Data processing method and device, electronic equipment and nonvolatile storage medium

Abstract

The application discloses a data processing method and device, electronic equipment and nonvolatile storage medium. The method comprises the following steps: acquiring multi-source customer data, and performing feature extraction on the multi-source customer data to obtain a customer feature vector; mapping the customer feature vector to a digital holographic coordinate system to obtain customer relationship holographic data, wherein in the customer relationship holographic data, the customer feature vector of each customer is mapped as a hologram element containing amplitude and phase attributes; performing image reconstruction on the customer relationship holographic data to obtain a customer relationship visualization image, wherein the customer relationship visualization image is used to represent the distribution relationship of the customers in the fusion feature space and the deviation degree of the customer data and the benchmark data. The application solves the technical problem that the visualization analysis method in the related art cannot comprehensively reflect the real and complex logical relationship between the multi-source data.

Description

Technical Field

[0001] This application relates to the field of big data analytics technology, and more specifically, to a data processing method, apparatus, electronic device, and non-volatile storage medium. Background Technology

[0002] With the rapid development of data science, big data has become an important part of modern society. Extracting valuable information from massive amounts of data and understanding its underlying meaning to better understand the market, customers, and business is a significant challenge. Visual analytics and insight enhancement technologies, as key means to address this challenge, have been widely researched and applied, mainly including data preprocessing, data visualization, and data insight. Among these, data visualization transforms complex data into graphics and images, leveraging the human visual system's high capacity for understanding visual symbols to help users intuitively perceive data characteristics.

[0003] However, data visualization methods in related technologies have obvious limitations in terms of the depth and interactivity of expressing data relationships: traditional visualization analysis is mostly a simple superposition of two-dimensional data or multiple two-dimensional data, which is difficult to fully reflect the real relationship between data; it usually relies on functions to perform point-to-point data association, which cannot present complex logical relationships; the analysis results are mostly static and fixed display formats, which cannot express the mutuality and dynamic interaction between data.

[0004] There is currently no effective solution to the above problems. Summary of the Invention

[0005] This application provides a data processing method, apparatus, electronic device, and non-volatile storage medium to at least solve the technical problem that visualization analysis methods in related technologies cannot fully reflect the real and complex logical relationships between multi-source data.

[0006] According to one aspect of the embodiments of this application, a data processing method is provided, comprising: acquiring multi-source customer data and extracting features from the multi-source customer data to obtain customer feature vectors, wherein the multi-source customer data includes heterogeneous data from multiple sources or of different types corresponding to multiple customers, and each customer corresponds to a customer feature vector; mapping the customer feature vectors to a digital holographic coordinate system to obtain customer relationship holographic data, wherein in the customer relationship holographic data, the customer feature vector of each customer is mapped to a holographic primitive containing amplitude attributes and phase attributes, the amplitude attribute being used to characterize the intensity of customer attribute values, and the phase attribute being used to characterize customer difference attributes; and performing image reconstruction on the customer relationship holographic data to obtain a customer relationship visualization image, wherein the image reconstruction is used to restore the complete information of the customer data based on the amplitude attributes and phase attributes, and the customer relationship visualization image is used to characterize the distribution relationship of customers in the fused feature space and the degree of deviation between the customer data and the benchmark data.

[0007] Optionally, feature extraction from multi-source customer data to obtain a customer feature vector includes: obtaining heterogeneous data of the same customer from multiple data sources to obtain multi-source customer data, wherein the data dimensions of the multi-source customer data include at least one of the following: consumption data, behavioral data, feedback data, and basic data; extracting features from the multi-source customer data of the same customer according to the data dimensions to obtain feature sub-vectors corresponding to each data dimension; and concatenating the feature sub-vectors corresponding to different data dimensions of the same customer to obtain a customer feature vector.

[0008] Optionally, mapping customer feature vectors to a digital holographic coordinate system to obtain customer relationship holographic data includes: establishing a digital holographic coordinate system comprising an object plane and a holographic recording plane, wherein the object plane corresponds to the original feature space, and the holographic recording plane corresponds to the fused feature space after data fusion processing; mapping each customer's customer feature vector to a first object optical complex amplitude on the object plane, wherein the first object optical complex amplitude is used to characterize the joint expression of data intensity and differential attributes of customer data in the original feature space; propagating and transforming the first object optical complex amplitude to the holographic recording plane through a propagation operator to obtain a second object optical complex amplitude on the holographic recording plane, wherein the second object optical complex amplitude is used to characterize the fused feature expression of customer data after propagation transformation; and superimposing and interfering the second object optical complex amplitude with a reference optical complex amplitude on the holographic recording plane to obtain customer relationship holographic data, wherein the reference optical complex amplitude is used to characterize the reference data corresponding to the customer data, and the superimposed interference is used to generate an interference term carrying phase difference information between the customer data and the reference data to measure the degree of deviation of the customer object relative to the reference.

[0009] Optionally, mapping each customer's feature vector to the first object light complex amplitude on the object plane includes: selecting key feature dimensions from the customer feature vector as the customer's coordinate information on the object plane; calculating the amplitude attribute corresponding to the customer based on the weight coefficients corresponding to each data dimension in the multi-source customer data and the customer's numerical performance in each data dimension; determining the phase attribute corresponding to the customer based on the customer's differentiated labels, wherein the differentiated labels are used to characterize at least one of the customer's preference type, risk level, or business status, and different customer preference types, risk levels, or business statuses correspond to different phase values; and constructing a complex amplitude expression based on the coordinate information, amplitude attribute, and phase attribute to obtain the first object light complex amplitude.

[0010] Optionally, propagating and transforming the first object optical complex amplitude to the holographic recording plane using a propagation operator to obtain the second object optical complex amplitude on the holographic recording plane includes: using a propagation operator to perform information loss compensation processing on the first object optical complex amplitude, wherein the information loss compensation processing is used to compensate for information loss in the process of multi-source customer data from the original channel to the processing layer; performing dimensional adjustment processing on the first object optical complex amplitude after information loss compensation processing to map the customer data in the original feature space to the corresponding coordinates in the fused feature space, wherein the dimensional adjustment processing includes at least one of the following: dimensionality reduction, feature scaling, and feature weight allocation; performing frequency domain fusion processing on the first object optical complex amplitude after dimensional adjustment processing, wherein the frequency domain fusion processing is used to achieve frequency feature superposition and fusion of data from different sources by converting the spatial domain data signal to the frequency domain; and mapping the first object optical complex amplitude after frequency domain fusion processing to the holographic recording plane to obtain the second object optical complex amplitude.

[0011] Optionally, on the holographic recording plane, the second object light complex amplitude and the reference light complex amplitude are superimposed and interfered to obtain customer relationship holographic data. This includes: superimposing and interfering the second object light complex amplitude and the reference light complex amplitude to obtain an interference light intensity distribution, wherein the interference light intensity distribution includes: an object light intrinsic intensity term, a reference light intrinsic intensity term, and an interference term. The object light intrinsic intensity term is used to characterize the intrinsic intensity of the customer data, the reference light intrinsic intensity term is used to characterize the intrinsic stability of the data reference, and the interference term is used to characterize the deviation of the customer data from the reference data; these are introduced sequentially. Multiple different phase shifts are used to obtain a set of interference intensity distributions at each phase shift, resulting in multiple sets of holographic intensity data. The phase shifts are used to adjust the phase of the reference optical complex amplitude to obtain the correlation features between customer data and reference data from different calibration perspectives. Based on the multiple sets of holographic intensity data, a multi-step phase shift algorithm is used to recover the complete phase information and complete amplitude information of the second object optical complex amplitude on the holographic recording plane, reconstructing the complete complex amplitude expression of the second object optical complex amplitude. Based on the complete complex amplitude expressions corresponding to multiple customers, customer relationship holographic data is generated.

[0012] Optionally, image reconstruction of customer relationship holographic data to obtain a customer relationship visualization image includes: propagating the complete complex amplitude expression of the customer relationship holographic data from the holographic recording plane to the object plane through a backpropagation transformation, wherein the backpropagation transformation is used to restore the fused feature data on the holographic recording plane to the distribution relationship of customers in the original feature space; on the object plane, determining the primitive brightness based on the amplitude component of the complete complex amplitude expression, determining the primitive phase feature based on the phase component of the complete complex amplitude expression, and determining the primitive position distribution based on the customer's feature coordinates, wherein the primitive brightness is used to characterize the intensity of the customer's attribute value, the primitive phase feature is used to characterize the customer's differentiated attributes, and the primitive position distribution is used to characterize the spatial distribution relationship of customers in the fused feature space; and fusing and rendering the primitive position distribution, primitive brightness, and primitive phase feature to generate a customer relationship visualization image.

[0013] Optionally, after obtaining the customer relationship visualization image, the method further includes: employing an insight enhancement network to perform deep insight analysis on the customer relationship visualization image and outputting deep insight results. The insight enhancement network is used to explore and analyze the customer relationship visualization image from different angles and dimensions to obtain deeper insights and understanding. The insight enhancement network is trained using a hierarchical frozen training strategy. The deep insight results include at least one of the following: customer segmentation results, preference mining results, and risk warning results. Based on the deep insight results, an automated service strategy is triggered. The automated service strategy includes at least one of the following: pushing churn intervention services to high-risk customers, pushing precise recommendation services to preferred customers, and pushing service upgrade plans to specific customer groups.

[0014] According to another aspect of the embodiments of this application, a data processing apparatus is also provided, comprising: a data representation module, configured to acquire multi-source customer data and extract features from the multi-source customer data to obtain customer feature vectors, wherein the multi-source customer data includes heterogeneous data from multiple sources or of different types corresponding to multiple customers, and each customer corresponds to a customer feature vector; a mapping processing module, configured to map the customer feature vectors to a digital holographic coordinate system to obtain customer relationship holographic data, wherein in the customer relationship holographic data, the customer feature vector of each customer is mapped to a holographic primitive containing amplitude attributes and phase attributes, the amplitude attribute being used to characterize the intensity of customer attribute values, and the phase attribute being used to characterize customer difference attributes; and a visualization module, configured to perform image reconstruction on the customer relationship holographic data to obtain a customer relationship visualization image, wherein the image reconstruction is used to restore the complete information of the customer data based on the amplitude attributes and phase attributes, and the customer relationship visualization image is used to characterize the distribution relationship of customers in the fused feature space and the degree of deviation between the customer data and the reference data.

[0015] According to another aspect of the embodiments of this application, an electronic device is also provided, including: a memory and a processor, the processor being configured to run a program stored in the memory, wherein the program executes a data processing method during runtime.

[0016] According to another aspect of the embodiments of this application, a non-volatile storage medium is also provided, the non-volatile storage medium including a stored computer program, wherein the device where the non-volatile storage medium is located executes a data processing method by running the computer program.

