Customer purchase power prediction method and related device

Abstract

The invention discloses a customer purchase power prediction method and a related device, and relates to the field of artificial intelligence, and the method comprises the steps: obtaining the purchase power data of a risk level when a customer purchases a product, the purchase power data obtained based on a scoring questionnaire and a final transaction operation, and the serialized purchase weight data; and respectively preprocessing the data into data of a client side and a product side. And performing nonlinear feature extraction on the client side data and the product side data based on the deep learning model of the attention mechanism to obtain a client side implicit factor model and a product side implicit factor model. Structural feature extraction is carried out on the two implicit factor models based on a probability matrix decomposition model, and a customer side implicit factor model and a product side implicit factor model are obtained; and combining the implicit factor models to optimize the deep learning model and the probability matrix decomposition model. And predicting the purchasing power of the target user by using the probability matrix decomposition model. A more accurate purchasing power reference is provided, and the user experience is improved.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a method and related device for predicting customer purchasing power. Background Art

[0002] With the increasing convenience of online financial services and the improvement of customer financial management awareness, more and more customers choose to purchase financial products online. Due to the inseparability of customer purchases of financial products and marketing services, marketers need to market the most suitable products for customers from a large number of financial categories and financial products and achieve ideal returns. How to provide marketers with accurate and reliable purchasing power references to improve marketing accuracy and customer experience has become an urgent problem to be solved. Summary of the invention

[0003] In view of the above problems, this application provides a customer purchasing power prediction method and related devices to achieve the purpose of improving the purchasing power prediction accuracy and enhancing the user experience. The specific scheme is as follows:

[0004] The first aspect of the present application provides a method for predicting customer purchasing power, comprising:

[0005] Obtaining first purchasing power data representing the risk level of customers when purchasing products during the statistical period, second purchasing power data obtained based on the scoring questionnaire and the final transaction operation, and serialized purchasing weight data;

[0006] Splitting the first purchasing power data into a customer side and a product side to obtain a customer side purchasing power matrix and a product side purchasing power matrix;

[0007] Preprocessing the second purchasing power data into a customer implicit factor query vector corresponding to the customer-side purchasing power matrix and a product implicit factor query vector corresponding to the product-side purchasing power matrix;

[0008] Splitting the serialized purchase weight data to obtain a customer-side purchase weight matrix and a product-side purchase weight matrix;

[0009] Inputting the customer-side data into a deep learning model based on an attention mechanism for nonlinear feature extraction to obtain a first customer-side implicit factor model, wherein the customer-side data includes: the customer-side purchasing power matrix, the customer implicit factor query vector, and the customer-side purchase weight matrix;

[0010] Inputting product-side data into the deep learning model for nonlinear feature extraction to obtain a first product-side implicit factor model, wherein the product-side data includes: the product-side purchasing power matrix, the product implicit factor query vector, and the product-side purchase weight matrix;

[0011] Based on the probability matrix decomposition model, structural features are extracted from the first customer-side implicit factor model and the first product-side implicit factor model respectively to obtain a second customer-side implicit factor model and a second product-side implicit factor model;

[0012] Combine the first client-side implicit factor model, the first product-side implicit factor model, the second client-side implicit factor model, and the second product-side implicit factor model to optimize the deep learning model and the probability matrix decomposition model;

[0013] The purchasing power of target users is predicted based on the optimized probability matrix decomposition model.

[0014] In a possible implementation, the first client-side implicit factor model, the first product-side implicit factor model, the second client-side implicit factor model, and the second product-side implicit factor model are combined to optimize the deep learning model and the probability matrix decomposition model, including:

[0015] Using the second client-side implicit factor model as a priori induced function of the first client-side implicit factor model, and performing training optimization on an objective function composed of the second client-side implicit factor model, the first client-side implicit factor model, and an output client-side result of the deep learning model;

[0016] The second product-side implicit factor model is used as a priori induced function of the first product-side implicit factor model, and the objective function composed of the second product-side implicit factor model, the first product-side implicit factor model and the output product-side result of the deep learning model is trained and optimized.

[0017] In a possible implementation, the first client-side implicit factor model, the first product-side implicit factor model, the second client-side implicit factor model, and the second product-side implicit factor model are combined to optimize the deep learning model and the probability matrix decomposition model, including:

[0018] The first customer-side implicit factor model and the first product-side implicit factor model are used as prior induction functions to train and optimize the objective function composed of the second customer-side implicit factor model, the first customer-side implicit factor model, the first product-side implicit factor model and the second product-side implicit factor model; the initial product-side implicit factor model and the initial customer-side implicit factor model of the probability matrix decomposition model are obtained according to the intermediate layer output of the deep learning model.

[0019] In a possible implementation, the deep learning model includes: a convolutional encoder, an implicit factor extractor, and a convolutional decoder, and the process of performing nonlinear feature extraction in the deep learning model includes:

[0020] Based on the convolution encoder, the client-side data and the product-side data are processed respectively to obtain client-side serialization feature data and product-side serialization feature data;

[0021] Based on the implicit factor extractor, feature extraction is performed on the client-side serialized feature data and the product-side feature data respectively to obtain the first client-side implicit factor model and the first product-side implicit factor model;

[0022] Based on the convolution decoder, the client-side serialized feature data and the product-side serialized feature data restored by the implicit factor extractor are restored to obtain the client-side purchasing power matrix and the product-side purchasing power matrix.

[0023] In a possible implementation, the convolutional encoder includes: a first convolutional encoding layer and a second convolutional encoding layer, the first convolutional encoding layer includes: at least one convolutional encoding unit, the convolutional encoding unit includes a first convolutional layer and a pooling layer connected in sequence; the second convolutional encoding layer includes: a first multi-head attention mechanism layer, a first residual connection and normalization layer, and a second convolutional layer; based on the convolutional encoder, the client-side data and the product-side data are processed respectively to obtain client-side serialized feature data and product-side serialized feature data, including:

[0024] Based on the first convolutional coding layer, feature extraction is performed on the customer-side purchasing power matrix and the customer-side purchase weight matrix to obtain customer-side purchasing power feature data;

[0025] Based on the first convolutional coding layer, feature extraction is performed on the product-side purchasing power matrix and the product-side purchasing weight matrix to obtain product-side purchasing power feature data;

[0026] Performing feature extraction on the customer-side purchasing power feature data and the customer implicit factor query vector based on the second convolutional coding layer to obtain the customer-side serialized feature data;

[0027] Based on the second convolutional coding layer, feature extraction is performed on the product-side purchasing power feature data and the product implicit factor query vector to obtain the product-side serialized feature data.

[0028] In a possible implementation, the implicit factor extractor includes: a fully connected layer and a third convolutional layer, and based on the implicit factor extractor, feature extraction is performed on the client-side serialized feature data and the product-side feature data respectively to obtain the first client-side implicit factor model and the first product-side implicit factor model, including:

[0029] Based on the fully connected layer, feature extraction is performed on the client-side serialized feature data and the product-side feature data respectively to obtain the first client-side implicit factor model and the first product-side implicit factor model;

[0030] The first client-side implicit factor model and the first product-side implicit factor model are respectively restored based on the third convolutional layer to obtain client-side feature mapping data and product-side feature mapping data.

[0031] In a possible implementation, the convolution decoder includes: a first convolution decoding layer and a second convolution decoding layer, the first convolution decoding layer includes: a second multi-head attention mechanism layer, a second residual connection and normalization layer, and a fourth convolution layer, the second convolution decoding layer includes convolution decoding units with the same number as the convolution encoding units, and each of the convolution decoding units includes: a fifth convolution layer and an upsampling layer connected in sequence; based on the convolution decoder, the client-side serialized feature data and the product-side serialized feature data restored by the implicit factor extractor are restored to obtain the client-side purchasing power matrix and the product-side purchasing power matrix, including:

[0032] Based on the first convolution decoding layer, the client-side feature mapping data and the client-side serialized feature data are restored to obtain the client-side purchasing power feature data;

[0033] Based on the first convolution decoding layer, the product-side feature mapping data and the product-side serialized feature data are restored to obtain the product-side purchasing power feature data;

[0034] Based on the first convolution decoding layer, the customer-side purchasing power feature data and the product-side purchasing power feature data are restored respectively to obtain the customer-side purchasing power matrix and the product-side purchasing power matrix.

