering processing; the full user feature group is stored in the specified path of the specified bucket in the ECS, stored according to the actual feature date, and the corresponding life cycle is set, and stored for about 3 months; the above process is scheduled daily for feature generation. Therefore, when obtaining the feature groups of the second customer group data and the sample customer group data, the corresponding feature groups can be read from the ECS specified directory according to the time slice of the activities participated in by each user in the second customer group data and the sample customer group data.

[0082] Step S30: sorting the recalled customer group data to obtain target customer group data.

[0083] After amplifying the customer group data based on the customer group data to be amplified and obtaining the recalled customer group data, the recalled customer group data is processed according to the customer group data processing mode to obtain the target customer group data. Specifically, if the customer group data processing mode is the seed list mode, the recalled customer group data is sorted to obtain the third customer group data, and the third customer group data is filtered to obtain the target customer group data. If the customer data processing mode is the historical strategy mode, the recalled customer data is predicted by the trained sorting model, and the recalled customer data is sorted and filtered according to the prediction results to obtain the target customer data. In this embodiment, the sorting model is used to predict the recalled customer data to determine the predicted value of the recalled customer data, so as to facilitate sorting and filtering the recalled customer data according to the predicted value to obtain the target customer data. The sorting model is constructed and trained using the Stacking model fusion method, specifically to construct a multi-layer model and use multiple prediction results to fit the final prediction result. Its advantage is that multiple different types of individual learners can be used as primary learners (such as logistic regression algorithm, random forest algorithm, GBDT+LR algorithm), and then a layer of final model is added through grid search and random search on the basis of the primary learner group to obtain the final prediction model. By sorting the recalled customer data, the customers produced in batches are high-quality customers, and the target customer data is obtained. The target customer data can be accurately promoted in a targeted manner to improve the efficiency of intelligent marketing. Stacking is an exquisite and complex strategy for model integration.

[0084] This embodiment provides a customer group data processing method, device, equipment and computer-readable storage medium, which receives a customer group data processing instruction, obtains customer group data to be amplified according to the customer group data processing instruction; amplifies customer group data based on the customer group data to be amplified to obtain recalled customer group data; and sorts the recalled customer group data to obtain target customer group data. This application first amplifies the customer group data to be processed in the customer group data processing instruction to obtain the recalled customer group data after the customer group amplification. It can efficiently and long-term batch produce recalled customer group data from the customer group data to be processed, and then sorts the recalled customer group data to sort the customers in the customer group data, so that the customers produced in batches are high-quality customers, and the target customer group data is obtained. It can accurately promote the target customer group data in a targeted manner, thereby improving the efficiency of intelligent marketing.

[0085] This embodiment can use a small batch of high-quality customer lists (i.e., customer group data to be amplified) to produce a large number of marketing high-quality customer lists (target customer group data) with amplification multiples, and at the same time build an efficient platform system for generating amplified lists based on seed lists, and provide list amplification functions for multiple scenarios. This application has a small demand for training sets, and only requires a small batch of seed lists for customer group mining and amplification: expand the number of marketing customer groups and improve marketing coverage through small and fine high-quality customer groups; combine customer characteristics with marketing strategies to improve feature expression capabilities: mine scene-related features, compare with the actual status before the implementation of the benchmark strategy, find out the influential features, and classify them for feature engineering processing; build for multiple scenarios, and adapt to the amplification function of various marketing scenarios; the platform amplification process is easy to operate, and is easy to use for business personnel who do not understand algorithms, and has the advantages of high reusability and efficiency.

[0086] This embodiment designs an asynchronous service architecture of "features - modeling - request". The general modeling process is: receiving a prediction request - extracting data - feature engineering - modeling training - model prediction - output. In the case of large data volumes of hundreds of millions of users, extracting data, feature engineering, and modeling training processes all consume a considerable amount of time. The delay between submitting a request and receiving the result significantly impacts the user experience and the timeliness of the user's participation in the next activity. When adopting an asynchronous structure of these three, and adding the pre-trained model link for existing scenarios, the time from submitting a request to receiving the list is greatly reduced, reducing the time it takes to obtain result feedback to approximately one hour.

[0087] This embodiment implements a decoupled design for multiple links in fan amplification work, including acquisition of target customer data, implementation of modeling, feature generation, and service-oriented deployment. It can perform hot upgrades and agile development optimization for requirements such as replacement of modeling features, replacement of sorting models, supplementation or replacement of scenarios, and adjustment of cluster parameters.