[0017] According to another aspect of the embodiments of this application, a computer program product is also provided, including a computer program that, when executed by a processor, implements the steps of a data processing method.

[0018] In this embodiment, multi-source customer data is acquired, and features are extracted from the multi-source customer data to obtain customer feature vectors. The multi-source customer data includes heterogeneous data from multiple sources or of different types corresponding to multiple customers, with each customer corresponding to a customer feature vector. These customer feature vectors are mapped to a digital holographic coordinate system to obtain customer relationship holographic data. In this customer relationship holographic data, each customer's feature vector is mapped to a holographic primitive containing amplitude and phase attributes. The amplitude attribute characterizes the intensity of the customer attribute value, and the phase attribute characterizes the customer's difference attribute. Image reconstruction is then performed on the customer relationship holographic data to obtain a visualized image of the customer relationship. Image reconstruction is used to restore the complete information of customer data based on amplitude and phase attributes. Customer relationship visualization images are used to characterize the distribution relationship of customers in the fused feature space and the degree of deviation between customer data and benchmark data. By mapping multi-source heterogeneous customer data into holographic primitives containing amplitude and phase attributes, and using propagation transformation and interference calculation in the principle of optical holography to achieve data fusion and phase recovery, the goal of completely preserving the strength of customer data attributes, differentiated features and deviation relationship from the benchmark in a unified multi-dimensional feature space is achieved. This solves the technical problem that visualization analysis methods in related technologies cannot fully reflect the real and complex logical relationship between multi-source data. Attached Figure Description

[0019] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0020] Figure 1 This is a hardware structure block diagram of a computer terminal (or electronic device) for implementing a data processing method according to an embodiment of this application;

[0021] Figure 2This is a schematic diagram of a data processing method flow according to an embodiment of this application;

[0022] Figure 3 This is a schematic diagram of a digital holographic coordinate system provided according to an embodiment of this application;

[0023] Figure 4 This is a schematic diagram of a network structure for enhancing digital holograms at multiple scales, according to an embodiment of this application.

[0024] Figure 5 This is a schematic diagram of the structure of a data processing device provided according to an embodiment of this application. Detailed Implementation

[0025] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0026] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0027] In related technologies, data visualization analysis generally employs functions for data association and uses graphs for visualization. These methods abstract and simplify data, transforming it into graphical or image-based presentations to improve readability and understandability. Examples include tables, histograms, scatter plots, line charts, bar charts, pie charts, area charts, flowcharts, bubble charts, and multiple data series or combinations of charts, such as timelines, Venn diagrams, data flow diagrams, and entity relationship diagrams. However, most of these visualization methods cannot fully present the characteristics and details of the data, failing to meet the needs of processing and analyzing large-scale, complex data. Furthermore, data insights struggle to automatically extract all useful information. The main drawbacks are as follows:

[0028] 1) Traditional visualization analysis is based on two-dimensional data or the superposition of multiple two-dimensional data, which cannot fully reflect the true relationship between the data.

[0029] 2) Generally, using functions, such as the VLOOKUP function, to find relationships between data only represents point-to-point relationships; complex logical relationships cannot be represented.

[0030] 3) The analysis results are generally static and fixed, and cannot express the interrelationships and complex interactions between data.

[0031] To address the aforementioned issues, this application provides a solution that utilizes the fundamental principles of holographic digital visualization technology. Each data item in the database is represented as a single primitive element, with a large dataset forming a data image. Simultaneously, the various attribute values ​​of the data are represented in a multidimensional format, allowing for observation and deeper analysis of the data from different dimensions. Holographic digital visualization technology transforms data into a visible form, highlighting important features, including commonalities and anomalies. These visualizations enable users to easily and quickly perceive salient aspects of their data. Visual representation enhances cognitive reasoning through perceptual reasoning, making analytical reasoning faster and more focused. A detailed explanation follows.

[0032] According to an embodiment of this application, a method embodiment for data processing is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0033] The methods and embodiments provided in this application can be executed on mobile terminals, computer terminals, or similar computing devices. Figure 1 A hardware structure block diagram of a computer terminal (or electronic device) for implementing a data processing method is shown. Figure 1 As shown, the computer terminal 10 (or electronic device) may include one or more processors 102 (shown as 102a, 102b, ..., 102n in the figure) 102 (processor 102 may include, but is not limited to, a microprocessor MCU or a programmable logic device FPGA, etc.), a memory 104 for storing data, and a transmission device 106 for communication functions. In addition, it may also include: a display, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of a BUS bus), a network interface, a power supply, and / or a camera. Those skilled in the art will understand that... Figure 1The structure shown is for illustrative purposes only and does not limit the structure of the aforementioned electronic device. For example, computer terminal 10 may also include... Figure 1 The more or fewer components shown, or having the same Figure 1 The different configurations shown.

[0034] It should be noted that the aforementioned one or more processors 102 and / or other data processing circuits are generally referred to herein as "data processing circuits". These data processing circuits may be embodied, in whole or in part, in software, hardware, firmware, or any other combination thereof. Furthermore, the data processing circuits may be a single, independent processing module, or may be integrated, in whole or in part, into any other element within the computer terminal 10 (or electronic device). As involved in the embodiments of this application, the data processing circuits serve as a processor control mechanism (e.g., selection of a variable resistor termination path connected to an interface).

[0035] The memory 104 can be used to store software programs and modules of application software, such as program instructions / data storage devices corresponding to the data processing method in this embodiment. The processor 102 executes various functional applications and data processing by running the software programs and modules stored in the memory 104, thereby implementing the aforementioned data processing method. The memory 104 may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include memory remotely located relative to the processor 102, and these remote memories can be connected to the computer terminal 10 via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0036] The transmission device 106 is used to receive or send data via a network. Specific examples of the network described above may include a wireless network provided by the communication provider of the computer terminal 10. In one example, the transmission device 106 includes a Network Interface Controller (NIC), which can connect to other network devices via a base station to communicate with the Internet. In another example, the transmission device 106 may be a Radio Frequency (RF) module, used for wireless communication with the Internet.

[0037] The display can be, for example, a touchscreen liquid crystal display (LCD) that allows the user to interact with the user interface of the computer terminal 10 (or electronic device).

[0038] Under the above operating environment, this application provides a data processing method. Figure 2This is a schematic diagram of a data processing method flow provided according to an embodiment of this application, such as... Figure 2 As shown, the method includes the following steps:

[0039] Step S202: Obtain multi-source customer data and extract features from the multi-source customer data to obtain customer feature vectors. The multi-source customer data includes heterogeneous data from multiple sources or of different types corresponding to multiple customers, and each customer corresponds to a customer feature vector.

[0040] Step S204: Map the customer feature vector to the digital holographic coordinate system to obtain customer relationship holographic data. In the customer relationship holographic data, the customer feature vector of each customer is mapped to a holographic primitive containing amplitude and phase attributes. The amplitude attribute is used to characterize the intensity of the customer attribute value, and the phase attribute is used to characterize the customer difference attribute.

[0041] Step S206: Perform image reconstruction on the holographic data of customer relationships to obtain a visual image of customer relationships. The image reconstruction is used to restore the complete information of customer data based on amplitude and phase attributes. The visual image of customer relationships is used to characterize the distribution relationship of customers in the fused feature space and the degree of deviation between customer data and benchmark data.

[0042] Through the above steps, by mapping multi-source heterogeneous customer data into holographic primitives containing amplitude and phase attributes, and by using propagation transformation and interference calculation in the principle of optical holography to achieve data fusion and phase recovery, the goal of completely preserving the strength of customer data attributes, differentiated features, and deviation from the benchmark in a unified multi-dimensional feature space is achieved. This solves the technical problem that visualization analysis methods in related technologies cannot fully reflect the real and complex logical relationships between multi-source data.

[0043] The data processing method in steps S202 to S206 of the embodiments of this application will be further described below.

[0044] This application embodiment extracts multi-source heterogeneous customer data into a unified high-dimensional feature vector and introduces a digital holographic coordinate system to map the customer data into holographic primitives containing amplitude and phase attributes. After propagation transformation, interferometric calculation, and image reconstruction, a customer relationship visualization image is generated. This method breaks through the limitations of traditional visualization analysis, which is confined to two-dimensional static display and can only perform point-to-point functional correlations. By simultaneously encoding the attribute strength and differential characteristics of the data in the complex domain and using an interferometric mechanism to quantify the deviation of the data from the benchmark, it achieves a three-dimensional dynamic representation of complete customer data information in a unified multi-dimensional feature space. Details are as follows.

[0045] First, after acquiring multi-source customer data, it is extracted into high-dimensional feature vectors for data representation, as follows.

[0046] In some embodiments of this application, feature extraction of multi-source customer data to obtain customer feature vectors includes: obtaining heterogeneous data of the same customer from multiple data sources to obtain multi-source customer data, wherein the data dimensions of the multi-source customer data include at least one of the following: consumption data, behavioral data, feedback data, and basic data; performing feature extraction on the multi-source customer data of the same customer according to the data dimensions to obtain feature sub-vectors corresponding to each data dimension; and concatenating the feature sub-vectors corresponding to different data dimensions of the same customer to obtain the customer feature vector.

[0047] Representing multi-source customer data is a fundamental step in achieving multi-source data fusion. Data in customer relationship management (CRM) scenarios is often scattered across multiple business systems, characterized by diverse sources and types. Directly analyzing this heterogeneous data will fail to achieve effective fusion due to inconsistent formats and units. Therefore, it is necessary to first convert various raw data into a unified-dimensional numerical representation, namely a customer feature vector. This assigns each customer a specific location coordinate in mathematical space, providing a structured data foundation for subsequent mapping to a digital holographic coordinate system.

[0048] Specifically, taking the scenario of segmented operation of individual customers by telecommunications operators as an example, the data dimensions of multi-source customer data can include, but are not limited to: consumption data, behavioral data, feedback data, and basic data. Consumption data is structured data, including the past 12 months' package fees, value-added service consumption, payment frequency and amount, and outstanding payment records; behavioral data is semi-structured data, including APP login frequency, service channels, data usage distribution, and broadband connection duration; feedback data is multimodal data such as text and images, including customer service call recordings transcribed into text, APP feedback screenshots, and questionnaire survey results; basic data is structured data, including customer age, gender, city, and years of service. These multiple types of data collectively constitute the multi-source customer data for the same customer.