[0035] A second aspect of the present application provides a customer purchasing power prediction device, comprising:

[0036] A purchasing power data acquisition module is used to acquire first purchasing power data representing the risk level of customers when purchasing products during a statistical period, second purchasing power data obtained based on a scoring questionnaire and a final transaction operation, and serialized purchasing weight data;

[0037] A first data preprocessing module, used for splitting the first purchasing power data into a customer side and a product side to obtain a customer side purchasing power matrix and a product side purchasing power matrix;

[0038] A second data preprocessing module, used for preprocessing the second purchasing power data into a customer implicit factor query vector corresponding to the customer-side purchasing power matrix and a product implicit factor query vector corresponding to the product-side purchasing power matrix;

[0039] A third data preprocessing module is used to split the serialized purchase weight data to obtain a customer-side purchase weight matrix and a product-side purchase weight matrix;

[0040] A client-side feature data extraction module, used for inputting client-side data into a deep learning model based on an attention mechanism to perform nonlinear feature extraction to obtain a first client-side implicit factor model, wherein the client-side data includes: the client-side purchasing power matrix, the client implicit factor query vector, and the client-side purchase weight matrix;

[0041] A product-side feature data extraction module, used for inputting product-side data into the deep learning model for nonlinear feature extraction to obtain a first product-side implicit factor model, wherein the product-side data includes: the product-side purchasing power matrix, the product implicit factor query vector and the product-side purchase weight matrix;

[0042] An implicit factor model extraction module, used to extract structural features of the first client-side implicit factor model and the first product-side implicit factor model respectively based on a probability matrix decomposition model to obtain a second client-side implicit factor model and a second product-side implicit factor model;

[0043] a joint optimization training module, used to jointly optimize the deep learning model and the probability matrix decomposition model by combining the first client-side implicit factor model, the first product-side implicit factor model, the second client-side implicit factor model and the second product-side implicit factor model; and

[0044] The purchasing power prediction module is used to predict the purchasing power of target users based on the optimized probability matrix decomposition model.

[0045] A third aspect of the present application provides a computer program product, including computer-readable instructions. When the computer-readable instructions are executed on an electronic device, the electronic device implements the customer purchasing power prediction method of the first aspect or any implementation of the first aspect.

[0046] A fourth aspect of the present application provides an electronic device, comprising at least one processor and a memory connected to the processor, wherein:

[0047] The memory is used to store computer programs;

[0048] The processor is used to execute the computer program so that the electronic device can implement the customer purchasing power prediction method of the first aspect or any implementation method of the first aspect.

[0049] A fifth aspect of the present application provides a computer storage medium, which carries one or more computer programs. When the one or more computer programs are executed by an electronic device, the electronic device can implement the customer purchasing power prediction method of the first aspect or any implementation of the first aspect.

[0050] By means of the above technical scheme, the customer purchasing power prediction method provided by the present application is to obtain the first purchasing power data representing the risk level of the customer when purchasing the product during the statistical period, the second purchasing power data obtained based on the scoring questionnaire and the final transaction operation, and the serialized purchase weight data, and pre-process them into risk interaction data on the customer side and the product side. On this basis, the deep learning model based on the attention mechanism performs nonlinear feature extraction on the customer side data and the product side data respectively to obtain the first customer side implicit factor model and the first product side implicit factor model. Then, based on the probability matrix decomposition model, the structural features of the first customer side implicit factor model and the first product side implicit factor model are extracted respectively to obtain the second customer side implicit factor model and the second product side implicit factor model. Combined with the above implicit factor models, the deep learning model and the probability matrix decomposition model are optimized. Finally, the purchasing power of the target user is predicted based on the optimized probability matrix decomposition model. It is achieved that while improving the model optimization efficiency, the prediction accuracy of the customer product purchasing power is improved, which can provide marketers with a more accurate purchasing power reference, thereby improving the user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] The above and other features, advantages and aspects of the embodiments of the present disclosure will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. Throughout the accompanying drawings, the same or similar reference numerals represent the same or similar elements. It should be understood that the drawings are schematic and the originals and elements are not necessarily drawn to scale.

[0052] Figure 1 An architectural diagram of a customer purchasing power prediction system provided for this application;

[0053] Figure 2 A structural diagram of a terminal provided for this application;

[0054] Figure 3 A structural diagram of a server provided for this application;

[0055] Figure 4A flowchart of a method for predicting customer purchasing power provided for this application;

[0056] Figure 5 Process diagram of feature extraction provided for this application;

[0057] Figure 6 A structural diagram of a customer purchasing power prediction device provided in this application;

[0058] Figure 7 A structural diagram of an electronic device provided in this application. DETAILED DESCRIPTION

[0059] The following describes the embodiments of the present application in conjunction with the drawings in the embodiments of the present application. The terms used in the implementation method section of the present application are only used to explain the specific embodiments of the present application, and are not intended to limit the present application.

[0060] The embodiments of the present application are described below in conjunction with the accompanying drawings. Those skilled in the art will appreciate that, with the development of technology and the emergence of new scenarios, the technical solutions provided in the embodiments of the present application are also applicable to similar technical problems.

[0061] The terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and need not be used to describe a specific order or sequential order. It should be understood that the terms used in this way can be interchangeable under appropriate circumstances, which is only to describe the distinction mode adopted by the objects of the same attributes when describing in the embodiments of the present application. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, so that the process, method, system, product or equipment comprising a series of units need not be limited to those units, but may include other units that are not clearly listed or inherent to these processes, methods, products or equipment.

[0062] See also Figure 1 , Figure 1 FIG. 1 shows a schematic diagram of the architecture of a customer purchasing power prediction system. The system may include a terminal 100 and a server 200. The server 200 may include one or more servers ( Figure 1 In the example, a server is included, and the server 200 can provide the method provided in the embodiment of the present application for one or more terminals.

[0063] Among them, an application for customer purchasing power prediction can be installed on the terminal 100. The above application and web page can provide an interface. The terminal 100 can receive relevant parameters entered by the user on the customer purchasing power prediction interface and send the above parameters to the server 200. The server 200 can obtain processing results based on the received parameters and return the processing results to the terminal 100.

[0064] It should be understood that in some optional implementations, the terminal 100 can also complete the action of obtaining the processing result based on the received parameters by itself without the cooperation of the server, and the embodiments of the present application are not limited to this.

[0065] Next describe Figure 1 The product form of the mid-terminal 100;

[0066] The terminal 100 in the embodiment of the present application can be a mobile phone, a tablet computer, a wearable device, a vehicle-mounted device, an augmented reality (AR) / virtual reality (VR) device, a laptop computer, an ultra-mobile personal computer (UMPC), a netbook, a personal digital assistant (PDA), etc., and the embodiment of the present application does not impose any limitation on this.

[0067] Figure 2 An optional hardware structure diagram of the terminal 100 is shown.

[0068] refer to Figure 2 As shown, the terminal 100 may include a radio frequency unit 110, a memory 120, an input unit 130, a display unit 140, a camera 150 (optional), an audio circuit 160 (optional), a speaker 161 (optional), a microphone 162 (optional), an earphone jack 163 (optional), a processor 170, an external interface 180, a power supply 190 and other components. Those skilled in the art will appreciate that Figure 2 These are merely examples of terminals or multi-function devices and do not constitute limitations on the terminals or multi-function devices, which may include more or fewer components than those shown in the figures, or combinations of certain components, or different components.