[0088] Further, refer to Figure 3Based on the first embodiment of the customer group data processing method of the present application, a second embodiment of the customer group data processing method of the present application is proposed. In the second embodiment, the step of amplifying the customer group data based on the customer group data to be amplified to obtain the recalled customer group data includes:

[0089] Step S21, filtering customer group data based on the processing scenario in the customer group data processing instruction to obtain first customer group data;

[0090] After obtaining the customer group data to be amplified according to the customer group data processing instruction, first determine the processing scenario in the customer group data processing instruction, and filter out the customer group data corresponding to the processing scenario from the overall customer group data formed by all potential customers to obtain the first customer group data. For example, if the bank's marketing personnel determines that the processing scenario in the customer group data processing instruction sent by the front-end amplification platform is customer group management, then it is determined that the marketing personnel needs to conduct precision marketing in the customer group management scenario, and then all customer group data except for the customer group management scenario are filtered out from the customer group data composed of all customers associated with the bank, and the remaining customer group data constitute the first customer group data. Alternatively, if the bank's marketing personnel determines that the processing scenario in the customer group data processing instruction sent by the front-end amplification platform is activation and stickiness promotion, then it is determined that the marketing personnel needs to conduct precision marketing in the activation and stickiness promotion scenario, and then all customer group data except for the activation and stickiness promotion scenario are filtered out from the customer group data composed of all customers associated with the bank, and the remaining customer group data constitute the first customer group data. The customer group data is sampled on the first customer group data based on the customer group data to be amplified to obtain sample customer group data, and the sample customer group data is filtered out from the first customer group data to obtain second customer group data.

[0091] Step S22: sampling the first customer group data based on the customer group data to be amplified to obtain sample customer group data, and filtering the sample customer group data from the first customer group data to obtain second customer group data;

[0092] After obtaining the first customer group data corresponding to the processing scenario, the customer group data processing mode samples the first customer group data in combination with the customer group data to be amplified, where the sampling includes positive sampling and negative sampling. Specifically, the customer group data processing mode samples customer group data similar to the customer group data to be amplified from the first customer group data as positive sample customer group data, and randomly selects a certain number of customer group data from customer group data other than the positive sample customer group data (i.e., customer group data that is dissimilar to the customer group data to be amplified) as negative sample customer group data, and the sample customer group data is composed of the positive sample customer group data and the negative sample customer group data. Specifically, if the customer group data processing mode is the seed list mode, negative sampling is directly performed proportionally on the first customer group data, and the data is merged into positive and negative samples to obtain sample customer group data; if the customer group data processing mode is the historical strategy mode, the users corresponding to the input activities and strategies are found in the first customer group data, and those that meet the standards are positive, and those that do not meet the standards are negative. The negative samples are sampled proportionally to obtain sample customer group data. Furthermore, the customer group data that is identical to the sample customer group data is filtered out from the first customer group data, and the remaining customer group data is used to form the second customer group data. The sample size of the second customer group data is evaluated, and the customer group data is amplified based on the evaluation result to obtain the recalled customer group data.

[0093] Step S23: performing a sample size assessment on the second customer group data, and amplifying the customer group data based on the assessment result to obtain the recalled customer group data.

[0094] After obtaining the second customer group data, the sample size of the second customer group data is first determined, and whether the sample size of the second customer group data is greater than the preset sample size threshold is evaluated. Furthermore, if the evaluation result is that the sample size of the second customer group data is greater than or equal to the preset sample size threshold, the second customer group data is recalled based on the trained recall model to obtain the recalled customer group data. On the contrary, if the evaluation result is that the sample size of the second customer group data is less than the preset sample size threshold, the second customer group data is determined as the recalled customer group data. The recalled customer group data obtained after the customer group data is amplified can be efficiently and long-term batch-produced from the customer group data to be processed, which facilitates the sorting of the recalled customer group data, so as to sort the customers in the customer group data, so that the batch-produced customers are high-quality customers, and the target customer group data is obtained. The target customer group data can be accurately promoted in a targeted manner, thereby improving the efficiency of intelligent marketing.

[0095] Furthermore, the steps of performing sample size evaluation on the second customer group data, amplifying the customer group data based on the evaluation result, and obtaining the recalled customer group data include:

[0096] Step S231: Compare the sample size of the second customer group data with a preset sample size threshold to obtain an evaluation result;

[0097] After obtaining the second customer group data, the sample size of the second customer group data is compared with the preset sample size threshold to determine the size relationship between the sample size of the second customer group data and the preset sample size threshold, and obtain an evaluation result including that the sample size of the second customer group data is greater than or equal to the preset sample size threshold, and the sample size of the second customer group data is less than the preset sample size threshold.