[0049] In this embodiment, feature extraction can be performed on multi-source customer data of the same customer according to data dimensions to obtain feature sub-vectors corresponding to each data dimension. The specific implementation method varies depending on the data type. For example, for consumption data, 15-dimensional numerical features such as average monthly consumption, frequency of package upgrades, proportion of value-added service consumption, and number of overdue payments can be extracted, and the difference in dimensions can be eliminated through Min-Max standardization to obtain the first feature sub-vector. For behavioral data, 20-dimensional sequence features such as weekly active days of the APP, peak traffic periods, and business processing response time can be extracted using time series models such as LSTM to obtain the second feature sub-vector. For feedback data, text data can generate a 768-dimensional semantic embedding vector through large language models such as LLaMA 2, and image data can be extracted with a 512-dimensional visual feature vector through CNN. The two are then merged to obtain the third feature sub-vector. For basic data, age segments, city levels, and network access years ranges can be converted into 10-dimensional coded features through discretization encoding to obtain the fourth feature sub-vector.

[0050] Then, the feature sub-vectors corresponding to different data dimensions of the same customer can be concatenated to obtain the customer feature vector. For example, the aforementioned 15-dimensional consumption feature sub-vector, 20-dimensional behavior feature sub-vector, 1280-dimensional feedback feature sub-vector, and 10-dimensional basic feature sub-vector can be concatenated to finally merge into a 1325-dimensional customer feature vector. This customer feature vector fully retains the core information of consumption, behavior, feedback, and basic data, and can serve as the data foundation for mapping the customer feature vector to a digital holographic coordinate system. This allows each customer to have a unique numerical representation in the multi-dimensional feature space, providing a computable structured input for subsequent holographic primitive generation, propagation transformation, and interference calculation.

[0051] After obtaining the structured data foundation of customer feature vectors, we can proceed to the core stage of multi-source data fusion processing. The following section details how to map customer feature vectors to a digital holographic coordinate system, and how to generate holographic data of customer relationships through propagation transformation and interferometric calculations, thereby preserving and reconstructing complete complex amplitude information of customer data.

[0052] In some embodiments of this application, mapping customer feature vectors to a digital holographic coordinate system to obtain customer relationship holographic data includes: establishing a digital holographic coordinate system comprising an object plane and a holographic recording plane, wherein the object plane corresponds to the original feature space, and the holographic recording plane corresponds to the fused feature space after data fusion processing; mapping each customer's customer feature vector to a first object optical complex amplitude on the object plane, wherein the first object optical complex amplitude is used to characterize the joint expression of data intensity and differential attributes of customer data in the original feature space; propagating and transforming the first object optical complex amplitude to the holographic recording plane through a propagation operator to obtain a second object optical complex amplitude on the holographic recording plane, wherein the second object optical complex amplitude is used to characterize the fused feature expression of customer data after propagation transformation; and superimposing and interfering the second object optical complex amplitude with a reference optical complex amplitude on the holographic recording plane to obtain customer relationship holographic data, wherein the reference optical complex amplitude is used to characterize the reference data corresponding to the customer data, and the superimposed interference is used to generate an interference term carrying phase difference information between the customer data and the reference data to measure the degree of deviation of the customer object relative to the reference.

[0053] In this embodiment of the application, to achieve deep fusion and visual representation of multi-source heterogeneous customer data, a digital holographic coordinate system can be established first. For example... Figure 3 As shown, the coordinate system includes the plane where the recorded object is located (i.e., the object plane) O and the plane where the CCD is located (i.e., the holographic recording plane) H, and the two planes are separated by a distance d. Light emitted from the object (in this embodiment, this refers to the complex amplitude expression of the client data in this feature space) travels a distance d along the Z-axis and is then received by the CCD, and the data is processed using... and Let the complex amplitude distribution of the light emitted by the object on the object plane and the holographic recording plane be denoted as follows: .

[0054] It should be noted that the recorded object here is not a physical entity, but a customer data object formed after data representation. The object plane O corresponds to the feature space where the original multi-source data is located, that is, the high-dimensional space where customer feature vectors are distributed. The holographic recording plane H corresponds to the fused feature space after the data has been fused. The coordinates of the object plane can correspond to key feature dimensions such as average monthly spending and number of active days of the app, either in their original or reduced dimensions, while the coordinates of the holographic plane can correspond to fused feature dimensions such as the consumption-behavior composite value and customer lifecycle stage. The distance d between the two planes is not a physical distance, but rather represents the processing mapping process from the original multi-source data to the fused data image, covering operations such as data cleaning, normalization, feature weighting, dimensionality reduction, frequency domain fusion, and benchmark calibration.

[0055] The specific formulas are shown in formulas (1), (2), and (3) below.

[0056] = exp(i (1)

[0057] = exp(i (2)

[0058] = exp( (3)

[0059] In formula (1), The first object light complex amplitude on object plane O is represented. In the data scenario of this application embodiment, object plane O corresponds to the original multi-source feature space. This formula is used to represent the original customer data primitives obtained in the data representation step as complex numbers. Wherein, These are the object plane coordinates, corresponding to the feature dimensions of the original data, for example... This can be mapped to customer spending amounts. This can be mapped to the number of customer inquiries; The amplitude represents the strength or importance of the data, such as the weight of a customer's spending or the business share of that customer group. Phase represents the differentiated attributes of the data, such as customer preference tags. It can be matched with the "data plan". This corresponds to "broadband preference". The phase is transformed into a complex form using Euler's formula, which facilitates subsequent fusion calculations.

[0060] In formula (2), The second object light complex amplitude is represented on the holographic recording plane H. In this embodiment, the holographic recording plane H represents the fused feature mapping space. This formula describes the complex value state of the original multi-source data in the fused space after cleaning, standardization, weighting, dimensionality reduction, or fusion processing. These are the holographic plane coordinates, corresponding to the feature dimensions of the data processing layer, for example... This can be mapped to the integrated consumption-behavior value. This can be mapped to customer lifecycle stages; and These represent the strength and variance characteristics of the processed data, such as the normalized strength of customer consumption after preprocessing and the corrected preference labels.

[0061] In formula (3), The reference light amplitude is indicated in this embodiment. In this application, the reference light is not a physical light source, but rather serves as a unified benchmark for data processing, such as an industry average, customer group average, standardized rules, or business calibration parameters. The reference light amplitude corresponds to the weight of the benchmark, such as the scaling factor during normalization; The reference light phase corresponds to the calibration parameters of the benchmark, such as correction values ​​for data deviations or system error compensation for data from different channels. The role of the reference light is to provide a comparison benchmark for customer data elements, enabling the system to extract the degree of deviation, risk status, or value difference of the customer relative to the benchmark.

[0062] Through the above three formulas, the light field propagation, reference light interference, and phase recovery processes in this application embodiment correspond to multi-source feature mapping, correlation calculation of customer data and benchmark data, and reconstruction of hidden relationships such as customer preferences, value intensity, and risk status, respectively. Specifically, formula (1) encodes the original customer data into a complex amplitude expression containing intensity and difference attributes; formula (2) describes the state of the complex amplitude in the fusion space after being transformed by the propagation operator; formula (3) introduces an industry benchmark as a reference, laying the foundation for subsequent interference calculation on the holographic recording plane and generating customer relationship holographic data carrying phase difference information.

[0063] By establishing and mapping a digital holographic coordinate system, this application's embodiments enable cross-domain migration from optical holography to data holography. Specifically, optical holography preserves the complete three-dimensional information of an object by recording the amplitude and phase of light waves. Similarly, customer relationship data needs to preserve both the "numerical magnitude" (amplitude) and the "difference from the benchmark" (phase) to fully characterize customer features. Utilizing the structure-preserving property of the propagation operator, customers with similar consumption behaviors or risk states in the original feature space maintain similar relationships in the fused feature space, ensuring that the customer association structure is not destroyed during data fusion. Ultimately, through interferometric computation, not only can the correlation measurement between multi-source data and benchmark data be achieved, but the quantitative expression of phase difference also provides a mathematical foundation for subsequent insights into customer preferences, risk state identification, and reconstruction of hidden relationships, effectively addressing the technical limitations of single data relationships and fragmented dimensions in related technologies.

[0064] To further illustrate that the above-mentioned digital holographic coordinate system is not only applicable to physical optics scenarios, the embodiments of this application can also be equivalently described as an algebraic data representation system consisting of an original multi-source feature space, a fused representation space, a structure-preserving mapping, a complex-valued representation function, a reference benchmark, and a multi-dimensional observation transformation, denoted as (V,W,Fd,UO,R,G).

[0065] Here, V represents the original multi-source data feature space. The aforementioned acquired consumption data, behavioral data, feedback data, and basic data can be represented as subspaces Vc, Vb, Vf, and Vs, respectively, and the overall feature space can be represented as V = Vc⊕Vb⊕Vf⊕Vs. The symbol ⊕ represents a direct sum structure, used to illustrate that data from different sources retain their independent attributes while being combined into a unified customer feature vector. W represents the fused holographic representation space, corresponding to the holographic recording plane H in the digital holographic coordinate system. There exists a mapping operator Fd:V→W, used to convert the original multi-source feature vector xi into the fused holographic primitive representation hi = Fd(xi). Here, the parameter d can be equivalently understood as the transformation scale or processing level from the original data space to the fused representation space, used to characterize the comprehensive effect of data cleaning, normalization, weighted fusion, dimensionality reduction, frequency domain transformation, and benchmark calibration.

[0066] For any customer or business object xi, a complex-valued representation function UO(xi) = Aiexp(jΦi) can be defined. This formula is used to transform ordinary multidimensional feature vectors into holographic primitives that simultaneously carry intensity information and differential attributes. Here, Ai is the amplitude, representing customer value, activity level, business weight, or data importance; Φi is the phase, representing customer preference, risk status, business type, or differentiation label. Thus, the "light emitted by an object" in the original optical description can be equivalently represented by a joint amplitude-phase expression of the customer data object in the complex domain.

[0067] The mapping Fd described above is not a simple concatenation of data from different sources, but rather a mapping that preserves the data relationship structure as much as possible. If two customer objects have similar consumption behaviors, similar service preferences, or similar risk states in the original space V, then after the Fd mapping, they should still maintain a similar or comparable relationship in the fusion space W, i.e., ρW(Fd(xi),Fd(xj))≈ρV(xi,xj). Therefore, this mapping is equivalent to the propagation and recording processes in digital holography, both used to preserve the association structure between the original objects in the target space.

[0068] The reference light R can be equivalently represented by a benchmark vector, benchmark function, or standardization rule in the fusion space W, such as industry averages, customer group averages, city service resource correction parameters, or business calibration parameters. The relationship between the customer primitive hi and the reference benchmark R can be represented by the inner product, similarity, or phase difference, for example...<hi, R> Or ΔΦi = Φi - ΦR. This relationship corresponds to the optical interference term, used to measure the degree of deviation of the customer object from the reference.