[0069] The input unit 130 can be used to receive input digital or character information, and generate key signal input related to the user settings and function control of the portable multifunctional device. Specifically, the input unit 130 may include a touch screen 131 (optional) and / or other input devices 132. The touch screen 131 can collect the user's touch operations on or near it (such as the user's operation on or near the touch screen using any suitable object such as fingers, joints, stylus, etc.), and drive the corresponding connection device according to a pre-set program. The touch screen can detect the user's touch action on the touch screen, convert the touch action into a touch signal and send it to the processor 170, and can receive and execute the command sent by the processor 170; the touch signal at least includes the touch point coordinate information. The touch screen 131 can provide an input interface and an output interface between the terminal 100 and the user. In addition, the touch screen can be implemented using multiple types such as resistive, capacitive, infrared and surface acoustic wave. In addition to the touch screen 131, the input unit 130 can also include other input devices. Specifically, other input devices 132 may include, but are not limited to, one or more of a physical keyboard, function keys (such as a volume control key, a switch key, etc.), a trackball, a mouse, a joystick, and the like.

[0070] Among them, the input device 132 can receive input data and the like.

[0071] The display unit 140 may be used to display information input by the user or provided to the user, various menus of the terminal 100, interactive interfaces, file displays, and / or playback of any multimedia files. In the embodiment of the present application, the display unit 140 may be used to display an interface for customer purchasing power prediction, processing results, etc.

[0072] The memory 120 can be used to store instructions and data. The memory 120 can mainly include an instruction storage area and a data storage area. The data storage area can store various data, such as multimedia files, texts, etc.; the instruction storage area can store software units such as operating systems, applications, instructions required for at least one function, or their subsets and extensions. It can also include a non-volatile random access memory; provide the processor 170 with hardware, software and data resources including management of computing and processing equipment, and support control software and applications. It is also used for the storage of multimedia files, and the storage of running programs and applications.

[0073] The processor 170 is the control center of the terminal 100. It uses various interfaces and lines to connect various parts of the entire terminal 100. By running or executing instructions stored in the memory 120 and calling data stored in the memory 120, it executes various functions of the terminal 100 and processes data, thereby controlling the terminal device as a whole. Optionally, the processor 170 may include one or more processing units; preferably, the processor 170 may integrate an application processor and a modem processor, wherein the application processor mainly processes the operating system, user interface and application program, and the modem processor mainly processes wireless communication. It is understandable that the above-mentioned modem processor may not be integrated into the processor 170. In some embodiments, the processor and the memory may be implemented on a single chip, and in some embodiments, they may also be implemented separately on separate chips. The processor 170 may also be used to generate corresponding operation control signals, send them to corresponding components of the computing and processing device, read and process data in the software, especially read and process data and programs in the memory 120, so that each functional module therein performs corresponding functions, thereby controlling the corresponding components to act according to the requirements of the instructions.

[0074] Among them, the memory 120 can be used to store software codes related to the customer purchasing power prediction method, the processor 170 can execute the steps of the customer purchasing power prediction method, and can also schedule other units (such as the above-mentioned input unit 130 and display unit 140) to achieve corresponding functions.

[0075] The radio frequency unit 110 (optional) can be used for receiving and sending information or receiving and sending signals during a call, for example, after receiving the downlink information of the base station, it is sent to the processor 170 for processing; in addition, the designed uplink data is sent to the base station. Generally, the RF circuit includes but is not limited to an antenna, at least one amplifier, a transceiver, a coupler, a low noise amplifier (Low Noise Amplifier, LNA), a duplexer, etc. In addition, the radio frequency unit 110 can also communicate with network devices and other devices through wireless communication. The wireless communication can use any communication standard or protocol, including but not limited to Global System of Mobile communication (Global System of Mobile communication, GSM), General Packet Radio Service (General Packet Radio Service, GPRS), Code Division Multiple Access (Code Division Multiple Access, CDMA), Wideband Code Division Multiple Access (Wideband Code Division Multiple Access, WCDMA), Long Term Evolution (Long Term Evolution, LTE), email, Short Messaging Service (SMS), etc.

[0076] In this embodiment of the present application, the RF unit 110 can send the customer's purchasing power data, purchase weight and other data to the server 200, and receive the processing results sent by the server 200.

[0077] It should be understood that the radio frequency unit 110 is optional and can be replaced by other communication interfaces, such as a network port.

[0078] The terminal 100 also includes a power supply 190 (such as a battery) for supplying power to various components. Preferably, the power supply can be logically connected to the processor 170 through a power management system, so that the power management system can manage functions such as charging, discharging, and power consumption.

[0079] The terminal 100 further includes an external interface 180 , which may be a standard Micro USB interface or a multi-pin connector, and may be used to connect the terminal 100 to communicate with other devices, or to connect a charger to charge the terminal 100 .

[0080] Although not shown, the terminal 100 may also include a flashlight, a wireless fidelity (WiFi) module, a Bluetooth module, sensors with different functions, etc., which are not described in detail here. Some or all of the methods described below may be applied in the following embodiments. Figure 2 In the terminal 100 shown.

[0081] Next describe Figure 1 The product form of the server 200;

[0082] Figure 3 A structural diagram of a server 200 is provided, such as Figure 3 As shown, the server 200 includes a bus 201, a processor 202, a communication interface 203, and a memory 204. The processor 202, the memory 204, and the communication interface 203 communicate with each other via the bus 201.

[0083] The bus 201 may be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus. The bus may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 3 Only one thick line is used in the diagram, but this does not mean that there is only one bus or only one type of bus.

[0084] The processor 202 may be any one or more of a central processing unit (CPU), a graphics processing unit (GPU), a microprocessor (MP), or a digital signal processor (DSP).

[0085] The memory 204 may include a volatile memory, such as a random access memory (RAM). The memory 204 may also include a non-volatile memory, such as a read-only memory (ROM), a flash memory, a hard disk drive (HDD), or a solid state drive (SSD).

[0086] The memory 204 may be used to store software codes related to the customer purchasing power prediction method, and the processor 202 may execute the steps of the customer purchasing power prediction method of the chip, and may also schedule other units to implement corresponding functions.

[0087] It should be understood that the above-mentioned terminal 100 and server 200 can be centralized or distributed devices, and the processors in the above-mentioned terminal 100 and server 200 (such as processor 170 and processor 202) can be hardware circuits (such as application specific integrated circuit (ASIC), field-programmable gate array (FPGA), general-purpose processor, digital signal processor (DSP), microprocessor or microcontroller, etc.), or a combination of these hardware circuits. For example, the processor can be a hardware system with an instruction execution function, such as a CPU, DSP, etc., or a hardware system without an instruction execution function, such as an ASIC, FPGA, etc., or a combination of the above-mentioned hardware systems without an instruction execution function and hardware systems with an instruction execution function.

[0088] At present, the SASRec prediction model based on Transformer (a deep learning model based on the attention mechanism) is widely used in the field of customer purchase product prediction. It uses the product sequence of customer interaction as input for modeling. In order to avoid the impact of the sparsity of customer interaction behavior, the learnable position vector is used as the sequence feature and added to the product input feature to obtain the input variable. The stacked single self-attention mechanism encoding layer is used to improve the optimization efficiency through Dropout, residual connection and variance normalization methods to reduce the problems of data overfitting and gradient disappearance caused by the deepening of the network layer. The final prediction result is obtained by matrix decomposition multiplication of the output modeled based on the product features of each customer and the product initialization encoding variable.

[0089] However, the input features of the SASRec prediction model are relatively simple. It only uses product serialization features for modeling, and lacks a modeling method for user serialization features. Moreover, the serialization features of products are not fully represented, and it is impossible to extract the multi-peak interests of users with multiple considerations. In reducing the biggest problem faced by collaborative filtering methods: data sparsity, the SASRec prediction model fails to fully utilize product context information to enrich product feature representation and reduce the impact of data sparsity.

