A user car purchase intention prediction method, device and equipment and storage medium

Abstract

The application provides a user car purchase intention prediction method and device, equipment and storage medium, and relates to the field of big data. The method comprises the following steps: obtaining first browsing information, second browsing information, call information and access information of a target user; establishing a plurality of prediction models through a long short-term memory network according to the first browsing information, the second browsing information, the call information and the access information, and obtaining a prediction probability output by each prediction model; performing weight assignment on each prediction model to obtain a model weight corresponding to each prediction model; and obtaining a car purchase intention probability of the target user according to the prediction probability output by each prediction model and the model weight corresponding to each prediction model. The method solves the problem that the existing online browsing car purchase intention model has low accuracy and cannot accurately predict the car purchase intention of a user.

Description

Technical Field

[0001] This application relates to the field of big data, and in particular to a method, apparatus, device, and storage medium for predicting users' car purchase intentions. Background Technology

[0002] In existing technologies, based on users' browsing information from automotive vertical media and automotive applications (APPs), a multi-layer feedforward (BP) neural network is used to establish an online browsing and car purchase intention model to identify potential car buyers.

[0003] However, existing online car purchase intention models only incorporate behavioral characteristics of browsing automotive media and automotive applications, resulting in low model accuracy and an inability to accurately predict users' car purchase intentions based on online car purchase intention models. Summary of the Invention

[0004] This application provides a method, apparatus, device, and storage medium for predicting user car purchase intentions, in order to solve the problem that existing online car purchase intention models have low accuracy and cannot accurately predict users' car purchase intentions.

[0005] Firstly, this application provides a method for predicting a user's car purchase intention, the method comprising:

[0006] The system obtains the target user's first browsing information, second browsing information, call information, and access information. The first browsing information is used to instruct the target user to browse the browsing history of automotive vertical media, the second browsing information is used to instruct the target user to browse the browsing history of automotive applications, the call information is used to instruct the target user to communicate with automotive sales personnel, and the access information is used to instruct the target user to visit automotive sales and service stores.

[0007] Based on the first browsing information, the second browsing information, the call information, and the access information, multiple prediction models are established through the Long Short-Term Memory Network, and the prediction probability output by each prediction model is obtained.

[0008] Assign weights to each prediction model to obtain the model weights corresponding to each prediction model;

[0009] Based on the predicted probability output by each prediction model and the corresponding model weight, the probability of the target user's car purchase intention is obtained.

[0010] In one possible design, based on the first browsing information, the second browsing information, call information, and access information, multiple prediction models are built using a Long Short-Term Memory (LSTM) network, and the prediction probabilities output by each prediction model are obtained, including:

[0011] Based on the first browsing information, the second browsing information, the call information, and the access information, multiple feature matrices are obtained;

[0012] Each feature matrix is ​​input into a long short-term memory network to obtain the predicted probability corresponding to each feature matrix.

[0013] In one possible design, based on the first browsing information, the second browsing information, the call information, and the access information, multiple feature matrices are obtained, including:

[0014] Based on the initial browsing information, multiple first feature matrices are obtained;

[0015] Based on the second browsing information and call information, multiple second feature matrices are obtained;

[0016] Based on the access information, multiple third feature matrices are obtained.

[0017] In one possible design, the first browsing information includes multiple first dates, as well as the first browsing duration, first browsing count, and first browsing traffic corresponding to each first date;

[0018] The second browsing information includes multiple second dates, as well as the second browsing duration, second browsing count, and second browsing traffic corresponding to each second date;

[0019] The call information includes multiple secondary dates, as well as the call duration and number of calls corresponding to each secondary date;

[0020] The visit information includes multiple third dates, as well as the duration and number of visits for each third date;

[0021] Based on the initial browsing information, several first feature matrices are obtained, including:

[0022] Based on multiple first dates, obtain the first one-hot matrix corresponding to each first date;

[0023] Based on multiple first dates, and the first browsing duration, first browsing count, and first browsing traffic corresponding to each first date, the first row matrix corresponding to each first unique hot matrix is ​​obtained;

[0024] Based on multiple first one-hot matrices and the first row matrix corresponding to each first one-hot matrix, the first feature matrix corresponding to each first one-hot matrix is ​​obtained;

[0025] Based on the second browsing information and call information, multiple second feature matrices are obtained, including:

[0026] Based on multiple second dates, obtain the second one-hot matrix corresponding to each second date;

[0027] Based on multiple second dates, and the second browsing duration, second browsing count, second browsing traffic, call duration, and call count corresponding to each second date, the second row matrix corresponding to each second one-hot matrix is ​​obtained;

[0028] Based on multiple second one-hot matrices and the second row matrix corresponding to each second one-hot matrix, the second feature matrix corresponding to each second one-hot matrix is ​​obtained;

[0029] Based on the access information, several third feature matrices were obtained, including:

[0030] Based on multiple third dates, obtain the third unique hot matrix corresponding to each third date;

[0031] Based on multiple third dates, and the corresponding store visit duration and number of visits for each third date, the third row matrix corresponding to each third unique hot matrix is ​​obtained;

[0032] Based on multiple third-independent-hot matrices and the third row matrix corresponding to each third-independent-hot matrix, the third characteristic matrix corresponding to each third-independent-hot matrix is ​​obtained.