[0098] Step S232: If the evaluation result shows that the sample size of the second customer group data is greater than or equal to the preset sample size threshold, the second customer group data is recalled based on the trained recall model to obtain recalled customer group data.

[0099] Furthermore, if the sample size assessment results indicate that the sample size of the second customer group data is greater than or equal to the preset sample size threshold, this indicates that the sample size of the second customer group data is large, and a recall of the second customer group data is required to quickly reduce the user dimension scale and reduce subsequent computational effort and time. Therefore, the second customer group data is input into a pre-trained recall model, which then performs a recall based on the second customer group data, obtaining the recalled customer group data with a reduced sample size.

[0100] Understandably, when the sample size cannot be reduced, the modeling algorithms in ML and MLlib in pyspark can be used for ranking modeling.

[0101] Furthermore, the step of recalling the second customer group data based on the trained recall model to obtain the recalled customer group data includes:

[0102] Step S2321, clustering the sample customer group data to obtain a seed user center;

[0103] Step S2322: Calculate the distance between the second customer group data and the seed user center based on the trained recall model to obtain customer group distance information;

[0104] Step S2323: Recall the second customer group data according to the customer group distance information to obtain recalled customer group data.

[0105] When recalling the second customer group data based on the trained recall model, the positive sample customer group data in the sample customer group data is clustered to obtain the seed user center. Specifically, the kmeans clustering algorithm is used to search for the central clustering point of the positive sample customer group data to reduce the positive sample customer group data to 1 / 8 or 500 of the original number, thereby obtaining the seed user center. Furthermore, the second customer group data and the seed user center are input into the trained recall model, and the recall model is used to calculate the distance between the second customer group data and the seed user center. Specifically, the recall model calculates the Euclidean distance between the second customer group data and the seed user center, i.e., the Euclidean distance between each customer in the second customer group data and any seed user center is calculated, and the recalled customer group data in the second customer group data whose minimum distance to any seed user center is within a distance threshold is obtained. If the recalled customer group data does not meet the requirements, the distance threshold is adjusted and the above steps are iterated to obtain and output the recalled customer group data that meets the requirements, where the distance threshold is a distance value set according to actual requirements.

[0106] It can be understood that before applying the recall model, it is necessary to initialize an instance of the random projection bucketed local sensitive hashing model (i.e., the initial recall model), and input historical users, positive sample users corresponding to historical users, recall threshold (default 50.0, adjusted according to the number of feature groups), maximum number (the maximum value between 100,000 and 20 times the required amplification number), minimum number (the maximum value between 1,000 and 2 times the required amplification number) and other data to train the hash model, so as to obtain a recall model with better results.

[0107] Furthermore, after the step of comparing the sample size of the second customer group data with a preset sample size threshold to obtain an evaluation result, the method further includes:

[0108] Step S233: If the evaluation result is that the sample size of the second customer group data is smaller than the preset sample size threshold, the second customer group data is determined as recalled customer group data.

[0109] Furthermore, if the sample size evaluation result obtained by comparison is that the sample size of the second customer group data is less than the preset sample size threshold, indicating that the sample size of the second customer group data is small, the second customer group data is directly determined as the recalled customer group data.

[0110] This embodiment first filters customer group data based on the processing scenario in the customer group data processing instruction to obtain first customer group data; then, based on the customer group data to be amplified, samples the first customer group data to obtain sample customer group data, and filters the sample customer group data from the first customer group data to obtain second customer group data; finally, a sample size assessment is performed on the second customer group data, and customer group data amplification is performed based on the assessment results to obtain recalled customer group data. This method allows for efficient and long-term batch production of recalled customer group data from the customer group data to be processed, facilitating customer group data sorting for the recalled customer group data, thereby ranking the customers in the customer group data and ensuring that the batch-produced customers are high-quality customers. This results in target customer group data, which can then be accurately targeted for promotion, improving the efficiency of intelligent marketing.

[0111] Furthermore, based on the first embodiment of the customer group data processing method of the present application, a third embodiment of the customer group data processing method of the present application is proposed. In the third embodiment, before the step of sorting the recalled customer group data to obtain the target customer group data, the method further includes:

[0112] Step A, obtaining the customer group data processing mode in the customer group data processing instruction;

[0113] After obtaining the recalled customer group data by performing customer group data amplification based on the customer group data to be amplified, the customer group data processing mode contained in the task parameters is obtained according to the analysis of the customer group data processing instructions, wherein the customer group data processing mode in this embodiment may include a seed list mode and a historical strategy mode, that is, identifying whether the customer group data processing mode is a seed list mode or a historical strategy mode.