[0090] In addition, although the SASRec prediction model can capture the nonlinear deep representations between customer-based products, it ignores the structured interaction characteristics between customer products. That is, it fails to effectively use the deep representations of customers and products to perceive the interaction characteristics of customer products, and thus fails to use the deep representations to more accurately adjust the customer and product implicit factor model vectors. It fails to fully perceive the local characteristics of extracted purchase information and the complementary global model characteristics, and cannot provide customers with more accurate purchasing power predictions in all scenarios.

[0091] In order to solve the above problems, the embodiment of the present application provides a method for predicting customer purchasing power. The method for predicting customer purchasing power of the embodiment of the present application is described in detail below with reference to the accompanying drawings.

[0092] Reference Figure 4 , Figure 4 A schematic diagram of a method for predicting customer purchasing power provided in an embodiment of the present application is shown in FIG. Figure 4 As shown, a customer purchasing power prediction method provided in an embodiment of the present application may include steps 401 to 409, and these steps are described in detail below.

[0093] 401. Obtain first purchasing power data representing the risk level of the customer when purchasing a product during a statistical period, second purchasing power data obtained based on a scoring questionnaire and a final transaction operation, and serialized purchasing weight data.

[0094] Specifically, the historical data of customers’ financial product purchases (including product risk level, purchase time, and product category, etc.) and customer profiles (including the validity period of customer reputation questionnaires, property limits, etc.) are collected and stored in the database. The risk interaction data obtained above is classified and normalized into 0-5 data representations, and the customer’s purchase time series weights are calculated and saved in the database. At the same time, the serialized weight matrix, risk scoring questionnaire, and customer transaction behavior secondary operation data (such as buying, selling, and other secondary operations) are obtained.

[0095] 402. Split the first purchasing power data into a customer side and a product side to obtain a customer side purchasing power matrix and a product side purchasing power matrix.

[0096] 403. Preprocess the second purchasing power data into a customer implicit factor query vector corresponding to the customer-side purchasing power matrix and a product implicit factor query vector corresponding to the product-side purchasing power matrix.

[0097] 404. Split the serialized purchase weight data to obtain a customer-side purchase weight matrix and a product-side purchase weight matrix.

[0098] Specifically, according to the features of the Transformer convolutional autoencoder and the parallel calculation features from the customer side and the product side, the customer-product purchasing power interaction matrix P stored in the above database is transformed into n×m Classification is processed into a customer-based square purchasing power matrix and product-based square purchasing power matrix

[0099] Among them, X i / Y j (i=1,2,...,n / j=1,2,...,m) represent the customer side purchasing power matrix and the product side purchasing power matrix respectively. ceil(·) means returning the smallest integer greater than or equal to the specified expression, n represents the number of rows, m represents the number of columns, U represents the customer side, and V represents the product side.

[0100] Use the customer product serialization weight preprocessor to process the serialized purchase weight data into the customer-side purchase weight matrix and the product-side purchase weight matrix, where the customer product serialization weight preprocessor is:

[0101]

[0102] Among them, X W represents the customer-side purchase weight matrix, Y W represents the product-side purchase weight matrix, B T The timestamp of the customer interaction, N T The current time stamp.

[0103] The customer risk assessment questionnaire can be divided into customer financial status assessment, investment experience assessment, investment purpose assessment, liquidity requirement assessment, return expectation assessment, risk preference assessment, risk loss tolerance assessment and the secondary operation dimension of the corresponding product after the customer holds a position from the subjective and objective dimensions, and refined into customer-based and product-based query input vectors. A two-layer fully connected neural network with an activation function and a softmax normalization function are used to extract the attention query vectors of customers and products.

[0104] In order to reduce the impact of data sparsity on the model-based collaborative filtering model, the implicit factor model of the deep learning model based on the attention mechanism is used to initialize the structured implicit factor model of the probability matrix decomposition model:

[0105] G U / G V =AttConvAEU L / 2 / AttConvAEV L / 2

[0106] in, (i=1,2,...,n / j=1,2,...,m) are the customer implicit factor model and product implicit factor model respectively. AttConvAE(·) represents the output matrix of the middle layer of the deep learning model, Represents a d-row i / j-column entity matrix, U L / 2 Represents the middle layer on the client side, V L / 2 Indicates the middle layer on the product side.

[0107] 405. Input the customer-side data into a deep learning model based on an attention mechanism to perform nonlinear feature extraction to obtain a first customer-side implicit factor model. The customer-side data includes: a customer-side purchasing power matrix, a customer implicit factor query vector, and a customer-side purchase weight matrix.

[0108] 406. Input the product side data into the deep learning model for nonlinear feature extraction to obtain a first product side implicit factor model. The product side data includes: a product side purchasing power matrix, a product implicit factor query vector, and a product side purchase weight matrix.

[0109] Specifically, the deep learning model based on the attention mechanism includes: a convolutional encoder, an implicit factor extractor and a convolutional decoder. The process of nonlinear feature extraction in the deep learning model includes:

[0110] The client-side data and the product-side data are processed based on the convolution encoder to obtain the client-side serialized feature data and the product-side serialized feature data. The client-side serialized feature data and the product-side feature data are extracted based on the implicit factor extractor to obtain the first client-side implicit factor model and the first product-side implicit factor model.

[0111] The client-side serialized feature data and the product-side serialized feature data restored by the implicit factor extractor are restored based on a convolution decoder to obtain a client-side purchasing power matrix and a product-side purchasing power matrix.

[0112] Among them, refer to Figure 5 The convolutional encoder shown includes: a first convolutional encoding layer and a second convolutional encoding layer, the first convolutional encoding layer includes: at least one convolutional encoding unit, the convolutional encoding unit includes a first convolutional layer and a pooling layer connected in sequence; the second convolutional encoding layer includes: a first multi-head attention mechanism layer, a first residual connection and normalization layer, and a second convolutional layer; based on the convolutional encoder, the client-side data and the product-side data are processed respectively to obtain the client-side serialized feature data and the product-side serialized feature data, including:

[0113] Based on the first convolutional coding layer, feature extraction is performed on the customer side purchasing power matrix and the customer side purchasing weight matrix to obtain customer side purchasing power feature data.

[0114] Based on the first convolutional coding layer, feature extraction is performed on the product-side purchasing power matrix and the product-side purchasing weight matrix to obtain product-side purchasing power feature data.

[0115] Based on the second convolutional coding layer, feature extraction is performed on the customer-side purchasing power feature data and the customer implicit factor query vector to obtain customer-side serialized feature data.

[0116] Based on the second convolutional coding layer, feature extraction is performed on the product-side purchasing power feature data and the product implicit factor query vector to obtain the product-side serialized feature data.

[0117] The implicit factor extractor includes: a fully connected layer and a third convolutional layer. Based on the implicit factor extractor, feature extraction is performed on the serialized feature data on the client side and the feature data on the product side, respectively, to obtain a first client side implicit factor model and a first product side implicit factor model, including:

[0118] Based on the fully connected layer, feature extraction is performed on the client-side serialized feature data and the product-side feature data respectively to obtain the first client-side implicit factor model and the first product-side implicit factor model.

[0119] Based on the third convolutional layer, the first client-side implicit factor model and the first product-side implicit factor model are restored respectively to obtain client-side feature mapping data and product-side feature mapping data.

[0120] The convolution decoder includes: a first convolution decoding layer and a second convolution decoding layer, the first convolution decoding layer includes: a second multi-head attention mechanism layer, a second residual connection and normalization layer, and a fourth convolution layer, the second convolution decoding layer includes convolution decoding units with the same number as the convolution encoding units, and each convolution decoding unit includes: a fifth convolution layer and an upsampling layer connected in sequence; based on the convolution decoder, the client-side serialized feature data and the product-side serialized feature data restored by the implicit factor extractor are restored to obtain the client-side purchasing power matrix and the product-side purchasing power matrix, including:

[0121] Based on the first convolution decoding layer, the client-side feature mapping data and the client-side serialized feature data are restored to obtain the client-side purchasing power feature data.