[0069] G represents a group of observation transformations acting on the fusion representation space W, including operations such as rotation, scaling, projection, dimension switching, time slicing, and customer group filtering. For any g∈G, g "hi" represents observing the same customer primitive from different perspectives. The aforementioned operations such as 3D dynamic interaction, multi-dimensional observation, rotation, scaling, and hierarchical filtering can be abstracted as a transformation group G acting on the fusion representation space W. For the customer primitive "hi", different operations can be represented as g·hi. These operations do not change the customer object itself, but only change the observation method and presentation angle. This is consistent with the mathematical meaning of group action, that is, the process does not change the essential identity of the customer object, but changes its visualization presentation method, corresponding to the process of observing data images from different angles in holographic visualization.

[0070] Therefore, the digital holographic coordinate system can be equivalently converted from optical language to algebraic language: the object plane O corresponds to the original multi-source feature space V; the holographic recording plane H corresponds to the fused representation space W; the complex amplitude of the object light corresponds to the complex-valued representation function UO; the propagation distance d corresponds to the transformation scale of the mapping operator Fd; the reference light corresponds to the reference vector or calibration rule R; and holographic observation and interaction correspond to the transformation group G. This equivalent description does not replace the original optical formulas, but rather explains the mathematical meaning of the formulas in multi-source data processing of customer relationships from the perspective of abstract algebra and data fusion.

[0071] This application borrows the concept of optical holography in data processing, primarily based on the isomorphism of mathematical structures, rather than a simple approximation of physical optical phenomena. The essence of optical holography is a set of mathematical tools describing how wave fields propagate, superimpose, and reconstruct in space. It uses complex numbers (amplitude and phase) to fully characterize the state of a signal and restores the original information through interference recording and diffraction reconstruction. Although customer relationship data is not a physical light wave, it can still be abstracted as an information field propagating in a high-dimensional space: the multi-source feature vectors of each customer constitute a signal source, and the process of data cleaning, dimensionality reduction, and fusion corresponds precisely to the mapping and propagation of the signal from the original space to the fusion space. Therefore, describing this process using optical formulas is essentially using a mature linear system theory to rigorously define the boundary conditions and information conservation relationships of data transformation, making the fusion of multi-source heterogeneous data no longer a simple table stitching, but a structured mapping with geometric and algebraic meaning.

[0072] The following provides a detailed description of each step in the process of mapping customer feature vectors to a digital holographic coordinate system in the embodiments of this application.

[0073] First, the customer feature vector of each customer is mapped to the first object light complex amplitude on the object plane. The specific steps are as follows.

[0074] In some embodiments of this application, mapping each customer's customer feature vector to a first object light complex amplitude on the object plane includes: selecting key feature dimensions from the customer feature vector as the customer's coordinate information on the object plane; calculating the amplitude attribute corresponding to the customer based on the weight coefficients corresponding to each data dimension in the multi-source customer data and the customer's numerical performance in each data dimension; determining the phase attribute corresponding to the customer based on the customer's differentiated label, wherein the differentiated label is used to characterize at least one of the customer's preference type, risk level, or business status, and different customer preference types, risk levels, or business statuses correspond to different phase values; and constructing a complex amplitude expression based on the coordinate information, amplitude attribute, and phase attribute to obtain the first object light complex amplitude.

[0075] In this embodiment, mapping the customer feature vector to the first object light complex amplitude on the object plane is a crucial initial step in multi-source data fusion processing. Traditional real-domain data representation can only characterize numerical magnitude and cannot simultaneously encode the differential attributes of the data. However, the amplitude-phase joint representation in the complex domain can simultaneously preserve the intensity and differences of customer data in the same mathematical object, retaining complete structural information for subsequent holographic interferometry calculations.

[0076] Specifically, key feature dimensions can be selected from the customer feature vector as the customer's coordinate information in the object plane. The object plane corresponds to the original multi-source feature space, and its coordinates can be mapped to the customer's core business indicators, such as average monthly spending, number of active days on the app, frequency of complaints, or package type code, etc., either in their original or reduced form, thus determining the position of the customer element in the original feature space. Based on the determined coordinate information, the amplitude attribute corresponding to the customer can be calculated according to the weight coefficients corresponding to each data dimension in the multi-source customer data, as well as the customer's numerical performance in each data dimension. This amplitude attribute is used to characterize the strength of the customer attribute value, such as customer value, data importance, or business weight. Taking the scenario of individual customers of telecommunications operators as an example, the weight of consumption data can be set to 0.4, behavioral data 0.3, feedback data 0.2, and basic data 0.1. After weighting and aggregating the customer's numerical performance in each dimension, the amplitude of the customer element can be obtained. The larger the amplitude, the higher the business value of the customer and the stronger the data influence, corresponding to a higher brightness or intensity in the hologram.

[0077] Subsequently, the phase attribute of a customer can be determined based on their differentiated tags. These differentiated tags characterize at least one of a customer's preference type, risk level, or business status; different preference types, risk levels, or business statuses correspond to different phase values. For example, a phase value of 0 can correspond to a data plan preference, a phase value of π / 2 to a broadband preference, and a phase value of π to a value-added service preference. Risk status or business type can also be mapped to different phase intervals. By encoding differentiated tags into phases, the system can distinguish customers with different attribute characteristics even when the numerical strength is the same. For example, two customers with the same average monthly spending may have drastically different phase values ​​due to different preference types.

[0078] Finally, based on the aforementioned coordinate information, amplitude attributes, and phase attributes, the system constructs a complex amplitude representation to obtain the first object light complex amplitude. This first object light complex amplitude is a joint amplitude-phase representation of the customer data object in the complex domain, where the amplitude component represents "how important the customer is" and the phase component represents "how the customer is different".

[0079] The amplitude-phase binary representation in optics provides a richer way to encode customer relationship data. In traditional customer analysis, data is usually presented only as real values ​​(such as spending amount, number of active days), which is equivalent to only recording amplitude or intensity, losing the structural differences carried by phase. In optics, phase determines the relative position and interference behavior of waveforms. Mapped to customer data, it can naturally encode differentiated attributes such as customer preference type, risk status, or channel preference. For example, two customers may have the same average monthly spending (equal amplitude), but one is traffic-oriented and the other is broadband-oriented. By assigning different phase values, the system can distinguish the essential differences between the two even when the numerical values ​​are equal. This complex-domain representation decouples the customer's value intensity and attribute differences into two independent mathematical dimensions, avoiding the misjudgment of customers as being of the same type due to similar numerical values ​​in traditional real-number analysis.

[0080] From an abstract algebraic perspective, this complex-valued representation function encodes customer objects in the original multi-source feature space into holographic primitives containing complete information. It preserves the independent attributes of consumption, behavior, feedback, and basic data while combining them into a unified expression of customer characteristics. If two customer objects have similar consumption behaviors or similar service preferences in the original space, their complex amplitude expressions remain mathematically similar, ensuring that the data association structure is not destroyed during subsequent propagation transformations. Through this complex-domain encoding, the system provides input containing complete phase information for subsequent propagation to the holographic recording plane and interference calculations with the reference light. This effectively solves the technical problem of lost differentiated attributes in traditional real-number analysis, achieving the technical effect of preserving complete customer characteristic information at the source of data fusion.

[0081] After obtaining the first object optical complex amplitude, the first object optical complex amplitude can be further propagated and transformed to the holographic recording plane through the propagation operator to obtain the second object optical complex amplitude on the holographic recording plane, as follows.

[0082] In some embodiments of this application, propagating and transforming the first object optical complex amplitude to the holographic recording plane using a propagation operator to obtain the second object optical complex amplitude on the holographic recording plane includes: using a propagation operator to perform information loss compensation processing on the first object optical complex amplitude, wherein the information loss compensation processing is used to compensate for information loss in the process of multi-source customer data from the original channel to the processing layer; performing dimensional adjustment processing on the first object optical complex amplitude after information loss compensation processing to map the customer data in the original feature space to the corresponding coordinates in the fused feature space, wherein the dimensional adjustment processing includes at least one of the following: dimensionality reduction, feature scaling, and feature weight allocation; performing frequency domain fusion processing on the first object optical complex amplitude after dimensional adjustment processing, wherein the frequency domain fusion processing is used to achieve frequency feature superposition and fusion of data from different sources by converting spatial domain data signals to the frequency domain; and mapping the first object optical complex amplitude after frequency domain fusion processing to the holographic recording plane to obtain the second object optical complex amplitude.

[0083] In this embodiment, the propagation operator transforms the first object light complex amplitude to the holographic recording plane to obtain the second object light complex amplitude on the holographic recording plane. This is the core step in mapping multi-source data from the original feature space to the fused feature space. The propagation operator is not a simple linear mapping, but a comprehensive transformation process that includes information loss compensation, dimension adjustment, and frequency domain fusion. Its mathematical essence corresponds to the Fresnel diffraction process of object light waves propagating from the object plane to the holographic recording plane in optical holography, specifically as shown in the following formula (4):

[0084] = exp[ ( + )] (4)

[0085] Specifically, the propagation operator can be used to compensate for information loss in the complex amplitude of the first object light. This information loss compensation process is used to compensate for the information loss in the process of multi-source customer data from the original channel to the processing layer, corresponding to the leading term in the above formula (4). In an optical sense, this term represents the amplitude attenuation and phase shift of light after it has traveled a certain distance. In a data context, it is equivalent to compensating for information loss when multi-source data is transmitted from the original channel (object plane) to the processing layer (holographic plane). For example, it can be used to fill in missing values ​​in APP login records, correct semantic deviations in feedback text, and filter cross-channel noise to ensure the integrity and accuracy of the original customer data, laying a reliable foundation for subsequent fusion calculations.

[0086] Subsequently, the system performs dimensionality adjustment processing on the first object light complex amplitude after information loss compensation processing to map the customer data in the original feature space to the corresponding coordinates in the fused feature space. This dimensionality adjustment processing includes at least one of dimensionality reduction, feature scaling, and feature weight allocation, corresponding to the phase term exp[ in the above formula (4). ( + )]and In optical terms, this term represents the phase change of light due to path difference during propagation; in data scenarios, it is equivalent to dimensional adjustment of multi-source data. For example, compressing a customer's 1325-dimensional high-dimensional attributes into two-dimensional coordinates centered on the comprehensive value of consumption and behavior and the customer lifecycle stage, eliminating dimensional differences through feature scaling, and reflecting the differences in importance of different data dimensions through feature weight allocation, so that customer data can obtain a concise and business-meaning coordinate expression in the fused feature space.

[0087] Then, the system can perform frequency domain fusion processing on the first object's complex amplitude after dimension adjustment. This frequency domain fusion processing is used to convert spatial domain data signals to the frequency domain, thereby achieving the superposition and fusion of frequency features from different sources, corresponding to the Fourier transform term F{} in the above formula. In an optical sense, the Fourier transform converts the optical signal in the spatial domain (x,y) into a frequency domain signal, simplifying propagation calculations; in data scenarios, it is equivalent to frequency domain fusion of multi-source data, such as extracting high-frequency interaction features from customer data on an APP and low-frequency stability features from offline business halls, allowing the frequency features of different sources to be independently superimposed in the frequency domain, avoiding conflicts and interference in time domain fusion, and then reconstructing a unified fusion feature expression through inverse transformation, thereby achieving true multi-source lossless fusion.