[0114] Furthermore, the step of sorting the recalled customer group data to obtain target customer group data includes:

[0115] Step B: If the customer group data processing mode is the historical strategy mode, the recalled customer group data is predicted using the trained sorting model, and the recalled customer group data is sorted and filtered according to the prediction result to obtain the target customer group data.

[0116] Furthermore, the recalled customer group data is processed according to the seed list mode or the historical strategy mode to obtain the target customer group data. Specifically, if the customer group data processing mode is the historical strategy mode, the classification model instance is initialized, the model instance and the parameter table are input to initialize the grid search instance, and the trained positive and negative sample features (i.e., the features corresponding to the sample customer group data) are input into the grid search for training to search for the best model parameters. The best sorting model after the search is used to predict the recalled user list (i.e., the recalled customer group data), the prediction results are output and sorted by the predicted value, and the output is limited to twice the required number (the required number is the number of people in the desired amplified list input by the user). Further, the output amplified user list is matched with the user channel preference table, the user's channel preferences are mapped, the users who meet the channel requirements are filtered out, and the number is limited to the required number to obtain the target customer group data representing the amplified user list.

[0117] Understandably, before applying the ranking model, it is necessary to first obtain the feature vectors of seed users and positive and negative sampled users from the historical data as a training dataset. The training dataset is then divided into training samples, validation samples, and test samples. The initial ranking algorithm is trained using the training samples, and then the trained ranking algorithm is validated using the validation samples to optimize the ranking algorithm parameters, obtain the optimal hyperparameters, and achieve model tuning. Finally, the multiple ranking models formed by the optimized ranking algorithm are evaluated using the test samples to obtain the current optimal ranking model. When needed, the trained ranking model can be used to predict the recalled customer data.

[0118] Furthermore, after the step of obtaining the customer group data processing mode in the customer group data processing instruction, the method further includes:

[0119] Step C: If the customer group data processing mode is the seed list mode, filtering the recalled customer group data to obtain target customer group data.

[0120] Furthermore, if it is determined that the customer group data processing mode is the seed list mode, the recalled customer group data is filtered to twice the required quantity and output to obtain the target customer group data.

[0121] This embodiment can sort the recalled customer group data according to the customer group data processing mode to sort the customers in the customer group data, so that the customers produced in batches are high-quality customers, and the target customer group data is obtained. The target customer group data can be accurately promoted in a targeted manner to improve the efficiency of intelligent marketing.

[0122] Furthermore, the present application also provides a customer data processing device.

[0123] Reference Figure 4 , Figure 4This is a functional module diagram of the first embodiment of the customer data processing device of this application.

[0124] The customer group data processing device includes:

[0125] An acquisition module 10 is configured to receive a customer group data processing instruction and acquire the customer group data to be amplified according to the customer group data processing instruction;

[0126] an amplification module 20 for amplifying customer group data based on the customer group data to be amplified to obtain recalled customer group data;

[0127] The sorting module 30 is used to sort the recalled customer group data to obtain target customer group data.

[0128] In addition, the present application also provides a computer-readable storage medium on which a customer group data processing program is stored. When the customer group data processing program is executed by a processor, the steps of each embodiment of the above-mentioned customer group data processing method are implemented.

[0129] In addition, the present application also provides a computer program product, including a computer program, which implements the steps of each embodiment of the above-mentioned customer data processing method when executed by a processor.

[0130] In the embodiments of the customer data processing device, computer-readable storage medium and computer program product of the present application, all technical features of the above-mentioned embodiments of the customer data processing method are included, and the descriptions and explanations are basically the same as those of the above-mentioned embodiments of the customer data processing method, and are not repeated here.

[0131] It should be noted that, in this document, the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, article, or apparatus comprising a series of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article, or apparatus comprising the element.

[0132] The serial numbers of the above embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0133] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus the necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, disk, CD), and includes a number of instructions for enabling a terminal device (which can be a fixed terminal, such as an IoT smart device, including smart air conditioners, smart lights, smart power supplies, smart routers and other smart homes; or a mobile terminal, including smart phones, wearable networked AR / VR devices, smart speakers, self-driving cars and many other networked devices) to execute the methods described in the various embodiments of the present application.