[0122] Based on the first convolution decoding layer, the product side feature mapping data and the product side serialized feature data are restored to obtain the product side purchasing power feature data;

[0123] Based on the first convolution decoding layer, the customer-side purchasing power feature data and the product-side purchasing power feature data are restored respectively to obtain the customer-side purchasing power matrix and the product-side purchasing power matrix.

[0124] Specifically, the first convolutional coding layer can use two convolutional layers and pooling layers (i.e., two convolutional coding units) from the perspective of customers and products respectively. Through the translation equilaterality, parameter sharing, and sparse interaction properties of the convolution operation, the number of parameters is greatly reduced and better local perception is obtained. And through the nonlinear activation function added after the convolution layer, better data representation and accurate deep local features are learned. After the activation function is output, in order to enhance the robustness of the algorithm, retain the main features after convolution, and reduce the size of the feature map to avoid overfitting, the maximum pooling layer can be used to realize feature extraction. The encoding formula of the first convolutional coding layer is:

[0125]

[0126] in, They represent the purchasing power feature data of the customer side and the purchasing power feature data of the product side, * represents the convolution operation, t1 represents the number of convolution kernels, ζ(·) represents the nonlinear activation function, ψ(·) represents the maximum pooling operation, and d t1 ×d t1 Represents the dimension of the entity matrix after the t1 convolution kernel two-dimensional convolution pooling operation, and Represent the two-dimensional convolution kernel weights of customers and products respectively, and Respectively and The bias, P ui represents the customer side purchasing power matrix, represents the product side purchase weight matrix, P vj represents the product side purchasing power matrix, Represents the product-side purchase weight matrix.

[0127] Then, the second convolutional coding layer is used to obtain the multi-dimensional deep implicit factor model of customer and product attention. Here, the second convolutional coding layer is obtained by replacing the feedforward network of the Transformer coding layer with a convolution operation (where the Transformer coding layer consists of a multi-head attention mechanism layer, a residual connection and a normalization layer, and a feedforward network layer) to obtain the customer-product multi-peak serialization features:

[0128]

[0129] in, Respectively represent the multi-head attention features on the customer side and the multi-head attention features on the product side, MutiAtt t2 (·) indicates that t2 self-attention encoding layers are used to obtain the multi-head attention output matrix. Q, K, and Val represent the input variables of the self-attention network in the multi-head attention mechanism layer. in represents the purchasing power characteristic data on the customer side or the purchasing power characteristic data on the product side, W is a matrix of the same order, represents the query input vector based on customer i and product j, which is refined from the customer reputation questionnaire and secondary purchase behavior. Softmax(·) is a column-normalized function, ζ(·) represents the activation function, and the Reshape(·) operation represents the matrix transformation. dimension matrix, where in order to prevent QK T The inner product is too large, so it is divided by the number of columns in the Q and K matrices d t1 The square root of . In order to prevent network degradation, residual connections and output value normalization LayerNorm methods are used.

[0130] They represent the serialized feature data on the client side and the serialized feature data on the product side respectively. It means that t3 three-dimensional convolution kernels are used to perform three-dimensional convolution operations on the client side and the product side respectively. express As the depth increases, the number of feature layers captured continues to increase, which makes the implicit factor features of customers and products continue to increase, and more deep nonlinear representations are learned.

[0131] The implicit factor extractor uses a fully connected layer to map the multi-peak deep nonlinear representation of customers and products (i.e., serialized feature data on the customer side and serialized feature data on the product side) to the customer and product implicit factor model space. Then, convolution is used instead of deconvolution operation to map the implicit factor space back to the customer product feature encoding space, laying the foundation for the next step of convolution decoder to restore the encoding matrix:

[0132]

[0133] in, denote the output matrices of the fully connected layers on the client side and the product side, respectively, where L / 2 denotes the middle layer of the second convolutional coding layer (i.e., the fully connected layer), FC(·) denotes a single-layer feedforward neural network with a nonlinear activation function, Reshape(·) indicates that the implicit factor space model is converted into a square convolutional layer input specification by using the same square processing mechanism. represents t3 two-dimensional convolution kernels on the client side and product side, Indicates the corresponding bias.

[0134] The convolution decoder corresponds to the reverse process of the convolution encoder above. It maps the feature vector of the implicit factor model space to the feature encoding space based on the previous step. The matrix is ​​used as the input of the query vector, and K and V of the multi-head attention mechanism are used As input. Since the query vector is used as the query layer after perception optimization, the query vector reduces the impact of data sparsity. The residual connection further obtains the multi-peak decoding features that the customer product pays more attention to after decoding, which can be restored to a richer customer product interaction matrix, improving the decoding efficiency and the customer product purchase accuracy prediction effect:

[0135]

[0136] Here, θ(·) represents the bilinear interpolation upsampling operation. Indicates using t2 self-attention decoding layers to obtain the multi-head attention output matrix, represents matrices of the same order, They represent the restored customer-side purchasing power characteristic data and product-side purchasing power characteristic data respectively, They represent the restored customer-side purchasing power matrix and product-side purchasing power matrix respectively. represents t1 two-dimensional convolution kernels on the client side and product side, Indicates the corresponding bias.

[0137] Finally, a two-dimensional convolution is used to restore the convolution to a two-dimensional matrix, and then a fully connected layer is used to restore it to a customer-based and product-based customer product purchasing power vector:

[0138]

[0139] where X′ i / Y′ j They represent the customer-side product purchasing power vector and the product-side purchasing power vector respectively, Represent the one-dimensional convolution kernels on the client side and product side respectively, Indicates the corresponding bias.

[0140] 407. Based on the probability matrix decomposition model, structural features are extracted from the first customer-side implicit factor model and the first product-side implicit factor model respectively to obtain a second customer-side implicit factor model and a second product-side implicit factor model.

[0141] Specifically, due to the good scalability of probability matrix decomposition, probability matrix decomposition can be used to predict customer purchasing power and mine the structured features of data in the customer-product matrix. The specific probability matrix decomposition model prediction formula is as follows:

[0142]

[0143] Among them, p(P|G U , G V , σ 2) represents the maximum likelihood function for the maximum a posteriori probability estimation of the probability matrix decomposition model parameters, P represents the purchasing power prediction, σ 2 is the variance, It means the mean The variance is σ 2 The Gaussian normal distribution probability density function, I ij is an indicator function indicating that when customer i purchases product j, it is 1, otherwise it is 0. i and p j They represent the customer purchasing power implicit factor model that captures the purchasing power characteristics of a specific product purchased by a customer, and the product purchase implicit factor model that captures the purchasing power characteristics of a specific customer for a product. represents the predicted purchasing power of customer i for product j, p ij The real purchasing power of customer i to purchase product j.

[0144] 408. Combine the first customer-side implicit factor model, the first product-side implicit factor model, the second customer-side implicit factor model, and the second product-side implicit factor model to optimize the deep learning model and the probability matrix decomposition model.

[0145] Specifically, when performing nonlinear deep feature optimization, the second client-side implicit factor model is used as a priori induced function of the first client-side implicit factor model, and the objective function composed of the second client-side implicit factor model, the first client-side implicit factor model and the output client-side results of the deep learning model is trained and optimized.

[0146] The second product-side implicit factor model is used as a priori induced function of the first product-side implicit factor model, and the objective function composed of the second product-side implicit factor model, the first product-side implicit factor model and the output product-side results of the deep learning model is trained and optimized.