[0088] Finally, the system maps the first object optical complex amplitude, after frequency domain fusion processing, onto the holographic recording plane to obtain the second object optical complex amplitude. This second object optical complex amplitude characterizes the fused feature expression of customer data after cleaning, normalization, feature weighting, dimensionality reduction, and frequency domain fusion. Its coordinates on the holographic recording plane reflect the customer's new position in the fused feature space. From the perspective of structure-preserving mapping, this propagation operator ensures that if two customer objects have similar consumption behaviors, similar service preferences, or similar risk states in the original feature space, they will still maintain a similar or comparable relationship in the fused feature space after mapping, thereby ensuring that the customer association structure is not destroyed during the data fusion process.

[0089] Through the aforementioned propagation transformation, the embodiments of this application can achieve a structured mapping of multi-source heterogeneous data from the original feature space to the fused feature space. The propagation process in optical holography is essentially a spatial transformation of the information field, and customer relationship data also requires systematic transformation processing to convert its dispersed original state into a unified fused state that can be analyzed uniformly. By employing a three-level processing approach—information loss compensation, dimension adjustment, and frequency domain fusion—the system not only preserves the complete information of the customer data but also solves the problem of feature conflicts among multi-source data through frequency domain separation and superposition. This achieves the technical effect of completely preserving the strength and differentiated characteristics of customer attributes in a unified fusion space, providing a reliable fusion feature input for subsequent interference calculations with the reference light on the holographic recording plane and the generation of customer relationship holographic data carrying phase difference information.

[0090] Furthermore, the complex amplitude of the second object light can be superimposed and interfered with the complex amplitude of the reference light on the holographic recording plane to obtain holographic data of customer relationship, as follows.

[0091] In some embodiments of this application, on the holographic recording plane, superimposing and interfering the second object light complex amplitude and the reference light complex amplitude to obtain customer relationship holographic data includes: superimposing and interfering the second object light complex amplitude and the reference light complex amplitude to obtain an interference light intensity distribution, wherein the interference light intensity distribution includes: an object light intrinsic intensity term, a reference light intrinsic intensity term, and an interference term, the object light intrinsic intensity term is used to characterize the intrinsic intensity of the customer data, the reference light intrinsic intensity term is used to characterize the intrinsic stability of the data reference, and the interference term is used to characterize the deviation of the customer data from the reference data; Multiple different phase shifts are introduced sequentially, and a set of interference intensity distributions is obtained under each phase shift, resulting in multiple sets of holographic intensity data. The phase shifts are used to adjust the phase of the reference optical complex amplitude to obtain the correlation features between customer data and reference data from different calibration perspectives. Based on the multiple sets of holographic intensity data, the complete phase information and complete amplitude information of the second object optical complex amplitude on the holographic recording plane are recovered through a multi-step phase shift algorithm, and the complete complex amplitude expression of the second object optical complex amplitude is reconstructed. Based on the complete complex amplitude expressions corresponding to multiple customers, customer relationship holographic data is generated.

[0092] Specifically, the complex amplitude of the second object light, which characterizes the fusion feature expression of the client, can be superimposed with the complex amplitude of the reference light, which characterizes the data benchmark, to obtain the interference light intensity distribution. This process corresponds to the light intensity distribution characterized by the following formula (5):

[0093] + + 2 R cos( - ) (5)

[0094] Among them, the object light intensity term Used to characterize the inherent strength of customer data, such as the square of the standardized value of preprocessed customer consumption data, reflecting the independent importance and intrinsic value of the customer data; (Referencing the intensity of light itself). Used to characterize the inherent stability of a data benchmark, such as the square of a standardized benchmark, to ensure that benchmark data such as industry averages or customer group averages remain stable during comparisons and to avoid interference from individual outliers; interference terms. 2 R cos( - )This is used to characterize the degree of deviation of customer data from benchmark data, and its core lies in carrying the object-light phase. (Data difference) and reference light phase (Benchmark calibration) difference information. This phase difference information is key to in-depth customer relationship insights. Traditional data storage often only retains intensity information and loses phase information, making it impossible to distinguish customers with the same numerical value but different attributes; the introduction of the interference term precisely makes up for this key difference dimension.

[0095] To extract the complete phase and amplitude information of the client data from the aforementioned interference intensity distribution, multiple different phase shifts can be introduced sequentially. Under each phase shift, a set of interference intensity distributions is obtained, resulting in multiple sets of holographic intensity data. Specifically, the phase shifts are typically taken as 0, π / 2, π, and 3π / 2, corresponding to four sets of holographic intensity data. , , , In data-driven mapping, these phase shifts are not physical phase shifts of light waves, but rather adjustments to the phase of the reference optical complex amplitude to obtain correlation characteristics between customer data and benchmark data from different calibration perspectives. For example, different phase shifts can be equivalent to using different industry benchmarks, business calibration parameters, time windows, or customer group averages as comparison scales (using the same customer data and the same reference benchmark under different phase shifts, only changing the known reference phase), thereby observing the relationship between the same customer object and the benchmark from multiple dimensions and avoiding the one-sidedness of comparing a single benchmark.

[0096] Based on the aforementioned sets of holographic intensity data, the complete phase and amplitude information of the second object's complex amplitude on the holographic recording plane can be recovered using a multi-step phase-shifting algorithm. The recovery process is as follows.

[0097] First, the phase direction of the customer data relative to the reference is calculated using the phase recovery formula, as shown in formula (6).

[0098] ( =arctan (6)

[0099] As an alternative implementation, the arctan function in formula (6) can also be replaced by the two-parameter arctangent function atan2.

[0100] Then, the complete complex amplitude expression of the second object light complex amplitude is obtained by reconstructing the complete complex amplitude reconstruction formula, as shown in the following formula (7).

[0101] = (7)

[0102] In customer data scenarios, this multi-step phase-shift algorithm is equivalent to recovering the customer's originally hidden differentiated attributes from the comparison results of multiple sets of customer data with a benchmark. These attributes include the customer's true preference direction, risk level, business need type, or deviation direction relative to the industry benchmark. The true characteristics of a customer often cannot be directly determined from a single data dimension and must be recovered by combining multiple sets of comparison results. Therefore, the multi-step phase-shift algorithm provides a mathematical framework for reconstructing complete phase information from intensity observations.

[0103] After obtaining the complete complex amplitude representation for each customer, customer relationship holographic data can be generated based on the complete complex amplitude representations for multiple customers. Specifically, each customer can be abstracted as a holographic primitive element, which includes at least object identifier, feature coordinates, amplitude attribute, phase attribute, and time or channel source information. Multiple primitive elements are arranged according to a unified coordinate mapping rule to form a data image for subsequent holographic calculations. This customer relationship holographic data is not an ordinary two-dimensional chart, but a data representation result containing multi-dimensional attributes such as location, intensity, phase, and source. The primitive position represents the distribution of customers or customer groups in the fused feature space, the primitive brightness or amplitude represents customer value or activity level, the phase difference represents customer preferences, risk status, or deviation from industry benchmarks, and local clustering areas represent customer groups with similar consumption behaviors, service needs, or risk characteristics.

[0104] From a theoretical perspective, optical interferometry provides a precise mathematical model for measuring the degree of deviation between customers and industry benchmarks. In optics, the contrast of interference fringes produced by the superposition of object and reference waves directly depends on the phase difference between the two waves. Applying this principle to data processing, the interference terms between customer and benchmark data map the degree of deviation to a continuous interval through a cosine relationship. This allows for the identification of high-value customers significantly better than the benchmark, as well as capturing customers at risk of churn who are significantly worse than the benchmark, while simultaneously preserving the directional information of the deviation. Compared to simple numerical subtraction or standardized scores in traditional statistics, the interferometry model encapsulates customers, benchmarks, and their relationship within a unified mathematical expression, making the difference itself a calculable, storable, and reconstructable core data asset.

[0105] Through the above-described superposition interference and phase recovery process, the embodiments of this application achieve the technical effect of fully preserving the strength of customer attributes, differentiated features, and deviation from the benchmark in a unified fusion space, providing high-quality data input containing complete phase information for subsequent image reconstruction and insight enhancement.

[0106] Then, the holographic data of customer relationships can be reconstructed to obtain a visualized image of the customer relationships, as shown below.

[0107] In some embodiments of this application, image reconstruction of customer relationship holographic data to obtain a customer relationship visualization image includes: propagating the complete complex amplitude expression of the customer relationship holographic data from the holographic recording plane to the object plane through a backpropagation transformation, wherein the backpropagation transformation is used to restore the fused feature data on the holographic recording plane to the distribution relationship of customers in the original feature space; on the object plane, determining the primitive brightness based on the amplitude component of the complete complex amplitude expression, determining the primitive phase feature based on the phase component of the complete complex amplitude expression, and determining the primitive position distribution based on the feature coordinates of the customers, wherein the primitive brightness is used to characterize the intensity of the customer's attribute value, the primitive phase feature is used to characterize the customer's differentiated attributes, and the primitive position distribution is used to characterize the spatial distribution relationship of customers in the fused feature space; and fusing and rendering the primitive position distribution, primitive brightness, and primitive phase feature to generate a customer relationship visualization image.

[0108] In this embodiment, reconstructing customer relationship holographic data to obtain a visualized image of customer relationships is a key step in restoring the fused customer data from the holographic recording plane into an observable and interactive visual representation. Specifically, the complete complex amplitude representation of the customer relationship holographic data can be propagated from the holographic recording plane to the object plane through a backpropagation transformation. This backpropagation transformation corresponds to the Fresnel reconstruction process in optical holography. First, in order to obtain a visualized image of the recorded object, the image is... Substituting into Fresnel's formula, the distance from the CCD plane can be calculated as follows: The image at the location is shown in the following formula (8):

[0109] = exp[ ( + )] (8)

[0110] In an optical sense, the above formula (8) is the Fresnel reconstruction formula, which is used to calculate the reconstructed image at the target distance based on the complex amplitude of the holographic recording plane. In a data scenario, it can be equivalent to reconstructing the customer primitives in the fusion representation space into a visualized data image, thereby obtaining observable results such as customer group distribution, preference clustering, and risk areas.