[0134] The above are only preferred embodiments of the present application and do not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the present application specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A customer group data processing method, characterized in that: The customer group data processing method includes: receiving a customer group data processing instruction, and obtaining the customer group data to be amplified according to the customer group data processing instruction; Amplify the customer group data based on the customer group data to be amplified to obtain recalled customer group data; The step of amplifying customer group data based on the customer group data to be amplified to obtain recalled customer group data includes: Filtering customer group data based on the processing scenario in the customer group data processing instruction to obtain first customer group data; sampling the first customer group data based on the customer group data to be amplified to obtain sample customer group data, and filtering the sample customer group data from the first customer group data to obtain second customer group data; evaluating the sample size of the second customer group data, and amplifying the customer group data based on the evaluation result to obtain recalled customer group data; Wherein, after filtering out the sample customer group data from the first customer group data to obtain the second customer group data, the step further includes: Classifying the second customer group data and the sample customer group data based on feature dimensions to obtain a number of basic dimension feature groups; obtaining, from the number of basic dimension feature groups, feature groups corresponding to the second customer group data and feature groups corresponding to the sample customer group data, respectively, and associating the feature groups with customers to amplify customer group data in combination with relevant feature groups for the processing scenario, the number of basic dimension feature groups including user portrait dimension features, scenario context features, and scenario behavior features; The recalled customer group data is sorted to obtain target customer group data.

2. The customer group data processing method according to claim 1, wherein: The steps of evaluating the sample size of the second customer group data, amplifying the customer group data based on the evaluation result, and obtaining the recalled customer group data include: Comparing the sample size of the second customer group data with a preset sample size threshold to obtain an evaluation result; If the evaluation result is that the sample size of the second customer group data is greater than or equal to the preset sample size threshold, customer group data is recalled on the second customer group data based on the trained recall model to obtain recalled customer group data.

3. The customer group data processing method according to claim 2, characterized in that: The step of recalling the second customer group data based on the trained recall model to obtain the recalled customer group data includes: Clustering the sample customer group data to obtain a seed user center; Calculating the distance between the second customer group data and the seed user center based on the trained recall model to obtain customer group distance information; The second customer group data is recalled according to the customer group distance information to obtain recalled customer group data.

4. The customer group data processing method according to claim 2, wherein: After the step of comparing the sample size of the second customer group data with a preset sample size threshold to obtain an evaluation result, the method further includes: If the evaluation result is that the sample size of the second customer group data is smaller than the preset sample size threshold, the second customer group data is determined as recalled customer group data.

5. The customer group data processing method according to claim 1, wherein: Before the step of sorting the recalled customer group data to obtain target customer group data, the method further includes: Obtaining a customer group data processing mode in the customer group data processing instruction; The step of sorting the recalled customer group data to obtain target customer group data includes: If the customer group data processing mode is the historical strategy mode, the recalled customer group data is predicted using the trained sorting model, and the recalled customer group data is sorted and filtered according to the prediction result to obtain the target customer group data.

6. The customer group data processing method according to claim 5, characterized in that: After the step of obtaining the customer group data processing mode in the customer group data processing instruction, the method further includes: If the customer group data processing mode is the seed list mode, the recalled customer group data is filtered to obtain target customer group data.

7. A customer data processing device, characterized in that: The customer group data processing device includes: an acquisition module, configured to receive a customer group data processing instruction and acquire the customer group data to be amplified according to the customer group data processing instruction; an amplification module, configured to amplify customer group data based on the customer group data to be amplified to obtain recalled customer group data; The amplification module is further configured to: filter customer group data based on the processing scenario in the customer group data processing instruction to obtain first customer group data; sample the first customer group data based on the customer group data to be amplified to obtain sample customer group data, and filter the sample customer group data from the first customer group data to obtain second customer group data; perform sample size assessment on the second customer group data, and amplify the customer group data based on the assessment result to obtain recalled customer group data; The amplification module is further configured to: classify the second customer group data and the sample customer group data by feature dimensions to obtain a number of basic dimension feature groups; obtain, from the number of basic dimension feature groups, a feature group corresponding to the second customer group data and a feature group corresponding to the sample customer group data, respectively, and associate the feature groups with customers to amplify the customer group data in combination with relevant feature groups for the processing scenario, wherein the number of basic dimension feature groups includes user portrait dimension features, scenario context features, and scenario behavior features; The sorting module is used to sort the recalled customer group data to obtain target customer group data.

8. A customer data processing device, characterized in that: The customer group data processing device includes a memory, a processor, and a customer group data processing program stored in the memory and executable on the processor. When the customer group data processing program is executed by the processor, the steps of the customer group data processing method according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a customer group data processing program, which, when executed by a processor, implements the steps of the customer group data processing method according to any one of claims 1 to 6.

Images