[0147] For example, in order to extract more accurate nonlinear deep features, the structured feature extraction method can be used. i and p j As a priori induction function of the nonlinear attention implicit factor model, the objective function of the deep learning model is trained and optimized:

[0148]

[0149] Among them, W U Represents the model parameters on the client side, It is W U The kth element of W V represents the model parameters on the product side, It is W V The kth element of ijThe indicator function indicates that when customer i purchases product j, it is 1, otherwise it is 0. Constant is another static constant. The L2 regularization function is used to avoid overfitting caused by prior-induced training. U represents the training parameter vector on the client side, λ V represents the training parameter vector on the product side, The client-side nonlinear local features obtained by the deep learning model, is the nonlinear local feature of the product side, X i is the real customer side purchasing power matrix, X′ i is the predicted customer purchasing power matrix, Y j is the real product-side purchasing power matrix, Y′ j is the predicted product-side purchasing power matrix, and λ is the training parameter vector of the deep learning model.

[0150] Similarly, in order to obtain more accurate customer-product structural features, the dot product of the customer purchasing power implicit factor model and the product purchase preference implicit factor model is used to obtain the probability matrix decomposition prediction result, that is, represents the predicted purchasing power of customer i for product j. In order to supplement the nonlinear deep local features, the attention implicit factor model extracted by the above deep learning model is used as p i and p j The prior induced function of , the following structured feature optimization objective function is obtained:

[0151]

[0152] In this way, the deep learning model with attention mechanism is organically integrated with the probabilistic matrix decomposition model, which reduces the overfitting and underfitting phenomena of the probabilistic matrix decomposer due to data sparsity and improves the accuracy of customer product purchasing power prediction.

[0153] 409. Predict the purchasing power of target users based on the optimized probability matrix decomposition model.

[0154] Specifically, after the above nonlinear deep feature optimizer and structured feature optimizer are optimized and converged, the customer-based implicit factor model p is obtained. i and the product-based implicit factor model p j Perform inner product operation to obtain customer i’s purchasing power score for product j

[0155]

[0156] Among them, ε i represents the user latent bias that captures the user bias for a specific item, ε jItem-latent offset that captures specific user-item bias.

[0157] This customer purchasing power prediction method uses a deep learning model based on the self-attention mechanism to capture the multi-modal dynamic deep features of customers and products in parallel using product serialization features and customer serialization features. In addition, the structured features captured by the probability matrix decomposition model and the context query features of the risk score questionnaire reduce the impact of data sparsity on the collaborative filtering model, promote the deep learning model to converge towards a more efficient direction, reduce the deviation of customer purchasing power prediction, and improve the accuracy of customer purchasing power score prediction.

[0158] In addition, the deep local features of customers and products extracted by the deep learning model with attention mechanism make up for the lack of nonlinear deep representation of the probability matrix decomposition implicit factor model, and promote the prior correction of the implicit factor probability space; on the other hand, based on the fact that the two share the same input source, the updated interaction characteristics of the probability matrix decomposition implicit factor model will collaboratively optimize the dual convolutional autoencoder model. The deep perception optimization layer promotes the joint perception optimization learning between models, fully explores the nonlinear deep features and structured characteristics of customer-product interaction data, learns the effective representation of sparse data, reduces data sparsity, and improves the accuracy of purchasing power score prediction.

[0159] On this basis, it can provide marketers with more accurate purchasing power references, so that the recommended products are more in line with the user's purchasing power, thereby improving user experience and the success rate of recommendations.

[0160] A method for predicting customer purchasing power provided by an embodiment of the present application is introduced above, and a device for executing the above method for predicting customer purchasing power is introduced below.

[0161] See also Figure 6 , Figure 6 This is a schematic diagram of the structure of a customer purchasing power prediction device provided in an embodiment of the present application. Figure 6 As shown, the customer purchasing power prediction device comprises:

[0162] The purchasing power data acquisition module 601 is used to obtain the first purchasing power data representing the risk level of the customer when purchasing the product during the statistical period, the second purchasing power data obtained based on the scoring questionnaire and the final transaction operation, and the serialized purchasing weight data.

[0163] The first data preprocessing module 602 is used to split the first purchasing power data into the customer side and the product side to obtain a customer side purchasing power matrix and a product side purchasing power matrix.

[0164] The second data preprocessing module 603 is used to preprocess the second purchasing power data into a customer implicit factor query vector corresponding to the customer-side purchasing power matrix and a product implicit factor query vector corresponding to the product-side purchasing power matrix.

[0165] The third data preprocessing module 604 is used to split the serialized purchase weight data to obtain a customer-side purchase weight matrix and a product-side purchase weight matrix.

[0166] The client-side feature data extraction module 605 is used to input the client-side data into a deep learning model based on the attention mechanism for nonlinear feature extraction to obtain a first client-side implicit factor model. The client-side data includes: the client-side purchasing power matrix, the client implicit factor query vector and the client-side purchase weight matrix.

[0167] The product-side feature data extraction module 606 is used to input the product-side data into the deep learning model for nonlinear feature extraction to obtain a first product-side implicit factor model. The product-side data includes: the product-side purchasing power matrix, the product implicit factor query vector and the product-side purchase weight matrix.

[0168] The implicit factor model extraction module 607 is used to extract structural features of the first client-side implicit factor model and the first product-side implicit factor model based on a probability matrix decomposition model to obtain a second client-side implicit factor model and a second product-side implicit factor model.

[0169] A joint optimization training module 608 is used to jointly optimize the deep learning model and the probability matrix decomposition model by combining the first client-side implicit factor model, the first product-side implicit factor model, the second client-side implicit factor model and the second product-side implicit factor model. And,

[0170] The purchasing power prediction module 609 is used to predict the purchasing power of the target user based on the optimized probability matrix decomposition model.

[0171] In a possible implementation, the process in which the joint optimization training module 608 jointly optimizes the deep learning model and the probability matrix decomposition model with the first client-side implicit factor model, the first product-side implicit factor model, the second client-side implicit factor model, and the second product-side implicit factor model includes:

[0172] Using the second client-side implicit factor model as a priori induced function of the first client-side implicit factor model, and performing training optimization on an objective function composed of the second client-side implicit factor model, the first client-side implicit factor model, and an output client-side result of the deep learning model;

[0173] The second product-side implicit factor model is used as a priori induced function of the first product-side implicit factor model, and the objective function composed of the second product-side implicit factor model, the first product-side implicit factor model and the output product-side result of the deep learning model is trained and optimized.

[0174] In a possible implementation, the process in which the joint optimization training module 608 jointly optimizes the deep learning model and the probability matrix decomposition model with the first client-side implicit factor model, the first product-side implicit factor model, the second client-side implicit factor model, and the second product-side implicit factor model includes:

[0175] The first customer-side implicit factor model and the first product-side implicit factor model are used as prior induction functions to train and optimize the objective function composed of the second customer-side implicit factor model, the first customer-side implicit factor model, the first product-side implicit factor model and the second product-side implicit factor model; the initial product-side implicit factor model and the initial customer-side implicit factor model of the probability matrix decomposition model are obtained according to the intermediate layer output of the deep learning model.

[0176] In a possible implementation, the deep learning model includes: a convolution encoder, an implicit factor extractor and a convolution decoder, and the process of performing nonlinear feature extraction in the client-side feature data extraction module 605 and the client-side feature data extraction module 606 deep learning model includes:

[0177] Based on the convolution encoder, the client-side data and the product-side data are processed respectively to obtain client-side serialization feature data and product-side serialization feature data;

[0178] Based on the implicit factor extractor, feature extraction is performed on the client-side serialized feature data and the product-side feature data respectively to obtain the first client-side implicit factor model and the first product-side implicit factor model;

[0179] Based on the convolution decoder, the client-side serialized feature data and the product-side serialized feature data restored by the implicit factor extractor are restored to obtain the client-side purchasing power matrix and the product-side purchasing power matrix.