[0111] Furthermore, substituting the aforementioned propagation expression (i.e., formula (4)) into the reconstruction formula, we obtain the following expanded form:

[0112] = exp[ ( + )] (9)

[0113] Formula (9) indicates that by substituting the multi-source data fusion mapping result into the data image reconstruction process, the intensity, phase, and spatial relationships of the original customer data can be reflected in the final customer relationship image. Further simplification yields:

[0114] = exp[ ( + )]

[0115] = exp[ ( + )]

[0116]

[0117] = exp[ ( + )] (in, = , = )

[0118] =- exp[ ( + )]

[0119]

[0120] Right now =- exp[ (1+ )( + )] (10)

[0121] Formula (10) is the simplified result of the aforementioned reconstruction formula, used to represent the complex amplitude distribution of the reconstructed image under specific conditions; in the data scenario, it corresponds to the formation result of the final customer relationship data image, that is, the position, intensity, phase difference and clustering relationship of customer primitives in the fusion space are reconstructed into an image expression that can be used for insight enhancement. When the reproduction distance is equal to the original propagation distance, the system can reconstruct a clear customer relationship data image.

[0122] The object light wave was obtained in , After the complex amplitude distribution on the plane, it is propagated in reverse. To the object plane, when reproducing distance = At that time, you can get = = Therefore =- , =- , =- A clear image can be reconstructed from this point, resulting in the final digital holographic reconstruction image.

[0123] Before image reconstruction, to recover the complete phase information of the client data from the interference light intensity distribution, the system needs to sequentially introduce multiple different phase shifts to generate multiple sets of digital holograms. Assuming the phase shift introduced into the reference light is θ, when θ is successively 0, α, β, and γ, according to the complex amplitude of the reference light, propagation transformation, and interference formula, the resulting digital hologram distributions are as follows:

[0124] + + +

[0125] + + exp(-i )+ exp(i )

[0126] + + exp(-i )+ exp(i )

[0127] + + exp(-i )+ exp(i )

[0128] The four sets of formulas above represent multiple digital holograms obtained by sequentially introducing different phase shifts into the reference light. In data-driven understanding, different phase shifts are equivalent to observing customer data multiple times from different reference angles or calibration conditions, such as using different industry averages, regional benchmarks, or business calibration parameters as references. Through joint calculation of multiple sets of observation results, the system can more stably recover the phase information and hidden difference features of customer primitives, avoiding the one-sidedness caused by a single benchmark comparison, thereby improving the completeness and accuracy of subsequent image reconstruction.

[0129] On the object plane, the system can determine the multidimensional attributes of visualized primitives based on the components expressed by the complete complex amplitude. Specifically, the brightness of the primitive can be determined based on the amplitude component. This brightness characterizes the intensity of customer attribute values, such as customer value, activity level, or business weight. The larger the amplitude, the higher the brightness of the primitive, intuitively reflecting the visual differences between high-value and low-value customers. The phase characteristics of the primitive can be determined based on the phase component. These phase characteristics characterize the differentiated attributes of customers, such as customer preference type, risk status, or degree of deviation from industry benchmarks. Different phase values ​​correspond to different color codes or texture features, enabling the system to intuitively distinguish between customers with traffic preferences, broadband preferences, and value-added service preferences. The positional distribution of the primitive can be determined based on the customer's feature coordinates. This positional distribution characterizes the spatial distribution relationship of customers in the fused feature space. Customers with similar consumption behaviors, service needs, or risk characteristics form local clusters in space, thus naturally presenting the clustering structure of customer groups.

[0130] After obtaining the aforementioned multi-dimensional primitive attributes, the system fuses and renders the primitive location distribution, primitive brightness, and primitive phase characteristics to generate a customer relationship visualization image. This customer relationship visualization image is not an ordinary two-dimensional chart, but a data expression result containing multi-dimensional attributes such as location, intensity, phase, and source. In specific implementation, the system can sequentially set phase shifts to generate multiple sets of digital holograms, substitute the holographic object light into the Fresnel formula for three-dimensional reconstruction, and display a three-dimensional customer cluster map on the holographic terminal. Users can switch between different dimensions such as consumption, behavior, and preferences to observe the distribution of customer clusters through rotation and zoom operations; click on any customer cluster through hierarchical filtering operations to display the core characteristics of the group, such as "25-35 year old urban traffic preference customers"; the system can also dynamically warn customers at risk of churn by flashing red markers, and display risk factors when the mouse hovers over them, such as "traffic usage decreased by 20% for 3 consecutive months".

[0131] The core advantage of optical holographic reconstruction lies in its ability to reconstruct a three-dimensional image of an object from a two-dimensional interference pattern by recording complete light wave information (amplitude and phase). This application extends this principle to data processing. Customer relationship holographic data also preserves the complete complex amplitude information of customer data; therefore, reverse propagation transformation can reconstruct multi-dimensional customer relationships that traditional two-dimensional reports cannot express. After frequency domain fusion and phase recovery, the customer data is restored into a three-dimensional data field with differences in depth, brightness, and color: the clustering relationship of customer groups is represented by spatial proximity, customer value by the brightness intensity of primitives, preference differences by phase-encoded hue or phase difference, and risk deviations are indicated by dynamic flashing or deformation. This visualization is no longer a mechanical stacking of multiple two-dimensional charts, but a unified information field that can be observed, scaled, and sliced ​​from any angle, conforming to the intuitive understanding of spatial relationships in human visual perception. Therefore, the 3D visualization image formed by fusing and rendering primitive position, brightness, and phase features allows analysts to intuitively perceive salient aspects of customer data. Visual representation enhances cognitive reasoning through perceptual reasoning, making the analytical reasoning process faster and more focused. This customer relationship visualization image serves as input for subsequent insight enhancement, enabling the system to further perform customer segmentation identification, churn risk warning, and automated service recommendations. This effectively solves the technical problem of traditional visualization analysis methods being limited to two-dimensional static displays and unable to fully reflect the real and complex relationships between data. Therefore, optical concepts not only provide the mathematical language for data processing but also the natural form of result presentation, ensuring that the entire chain from raw data to business insights operates self-consistently under the same set of geometric and physical intuitions.

[0132] After obtaining the customer relationship visualization image, in order to further explore the deeper patterns behind the data, this application embodiment can also use an insight enhancement network to perform in-depth insight analysis on the customer relationship visualization image, as detailed below.

[0133] In some embodiments of this application, after obtaining the customer relationship visualization image, the method further includes: employing an insight enhancement network to perform deep insight analysis on the customer relationship visualization image and outputting deep insight results. The insight enhancement network is used to explore and analyze the customer relationship visualization image from different angles and dimensions to obtain deeper insights and understanding. The insight enhancement network is trained using a hierarchical frozen training strategy. The deep insight results include at least one of the following: customer segmentation results, preference mining results, and risk warning results. Based on the deep insight results, an automated service strategy is triggered. The automated service strategy includes at least one of the following: pushing churn intervention services to high-risk customers, pushing precise recommendation services to preferred customers, and pushing service upgrade plans to specific customer groups.

[0134] Specifically, such as Figure 4 As shown, the insight enhancement network in this embodiment can be extended based on the improved U-Net network structure, and includes three core modules: jump connections, output convolutional layers, and hierarchical segmentation blocks (HS-Blocks). The jump connections are used to connect the feature maps output by shallow and deep networks through channels, allowing the deep network to retain useful shallow information while avoiding gradient vanishing and improving network performance. The output convolutional layer consists of a single convolutional layer and an activation layer, used to output a single-channel insight enhancement image. The hierarchical segmentation blocks are used to implement multi-scale feature extraction. Figure 4 The numbers above the matrix box indicate the number of channels in the current feature map. "1" indicates that both the input and output are grayscale images, while "40", "20", and "20" represent the channel configurations of different levels in the improved U-Net structure, respectively.

[0135] This insight enhancement network is used to explore and analyze customer relationship visualizations from different angles and dimensions to gain deeper insights and understanding. To achieve this goal, in this embodiment, the insight enhancement network can be trained using a layered freeze-training strategy. The core idea of ​​this strategy is that feature maps extracted by training and enhancing digital holograms at a certain scale are equally valuable to feature maps at other scales. By freezing the trained parameters layer by layer, redundant computation can be avoided, reducing the total number of network parameters, enabling a single deep learning model to perform insight enhancement on customer relationship visualizations at multiple scales.

[0136] For example, a first-layer network can be trained to process 256×256 scale images with its parameters fixed. Then, a second-layer network trained to process 512×512 scale images can be overlaid, updating only the second-layer parameters and keeping them fixed. Finally, a third-layer network trained to process 640×480 scale images can be overlaid, updating only the third-layer parameters while keeping the first and second-layer parameters unchanged. Taking a telecommunications operator's individual customer scenario as an example, the system can select 200,000 customer holograms (covering the three scales mentioned above) and train them with a 70% training set and a 30% test set ratio. This layered, frozen training strategy allows the model to gradually acquire multi-scale feature extraction capabilities.

[0137] Through the trained insight enhancement network, the system performs in-depth insight analysis on the visualized customer relationship image, outputting in-depth insight results, which include at least one of the following: customer segmentation results, preference mining results, and risk warning results. Specifically, customer segmentation results can identify multiple core customer groups, such as high-value broadband users, young data users, elderly basic service users, and potential upgrade users; preference mining results can discover the behavioral preferences and demand characteristics of specific customer groups, such as "25-35 year old data-preference customers" frequently using short video apps and having a strong demand for targeted data packages, or "high-value broadband users" paying attention to network stability and willing to pay for gigabit upgrades; risk warning results can accurately identify high-risk customers and their core risk factors, such as "untimely handling of broadband failure complaints" or "package fees 20% higher than similar customers," etc.

[0138] Furthermore, based on the aforementioned in-depth insights, automated service strategies can be triggered. These strategies may include, but are not limited to: pushing churn intervention services to high-risk customers, pushing targeted recommendation services to customers with preferences, and pushing service upgrade plans to specific customer groups. For example, churn intervention services can push free "on-site broadband fault detection" services and package coupons (such as a monthly fee reduction of 10 yuan for 3 consecutive months) to high-risk customers; targeted recommendation services can recommend "short video targeted data packages" to customers with data usage preferences and push "gigabit broadband upgrade discounts" to high-value broadband users; service upgrade plans can simplify the APP operation interface and add voice navigation functions for elderly users, or open a "one-click business processing" green channel for active online customers.

[0139] This application's embodiments utilize an insight-enhancing network to perform deep learning on customer relationship visualization images. This not only automatically explores hidden patterns in the data from multiple angles and dimensions, but also adapts to various data scales while maintaining model simplicity through a hierarchical frozen training strategy. This effectively extracts deep insights such as customer segmentation, preference mining, and risk warning, directly triggering corresponding automated service strategies. This design tightly integrates data visualization with business execution, forming a complete closed loop from data fusion and insight analysis to service triggering. It achieves the technical effect of improving customer satisfaction and business conversion rates, effectively solving the technical problems of traditional visualization analysis methods being unable to automatically uncover deep patterns and support accurate business decision-making.