[0180] In a possible implementation, the convolutional encoder includes: a first convolutional encoding layer and a second convolutional encoding layer, the first convolutional encoding layer includes: at least one convolutional encoding unit, the convolutional encoding unit includes a first convolutional layer and a pooling layer connected in sequence; the second convolutional encoding layer includes: a first multi-head attention mechanism layer, a first residual connection and normalization layer, and a second convolutional layer; the client-side feature data extraction module 605 and the client-side feature data extraction module 606 process the client-side data and the product-side data respectively based on the convolutional encoder to obtain the client-side serialized feature data and the product-side serialized feature data, including:

[0181] Based on the first convolutional coding layer, feature extraction is performed on the customer-side purchasing power matrix and the customer-side purchase weight matrix to obtain customer-side purchasing power feature data;

[0182] Based on the first convolutional coding layer, feature extraction is performed on the product-side purchasing power matrix and the product-side purchasing weight matrix to obtain product-side purchasing power feature data;

[0183] Performing feature extraction on the customer-side purchasing power feature data and the customer implicit factor query vector based on the second convolutional coding layer to obtain the customer-side serialized feature data;

[0184] Based on the second convolutional coding layer, feature extraction is performed on the product-side purchasing power feature data and the product implicit factor query vector to obtain the product-side serialized feature data.

[0185] In a possible implementation, the implicit factor extractor includes: a fully connected layer and a third convolutional layer, and the client-side feature data extraction module 605 and the client-side feature data extraction module 606 perform feature extraction on the client-side serialized feature data and the product-side feature data based on the implicit factor extractor, respectively, to obtain the first client-side implicit factor model and the first product-side implicit factor model. The process includes:

[0186] Based on the fully connected layer, feature extraction is performed on the client-side serialized feature data and the product-side feature data respectively to obtain the first client-side implicit factor model and the first product-side implicit factor model;

[0187] The first client-side implicit factor model and the first product-side implicit factor model are respectively restored based on the third convolutional layer to obtain client-side feature mapping data and product-side feature mapping data.

[0188] In a possible implementation, the convolution decoder includes: a first convolution decoding layer and a second convolution decoding layer, the first convolution decoding layer includes: a second multi-head attention mechanism layer, a second residual connection and normalization layer, and a fourth convolution layer, the second convolution decoding layer includes convolution decoding units with the same number as the convolution encoding units, and each of the convolution decoding units includes: a fifth convolution layer and an upsampling layer connected in sequence; the client-side feature data extraction module 605 and the client-side feature data extraction module 606 restore the client-side serialized feature data and the product-side serialized feature data restored by the implicit factor extractor based on the convolution decoder to obtain the client-side purchasing power matrix and the product-side purchasing power matrix. The process includes:

[0189] Based on the first convolution decoding layer, the client-side feature mapping data and the client-side serialized feature data are restored to obtain the client-side purchasing power feature data;

[0190] Based on the first convolution decoding layer, the product-side feature mapping data and the product-side serialized feature data are restored to obtain the product-side purchasing power feature data;

[0191] Based on the first convolution decoding layer, the customer-side purchasing power feature data and the product-side purchasing power feature data are restored respectively to obtain the customer-side purchasing power matrix and the product-side purchasing power matrix.

[0192] The present application also provides an electronic device in an embodiment. Figure 7 As shown, it shows a schematic diagram of the structure of an electronic device suitable for implementing the embodiments of the present application. The electronic device in the embodiments of the present application may include but is not limited to fixed terminals such as mobile phones, notebook computers, PDAs (personal digital assistants), PADs (tablet computers), desktop computers, etc. Figure 7 The electronic device shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0193] like Figure 7 As shown, the electronic device may include a processing device (e.g., a central processing unit, a graphics processing unit, etc.) 701, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 702 or a program loaded from a storage device 708 into a random access memory (RAM) 703. When the electronic device is powered on, various programs and data required for the operation of the electronic device are also stored in the RAM 703. The processing device 701, the ROM 702, and the RAM 703 are connected to each other via a bus 704. An input / output (I / O) interface 705 is also connected to the bus 704.

[0194] Typically, the following devices may be connected to the I / O interface 705: an input device 706 including, for example, a touch screen, a touch pad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; an output device 707 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; a storage device 708 including, for example, a memory card, a hard disk, etc.; and a communication device 709. The communication device 709 may allow the electronic device to communicate with other devices wirelessly or by wire to exchange data. Although Figure 7 An electronic device having various devices is shown, but it should be understood that it is not required to implement or possess all the devices shown. More or fewer devices may be implemented or possessed instead.

[0195] An embodiment of the present application also provides a computer program product including computer-readable instructions. When the computer-readable instructions are executed on an electronic device, the electronic device implements any one of the customer purchasing power prediction methods provided in the embodiments of the present application.

[0196] A computer-readable storage medium is also provided in an embodiment of the present application. The storage medium carries one or more computer programs. When the one or more computer programs are executed by an electronic device, the electronic device can implement any customer purchasing power prediction method provided in the embodiment of the present application.

[0197] It is understandable that before using the technical solutions disclosed in the embodiments of the present disclosure, the types, scope of use, usage scenarios, etc. of the personal information involved in the present disclosure should be informed to the user and the user's authorization should be obtained in an appropriate manner in accordance with relevant laws and regulations.

[0198] It should also be noted that the device embodiments described above are merely schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed over multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. In addition, in the drawings of the device embodiments provided by the present application, the connection relationship between the modules indicates that there is a communication connection between them, which may be specifically implemented as one or more communication buses or signal lines.

[0199] Through the description of the above implementation mode, the technicians in the field can clearly understand that the present application can be implemented by means of software plus necessary general hardware, and of course, it can also be implemented by special hardware including special integrated circuits, special CPUs, special memories, special components, etc. In general, all functions completed by computer programs can be easily implemented by corresponding hardware, and the specific hardware structure used to implement the same function can also be various, such as analog circuits, digital circuits or special circuits. However, for the present application, software program implementation is a better implementation mode in more cases. Based on such an understanding, the technical solution of the present application is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a readable storage medium, such as a computer floppy disk, a U disk, a mobile hard disk, a ROM, a RAM, a disk or an optical disk, etc., including a number of instructions to enable a computer device (which can be a personal computer, a training device, or a network device, etc.) to execute the methods described in each embodiment of the present application.

[0200] In the above embodiments, all or part of the embodiments may be implemented by software, hardware, firmware or any combination thereof. When implemented by software, all or part of the embodiments may be implemented in the form of a computer program product.

[0201] The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions may be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions may be transmitted from a website site, a computer, a training device, or a data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) mode to another website site, computer, training device, or data center. The computer-readable storage medium may be any available medium that a computer can store or a data storage device such as a training device, a data center, etc. that includes one or more available media integrations. The available medium may be a magnetic medium, (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state drive (SSD)), etc.

Claims

1. A method for predicting customer purchasing power, characterized in that: include: Obtaining first purchasing power data representing the risk level of customers when purchasing products during the statistical period, second purchasing power data obtained based on the scoring questionnaire and the final transaction operation, and serialized purchasing weight data; Splitting the first purchasing power data into a customer side and a product side to obtain a customer side purchasing power matrix and a product side purchasing power matrix; Preprocessing the second purchasing power data into a customer implicit factor query vector corresponding to the customer-side purchasing power matrix and a product implicit factor query vector corresponding to the product-side purchasing power matrix; Splitting the serialized purchase weight data to obtain a customer-side purchase weight matrix and a product-side purchase weight matrix; Inputting the customer-side data into a deep learning model based on an attention mechanism for nonlinear feature extraction to obtain a first customer-side implicit factor model, wherein the customer-side data includes: the customer-side purchasing power matrix, the customer implicit factor query vector, and the customer-side purchase weight matrix; Inputting product-side data into the deep learning model for nonlinear feature extraction to obtain a first product-side implicit factor model, wherein the product-side data includes: the product-side purchasing power matrix, the product implicit factor query vector, and the product-side purchase weight matrix; Based on the probability matrix decomposition model, structural features are extracted from the first customer-side implicit factor model and the first product-side implicit factor model respectively to obtain a second customer-side implicit factor model and a second product-side implicit factor model; Combine the first client-side implicit factor model, the first product-side implicit factor model, the second client-side implicit factor model, and the second product-side implicit factor model to optimize the deep learning model and the probability matrix decomposition model; The purchasing power of target users is predicted based on the optimized probability matrix decomposition model.

2. The method for predicting customer purchasing power according to claim 1, characterized in that: The deep learning model and the probability matrix decomposition model are optimized by combining the first client-side implicit factor model, the first product-side implicit factor model, the second client-side implicit factor model, and the second product-side implicit factor model, including: Using the second client-side implicit factor model as a priori induced function of the first client-side implicit factor model, and performing training optimization on an objective function composed of the second client-side implicit factor model, the first client-side implicit factor model, and an output client-side result of the deep learning model; The second product-side implicit factor model is used as a priori induced function of the first product-side implicit factor model, and the objective function composed of the second product-side implicit factor model, the first product-side implicit factor model and the output product-side result of the deep learning model is trained and optimized.

3. The method for predicting customer purchasing power according to claim 1, characterized in that: The deep learning model and the probability matrix decomposition model are optimized by combining the first client-side implicit factor model, the first product-side implicit factor model, the second client-side implicit factor model, and the second product-side implicit factor model, including: The first customer-side implicit factor model and the first product-side implicit factor model are used as prior induction functions to train and optimize the objective function composed of the second customer-side implicit factor model, the first customer-side implicit factor model, the first product-side implicit factor model and the second product-side implicit factor model; the initial product-side implicit factor model and the initial customer-side implicit factor model of the probability matrix decomposition model are obtained according to the intermediate layer output of the deep learning model.

4. The method for predicting customer purchasing power according to claim 1, characterized in that: The deep learning model includes: a convolution encoder, an implicit factor extractor and a convolution decoder. The process of extracting nonlinear features in the deep learning model includes: Based on the convolution encoder, the client-side data and the product-side data are processed respectively to obtain client-side serialization feature data and product-side serialization feature data; Based on the implicit factor extractor, feature extraction is performed on the client-side serialized feature data and the product-side feature data respectively to obtain the first client-side implicit factor model and the first product-side implicit factor model; Based on the convolution decoder, the client-side serialized feature data and the product-side serialized feature data restored by the implicit factor extractor are restored to obtain the client-side purchasing power matrix and the product-side purchasing power matrix.

5. The method for predicting customer purchasing power according to claim 4, characterized in that: The convolutional encoder includes: a first convolutional encoding layer and a second convolutional encoding layer, the first convolutional encoding layer includes: at least one convolutional encoding unit, the convolutional encoding unit includes a first convolutional layer and a pooling layer connected in sequence; the second convolutional encoding layer includes: a first multi-head attention mechanism layer, a first residual connection and normalization layer, and a second convolutional layer; based on the convolutional encoder, the client-side data and the product-side data are processed respectively to obtain client-side serialized feature data and product-side serialized feature data, including: Based on the first convolutional coding layer, feature extraction is performed on the customer-side purchasing power matrix and the customer-side purchase weight matrix to obtain customer-side purchasing power feature data; Based on the first convolutional coding layer, feature extraction is performed on the product-side purchasing power matrix and the product-side purchasing weight matrix to obtain product-side purchasing power feature data; Performing feature extraction on the customer-side purchasing power feature data and the customer implicit factor query vector based on the second convolutional coding layer to obtain the customer-side serialized feature data; Based on the second convolutional coding layer, feature extraction is performed on the product-side purchasing power feature data and the product implicit factor query vector to obtain the product-side serialized feature data.

6. The method for predicting customer purchasing power according to claim 5, characterized in that: The implicit factor extractor includes: a fully connected layer and a third convolutional layer. Based on the implicit factor extractor, feature extraction is performed on the client-side serialized feature data and the product-side feature data respectively to obtain the first client-side implicit factor model and the first product-side implicit factor model, including: Based on the fully connected layer, feature extraction is performed on the client-side serialized feature data and the product-side feature data respectively to obtain the first client-side implicit factor model and the first product-side implicit factor model; The first client-side implicit factor model and the first product-side implicit factor model are respectively restored based on the third convolutional layer to obtain client-side feature mapping data and product-side feature mapping data.

7. The method for predicting customer purchasing power according to claim 6, characterized in that: The convolution decoder includes: a first convolution decoding layer and a second convolution decoding layer, the first convolution decoding layer includes: a second multi-head attention mechanism layer, a second residual connection and normalization layer, and a fourth convolution layer, the second convolution decoding layer includes convolution decoding units with the same number as the convolution encoding units, and each of the convolution decoding units includes: a fifth convolution layer and an upsampling layer connected in sequence; based on the convolution decoder, the client-side serialized feature data and the product-side serialized feature data restored by the implicit factor extractor are restored to obtain the client-side purchasing power matrix and the product-side purchasing power matrix, including: Based on the first convolution decoding layer, the client-side feature mapping data and the client-side serialized feature data are restored to obtain the client-side purchasing power feature data; Based on the first convolution decoding layer, the product-side feature mapping data and the product-side serialized feature data are restored to obtain the product-side purchasing power feature data; Based on the first convolution decoding layer, the customer-side purchasing power feature data and the product-side purchasing power feature data are restored respectively to obtain the customer-side purchasing power matrix and the product-side purchasing power matrix.

8. A customer purchasing power prediction device, characterized in that: include: A purchasing power data acquisition module is used to acquire first purchasing power data representing the risk level of customers when purchasing products during a statistical period, second purchasing power data obtained based on a scoring questionnaire and a final transaction operation, and serialized purchasing weight data; A first data preprocessing module, used for splitting the first purchasing power data into a customer side and a product side to obtain a customer side purchasing power matrix and a product side purchasing power matrix; A second data preprocessing module, used for preprocessing the second purchasing power data into a customer implicit factor query vector corresponding to the customer-side purchasing power matrix and a product implicit factor query vector corresponding to the product-side purchasing power matrix; A third data preprocessing module is used to split the serialized purchase weight data to obtain a customer-side purchase weight matrix and a product-side purchase weight matrix; A client-side feature data extraction module, used for inputting client-side data into a deep learning model based on an attention mechanism to perform nonlinear feature extraction to obtain a first client-side implicit factor model, wherein the client-side data includes: the client-side purchasing power matrix, the client implicit factor query vector, and the client-side purchase weight matrix; A product-side feature data extraction module, used for inputting product-side data into the deep learning model for nonlinear feature extraction to obtain a first product-side implicit factor model, wherein the product-side data includes: the product-side purchasing power matrix, the product implicit factor query vector and the product-side purchase weight matrix; An implicit factor model extraction module, used to extract structural features of the first client-side implicit factor model and the first product-side implicit factor model respectively based on a probability matrix decomposition model to obtain a second client-side implicit factor model and a second product-side implicit factor model; a joint optimization training module, used to jointly optimize the deep learning model and the probability matrix decomposition model by combining the first client-side implicit factor model, the first product-side implicit factor model, the second client-side implicit factor model and the second product-side implicit factor model; and The purchasing power prediction module is used to predict the purchasing power of target users based on the optimized probability matrix decomposition model.

9. An electronic device, characterized in that: The method comprises at least one processor and a memory connected to the processor, wherein: The memory is used to store computer programs; The processor is used to execute the computer program so that the electronic device can implement the customer purchasing power prediction method as described in any one of claims 1 to 7.

10. A computer storage medium, characterized in that: The storage medium carries one or more computer programs, and when the one or more computer programs are executed by an electronic device, the electronic device can implement the customer purchasing power prediction method as described in any one of claims 1 to 7.

Images