[0140] This application, based on the isomorphism of mathematical structures, transfers the principles of optical holography to the field of multi-source customer relationship data processing. Specifically, optical holography preserves the complete three-dimensional information of an object by recording the amplitude and phase of light waves. Similarly, customer relationship data needs to simultaneously preserve both "numerical intensity" and "differentiated attributes" to fully characterize customer features. Traditional visualization analysis can only record data size in the real number domain, losing crucial "phase" information such as customer preferences and risk status. Furthermore, limited by two-dimensional static displays and point-to-point functional relationships, it cannot fully reflect the true relationships, complex logic, and dynamic interactions between data. This application, leveraging the mathematical framework of optical holography, encodes multi-source heterogeneous customer data into a complex amplitude expression containing amplitude and phase. Through propagation transformation, interferometric calculation, and phase recovery, it fully preserves the attribute intensity, differential characteristics, and deviations from the benchmark of customer data in a unified multi-dimensional feature space. This elevates data fusion from simple table splicing to a structured mapping with geometric and algebraic meaning, making the differences themselves a computable, storable, and reconstructable core data asset.

[0141] Based on this, this application constructs a complete technical solution covering data processing, result presentation, and insight enhancement, including: establishing a digital holographic coordinate system to map the original multi-source feature space to the fused representation space; introducing a reference light as an industry benchmark and quantifying the degree of customer deviation relative to the benchmark through an interference mechanism; reconstructing images based on the Fresnel diffraction formula to generate a 3D customer relationship visualization image that supports rotation, scaling, hierarchical filtering, and dynamic early warning, achieving a highly interactive business model and natural and realistic visual restoration; further adopting an improved U-Net network structure and using a hierarchical freeze training strategy to enhance insights into digital holograms at multiple scales, deeply exploring data patterns from different angles and dimensions, outputting customer segmentation, preference mining, and risk warning results, and triggering automated service strategies, forming a complete closed loop from data fusion and visualization insights to service triggering.

[0142] By transforming data into a visible form through holographic digital visualization technology, commonalities and anomalies within customer groups can be highlighted, allowing users to easily and quickly perceive significant aspects of the data. Visual representation enhances cognitive reasoning through perceptual reasoning, making analytical reasoning faster and more focused. The three-dimensional dynamic interactive visualization breaks through the limitations of traditional two-dimensional charts, achieving a more natural and realistic visual reproduction. Through comprehensive information collection and perception of customer behavior in the real world, combined with holographic digital visualization analysis, real-time calculation and simulation of business operations can be performed, and future trends can be effectively predicted and intervened through digital calculation. The improved U-Net network further enhances deep insight capabilities, achieving efficient analysis of multi-scale data while reducing model parameters, ultimately effectively improving customer satisfaction and business conversion rates.

[0143] According to an embodiment of this application, an embodiment of a data processing apparatus is also provided. Figure 5 This is a schematic diagram of the structure of a data processing device according to an embodiment of this application. Figure 5 As shown, the device includes:

[0144] The data representation module 50 is used to acquire multi-source customer data and extract features from the multi-source customer data to obtain customer feature vectors. The multi-source customer data includes heterogeneous data from multiple sources or of different types corresponding to multiple customers, and each customer corresponds to a customer feature vector.

[0145] The mapping processing module 52 is used to map customer feature vectors to a digital holographic coordinate system to obtain customer relationship holographic data. In the customer relationship holographic data, each customer's customer feature vector is mapped to a holographic primitive containing amplitude and phase attributes. The amplitude attribute is used to characterize the intensity of the customer attribute value, and the phase attribute is used to characterize the customer difference attribute.

[0146] The visualization module 54 is used to perform image reconstruction on the holographic data of customer relationships to obtain a customer relationship visualization image. The image reconstruction is used to restore the complete information of the customer data based on the amplitude and phase attributes. The customer relationship visualization image is used to characterize the distribution relationship of customers in the fused feature space and the degree of deviation between the customer data and the benchmark data.

[0147] Optionally, feature extraction from multi-source customer data to obtain a customer feature vector includes: obtaining heterogeneous data of the same customer from multiple data sources to obtain multi-source customer data, wherein the data dimensions of the multi-source customer data include at least one of the following: consumption data, behavioral data, feedback data, and basic data; extracting features from the multi-source customer data of the same customer according to the data dimensions to obtain feature sub-vectors corresponding to each data dimension; and concatenating the feature sub-vectors corresponding to different data dimensions of the same customer to obtain a customer feature vector.

[0148] Optionally, mapping customer feature vectors to a digital holographic coordinate system to obtain customer relationship holographic data includes: establishing a digital holographic coordinate system comprising an object plane and a holographic recording plane, wherein the object plane corresponds to the original feature space, and the holographic recording plane corresponds to the fused feature space after data fusion processing; mapping each customer's customer feature vector to a first object optical complex amplitude on the object plane, wherein the first object optical complex amplitude is used to characterize the joint expression of data intensity and differential attributes of customer data in the original feature space; propagating and transforming the first object optical complex amplitude to the holographic recording plane through a propagation operator to obtain a second object optical complex amplitude on the holographic recording plane, wherein the second object optical complex amplitude is used to characterize the fused feature expression of customer data after propagation transformation; and superimposing and interfering the second object optical complex amplitude with a reference optical complex amplitude on the holographic recording plane to obtain customer relationship holographic data, wherein the reference optical complex amplitude is used to characterize the reference data corresponding to the customer data, and the superimposed interference is used to generate an interference term carrying phase difference information between the customer data and the reference data to measure the degree of deviation of the customer object relative to the reference.

[0149] Optionally, mapping each customer's feature vector to the first object light complex amplitude on the object plane includes: selecting key feature dimensions from the customer feature vector as the customer's coordinate information on the object plane; calculating the amplitude attribute corresponding to the customer based on the weight coefficients corresponding to each data dimension in the multi-source customer data and the customer's numerical performance in each data dimension; determining the phase attribute corresponding to the customer based on the customer's differentiated labels, wherein the differentiated labels are used to characterize at least one of the customer's preference type, risk level, or business status, and different customer preference types, risk levels, or business statuses correspond to different phase values; and constructing a complex amplitude expression based on the coordinate information, amplitude attribute, and phase attribute to obtain the first object light complex amplitude.

[0150] Optionally, propagating and transforming the first object optical complex amplitude to the holographic recording plane using a propagation operator to obtain the second object optical complex amplitude on the holographic recording plane includes: using a propagation operator to perform information loss compensation processing on the first object optical complex amplitude, wherein the information loss compensation processing is used to compensate for information loss in the process of multi-source customer data from the original channel to the processing layer; performing dimensional adjustment processing on the first object optical complex amplitude after information loss compensation processing to map the customer data in the original feature space to the corresponding coordinates in the fused feature space, wherein the dimensional adjustment processing includes at least one of the following: dimensionality reduction, feature scaling, and feature weight allocation; performing frequency domain fusion processing on the first object optical complex amplitude after dimensional adjustment processing, wherein the frequency domain fusion processing is used to achieve frequency feature superposition and fusion of data from different sources by converting the spatial domain data signal to the frequency domain; and mapping the first object optical complex amplitude after frequency domain fusion processing to the holographic recording plane to obtain the second object optical complex amplitude.

[0151] Optionally, on the holographic recording plane, the second object light complex amplitude and the reference light complex amplitude are superimposed and interfered to obtain customer relationship holographic data. This includes: superimposing and interfering the second object light complex amplitude and the reference light complex amplitude to obtain an interference light intensity distribution, wherein the interference light intensity distribution includes: an object light intrinsic intensity term, a reference light intrinsic intensity term, and an interference term. The object light intrinsic intensity term is used to characterize the intrinsic intensity of the customer data, the reference light intrinsic intensity term is used to characterize the intrinsic stability of the data reference, and the interference term is used to characterize the deviation of the customer data from the reference data; these are introduced sequentially. Multiple different phase shifts are used to obtain a set of interference intensity distributions at each phase shift, resulting in multiple sets of holographic intensity data. The phase shifts are used to adjust the phase of the reference optical complex amplitude to obtain the correlation features between customer data and reference data from different calibration perspectives. Based on the multiple sets of holographic intensity data, a multi-step phase shift algorithm is used to recover the complete phase information and complete amplitude information of the second object optical complex amplitude on the holographic recording plane, reconstructing the complete complex amplitude expression of the second object optical complex amplitude. Based on the complete complex amplitude expressions corresponding to multiple customers, customer relationship holographic data is generated.

[0152] Optionally, image reconstruction of customer relationship holographic data to obtain a customer relationship visualization image includes: propagating the complete complex amplitude expression of the customer relationship holographic data from the holographic recording plane to the object plane through a backpropagation transformation, wherein the backpropagation transformation is used to restore the fused feature data on the holographic recording plane to the distribution relationship of customers in the original feature space; on the object plane, determining the primitive brightness based on the amplitude component of the complete complex amplitude expression, determining the primitive phase feature based on the phase component of the complete complex amplitude expression, and determining the primitive position distribution based on the customer's feature coordinates, wherein the primitive brightness is used to characterize the intensity of the customer's attribute value, the primitive phase feature is used to characterize the customer's differentiated attributes, and the primitive position distribution is used to characterize the spatial distribution relationship of customers in the fused feature space; fusing and rendering the primitive position distribution, primitive brightness, and the primitive phase feature to generate a customer relationship visualization image.

[0153] Optionally, after obtaining the customer relationship visualization image, the method further includes: employing an insight enhancement network to perform deep insight analysis on the customer relationship visualization image and outputting deep insight results. The insight enhancement network is used to explore and analyze the customer relationship visualization image from different angles and dimensions to obtain deeper insights and understanding. The insight enhancement network is trained using a hierarchical frozen training strategy. The deep insight results include at least one of the following: customer segmentation results, preference mining results, and risk warning results. Based on the deep insight results, an automated service strategy is triggered. The automated service strategy includes at least one of the following: pushing churn intervention services to high-risk customers, pushing precise recommendation services to preferred customers, and pushing service upgrade plans to specific customer groups.

[0154] It should be noted that each module in the above data processing device can be a program module (e.g., a set of program instructions to implement a certain function) or a hardware module. For the latter, it can be manifested in the following forms, but is not limited to them: each of the above modules is manifested as a processor, or the functions of each of the above modules are implemented by a processor.

[0155] It should be noted that the data processing device provided in this embodiment can be used to perform... Figure 2 The data processing method shown above is also applicable to the embodiments of this application, and will not be repeated here.

[0156] This application embodiment also provides a non-volatile storage medium, which includes a stored computer program. The device containing the non-volatile storage medium executes the following data processing method by running the computer program: acquiring multi-source customer data and extracting features from the multi-source customer data to obtain customer feature vectors. The multi-source customer data includes heterogeneous data from multiple sources or of different types corresponding to multiple customers, with each customer corresponding to a customer feature vector. The customer feature vectors are mapped to a digital holographic coordinate system to obtain customer relationship holographic data. In the customer relationship holographic data, each customer's customer feature vector is mapped to a holographic primitive containing amplitude and phase attributes. The amplitude attribute characterizes the intensity of the customer attribute value, and the phase attribute characterizes the customer's difference attribute. Image reconstruction is performed on the customer relationship holographic data to obtain a customer relationship visualization image. The image reconstruction is used to restore the complete information of the customer data based on the amplitude and phase attributes, and the customer relationship visualization image is used to characterize the distribution relationship of customers in the fused feature space and the degree of deviation between the customer data and the benchmark data.

[0157] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the data processing method described in various embodiments of this application: acquiring multi-source customer data and extracting features from the multi-source customer data to obtain customer feature vectors, wherein the multi-source customer data contains heterogeneous data from multiple sources or of different types corresponding to multiple customers, and each customer corresponds to a customer feature vector; mapping the customer feature vectors to a digital holographic coordinate system to obtain customer relationship holographic data, wherein in the customer relationship holographic data, the customer feature vector of each customer is mapped to a holographic primitive containing amplitude and phase attributes, the amplitude attribute being used to characterize the intensity of customer attribute values, and the phase attribute being used to characterize customer difference attributes; performing image reconstruction on the customer relationship holographic data to obtain a customer relationship visualization image, wherein the image reconstruction is used to restore the complete information of the customer data based on the amplitude and phase attributes, and the customer relationship visualization image is used to characterize the distribution relationship of customers in the fused feature space and the degree of deviation between the customer data and the benchmark data.

[0158] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0159] In the above embodiments of this application, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0160] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units can be a logical functional division, and in actual implementation, there may be other division methods. For instance, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling, direct coupling, or communication connection may be through some interfaces; the indirect coupling or communication connection between units or modules may be electrical or other forms.

[0161] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0162] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0163] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, read-only memory (ROM), random access memory (RAM), portable hard drive, magnetic disk, or optical disk.

[0164] The above description is only a preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A data processing method, characterized in that, include: Acquire multi-source customer data and extract features from the multi-source customer data to obtain customer feature vectors. The multi-source customer data includes heterogeneous data from multiple sources or of different types corresponding to multiple customers, and each customer corresponds to one customer feature vector. The customer feature vector is mapped to a digital holographic coordinate system to obtain customer relationship holographic data. In the customer relationship holographic data, the customer feature vector of each customer is mapped to a holographic primitive containing amplitude and phase attributes. The amplitude attribute is used to characterize the intensity of the customer attribute value, and the phase attribute is used to characterize the customer difference attribute. The customer relationship holographic data is reconstructed to obtain a customer relationship visualization image. The image reconstruction is used to restore the complete information of the customer data based on the amplitude attribute and the phase attribute. The customer relationship visualization image is used to characterize the distribution relationship of customers in the fused feature space and the degree of deviation between the customer data and the benchmark data.

2. The data processing method according to claim 1, characterized in that, Feature extraction is performed on the multi-source customer data to obtain customer feature vectors, including: Heterogeneous data of the same customer is obtained from multiple data sources to obtain the multi-source customer data, wherein the data dimensions of the multi-source customer data include at least one of the following: consumption data, behavioral data, feedback data, and basic data; For the multi-source customer data of the same customer, feature extraction is performed according to the data dimensions to obtain the feature sub-vectors corresponding to the data of each data dimension; The feature vectors corresponding to different data dimensions of the same customer are concatenated to obtain the customer feature vector.

3. The data processing method according to claim 1, characterized in that, Mapping the customer feature vectors to a digital holographic coordinate system yields customer relationship holographic data, including: A digital holographic coordinate system comprising an object plane and a holographic recording plane is established, wherein the object plane corresponds to the original feature space, and the holographic recording plane corresponds to the fused feature space after data fusion processing; The customer feature vector of each customer is mapped to the first object light complex amplitude on the object plane, wherein the first object light complex amplitude is used to characterize the joint expression of data intensity and differential attributes of customer data in the original feature space; The first object optical complex amplitude is propagated and transformed to the holographic recording plane by a propagation operator to obtain the second object optical complex amplitude on the holographic recording plane. The second object optical complex amplitude is used to characterize the fusion feature expression of the customer data after propagation transformation. On the holographic recording plane, the second object light complex amplitude and the reference light complex amplitude are superimposed and interfered to obtain the customer relationship holographic data. The reference light complex amplitude is used to characterize the reference data corresponding to the customer data, and the superimposed interference is used to generate an interference term carrying the phase difference information between the customer data and the reference data to measure the degree of deviation of the customer object from the reference.

4. The data processing method according to claim 3, characterized in that, Mapping each customer's customer feature vector to the first object light complex amplitude on the object plane includes: Key feature dimensions are selected from the customer feature vector to serve as the customer's coordinate information in the object plane; Based on the weight coefficients corresponding to each data dimension in the multi-source customer data, and the numerical performance of the customer in each data dimension, the amplitude attribute corresponding to the customer is calculated; Based on the customer's differentiated tags, the phase attribute corresponding to the customer is determined, wherein the differentiated tags are used to characterize at least one of the customer's preference type, risk level, or business status, and different customer preference types, risk levels, or business statuses correspond to different phase values; Based on the coordinate information, the amplitude attribute, and the phase attribute, a complex amplitude expression is constructed to obtain the complex amplitude of the first object light.

5. The data processing method according to claim 3, characterized in that, The first object optical complex amplitude is propagated and transformed onto the holographic recording plane using a propagation operator to obtain the second object optical complex amplitude on the holographic recording plane, including: The propagation operator is used to perform information loss compensation processing on the complex amplitude of the first object light, wherein the information loss compensation processing is used to compensate for the information loss of multi-source customer data in the process from the original channel to the processing layer; The first object light complex amplitude after the information loss compensation processing is subjected to dimension adjustment processing to map the customer data in the original feature space to the corresponding coordinates in the fused feature space. The dimension adjustment processing includes at least one of the following: dimension reduction, feature scaling, and feature weight allocation. The first object light complex amplitude after the dimensional adjustment process is subjected to frequency domain fusion processing, wherein the frequency domain fusion processing is used to convert the spatial domain data signal to the frequency domain in order to achieve the superposition and fusion of frequency features of data from different sources. The first object optical complex amplitude, after undergoing the frequency domain fusion processing, is mapped onto the holographic recording plane to obtain the second object optical complex amplitude.

6. The data processing method according to claim 3, characterized in that, On the holographic recording plane, the second object light complex amplitude and the reference light complex amplitude are superimposed and interfered to obtain the customer relationship holographic data, including: The second object light complex amplitude and the reference light complex amplitude are superimposed and interfered to obtain an interference light intensity distribution, wherein the interference light intensity distribution includes: an object light intrinsic intensity term, a reference light intrinsic intensity term, and an interference term. The object light intrinsic intensity term is used to characterize the intrinsic intensity of the customer data, the reference light intrinsic intensity term is used to characterize the intrinsic stability of the data reference, and the interference term is used to characterize the deviation of the customer data from the reference data. Multiple different phase shifts are introduced sequentially, and a set of interference intensity distributions is obtained under each phase shift to obtain multiple sets of holographic light intensity data. The phase shifts are used to adjust the phase of the reference light complex amplitude to obtain the correlation characteristics between customer data and reference data from different calibration perspectives. Based on the multiple sets of holographic light intensity data, the complete phase information and complete amplitude information of the second object light complex amplitude on the holographic recording plane are recovered by a multi-step phase-shifting algorithm, and the complete complex amplitude expression of the second object light complex amplitude is reconstructed. Based on the complete complex amplitude expression corresponding to multiple customers, the customer relationship holographic data is generated.

7. The data processing method according to claim 6, characterized in that, Image reconstruction is performed on the holographic data of customer relationships to obtain a visual image of customer relationships, including: The complete complex amplitude representation of the customer relationship holographic data is propagated from the holographic recording plane to the object plane through a backpropagation transformation, wherein the backpropagation transformation is used to restore the fused feature data on the holographic recording plane to the distribution relationship of customers in the original feature space; On the object plane, the brightness of the primitive is determined based on the amplitude component expressed by the complete complex amplitude, the phase feature of the primitive is determined based on the phase component expressed by the complete complex amplitude, and the position distribution of the primitive is determined based on the feature coordinates of the customer. The brightness of the primitive is used to characterize the intensity of the customer's attribute value, the phase feature of the primitive is used to characterize the differentiated attributes of the customer, and the position distribution of the primitive is used to characterize the spatial distribution relationship of the customer in the fused feature space. The location distribution of the graphic elements, the brightness of the graphic elements, and the phase features of the graphic elements are fused and rendered to generate the customer relationship visualization image.

8. The data processing method according to claim 1, characterized in that, After obtaining the customer relationship visualization image, the method further includes: An insight enhancement network is used to perform deep insight analysis on the customer relationship visualization image and output deep insight results. The insight enhancement network is used to explore and analyze the customer relationship visualization image from different angles and dimensions to obtain deeper insights and understanding. The insight enhancement network is trained through a hierarchical frozen training strategy. The deep insight results include at least one of the following: customer segmentation results, preference mining results, and risk warning results. Based on the deep insights obtained, an automated service strategy is triggered, wherein the automated service strategy includes at least one of the following: pushing churn intervention services to high-risk customers, pushing precise recommendation services to preferred customers, and pushing service upgrade plans to specific customer groups.

9. A data processing apparatus, characterized in that, include: The data representation module is used to acquire multi-source customer data and extract features from the multi-source customer data to obtain customer feature vectors. The multi-source customer data includes heterogeneous data from multiple sources or of different types corresponding to multiple customers, and each customer corresponds to one customer feature vector. The mapping processing module is used to map the customer feature vector to a digital holographic coordinate system to obtain customer relationship holographic data. In the customer relationship holographic data, the customer feature vector of each customer is mapped to a holographic primitive containing amplitude attribute and phase attribute. The amplitude attribute is used to characterize the intensity of the customer attribute value, and the phase attribute is used to characterize the customer difference attribute. The visualization module is used to perform image reconstruction on the customer relationship holographic data to obtain a customer relationship visualization image. The image reconstruction is used to restore the complete information of the customer data based on the amplitude attribute and the phase attribute. The customer relationship visualization image is used to characterize the distribution relationship of customers in the fused feature space and the degree of deviation between the customer data and the benchmark data.

10. An electronic device, characterized in that, include: A memory and a processor, the processor being configured to run a program stored in the memory, wherein the program, when executed, performs the data processing method according to any one of claims 1 to 8.

11. A non-volatile storage medium, characterized in that, The non-volatile storage medium includes a stored computer program, wherein the device containing the non-volatile storage medium executes the data processing method according to any one of claims 1 to 8 by running the computer program.

12. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the data processing method according to any one of claims 1 to 8.

Images