[0033] In one possible design, each feature matrix includes a daily correlation feature matrix and a weekly correlation feature matrix;

[0034] Each feature matrix is ​​input into the Long Short-Term Memory network to obtain the predicted probability corresponding to each feature matrix, including:

[0035] Each daily correlation feature matrix and the corresponding weekly correlation feature matrix are input into the Long Short-Term Memory network to obtain the prediction probability corresponding to each daily correlation feature matrix.

[0036] In one possible design, weights are assigned to each prediction model to obtain the model weights corresponding to each prediction model, including:

[0037] The objective weighting method is used to assign weights to each prediction model, thus obtaining the model weights corresponding to each prediction model.

[0038] In one possible design, after obtaining the target user's car purchase intention probability based on the predicted probability output by each prediction model and the model weights corresponding to each prediction model, the method further includes:

[0039] Based on the probability of car purchase intention, a target marketing plan is determined from a variety of pre-stored marketing plans, and then pushed to the target users.

[0040] Secondly, this application provides a user's car purchase intention prediction device, comprising:

[0041] The user information acquisition module is used to acquire the target user's first browsing information, second browsing information, call information, and access information. The first browsing information is used to instruct the target user to browse the browsing history of automotive vertical media, the second browsing information is used to instruct the target user to browse the browsing history of automotive applications, the call information is used to instruct the target user to communicate with automotive sales personnel, and the access information is used to instruct the target user to visit automotive sales and service stores.

[0042] The prediction probability determination module is used to establish multiple prediction models through a long short-term memory network based on the first browsing information, the second browsing information, call information, and access information, and to obtain the prediction probability output by each prediction model.

[0043] The model weight determination module is used to assign weights to each prediction model and obtain the model weights corresponding to each prediction model.

[0044] The car purchase intention determination module is used to obtain the car purchase intention probability of the target user based on the predicted probability output by each prediction model and the model weight corresponding to each prediction model.

[0045] Thirdly, this application provides an electronic device, including: a processor, and a memory communicatively connected to the processor;

[0046] The memory stores the instructions that the computer executes;

[0047] When the processor executes computer execution instructions stored in memory, it is used to implement a user car purchase intention prediction method according to the first aspect of the invention.

[0048] Fourthly, this application provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement a user car purchase intention prediction method according to the first aspect of the invention.

[0049] This application provides a method, apparatus, device, and storage medium for predicting user car purchase intention. It acquires first browsing information, second browsing information, call information, and access information of a target user. The first browsing information indicates the target user's browsing history on automotive media; the second browsing information indicates the target user's browsing history on automotive applications; the call information indicates the target user's call history with car sales personnel; and the access information indicates the target user's visit to a car dealership. Based on this information, multiple prediction models are established using a Long Short-Term Memory (LSTM) network, and the prediction probability output by each model is obtained. Weights are then assigned to each prediction model to obtain its corresponding model weight. Finally, based on the prediction probability output by each model and its corresponding model weight, the target user's car purchase intention probability is obtained. The following technical effects were achieved: By incorporating call information between car sales representatives and target users, as well as visit information from target users' visits to car dealerships, the acquired information became more comprehensive, solving the problem of incomplete information leading to an inability to accurately determine the probability of a target user's car purchase intention. Furthermore, by inputting the target user's first browsing, second browsing, call, and visit information into an existing long short-term memory network model, multiple prediction models for different behavioral stages were constructed for each target user. This addressed the problem of a limited number of prediction models and the lack of model construction tailored to different behavioral stages, thus preventing further improvement in the accuracy of the prediction models. Finally, by assigning weights to each prediction model based on the car purchase information data from the prediction models at different behavioral stages, and by weighted summing of the prediction probabilities of the prediction models for different behavioral stages, the probability of each target user's car purchase intention was derived, resolving the issue of not being able to accurately determine the probability of a target user's car purchase intention. Attached Figure Description

[0050] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0051] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0052] Figure 1 This is a schematic diagram of the system architecture of the user car purchase intention prediction method provided in the embodiments of this application;

[0053] Figure 2 This is a framework diagram of the application scenario for the user car purchase intention prediction method provided in the embodiments of this application;

[0054] Figure 3 A flowchart illustrating the user car purchase intention prediction method provided in this application embodiment. Figure 1 ;

[0055] Figure 4 A flowchart illustrating the user car purchase intention prediction method provided in this application embodiment. Figure 2 ;

[0056] Figure 5 A flowchart illustrating the user car purchase intention prediction method provided in this application embodiment. Figure 3 ;

[0057] Figure 6 A schematic diagram of the third feature matrix provided in an embodiment of this application;

[0058] Figure 7 A flowchart illustrating the user car purchase intention prediction method provided in this application embodiment. Figure 4 ;

[0059] Figure 8 A flowchart illustrating the user car purchase intention prediction method provided in this application embodiment. Figure 5 ;

[0060] Figure 9 A flowchart illustrating the user car purchase intention prediction method provided in this application embodiment. Figure 6 ;

[0061] Figure 10 A flowchart illustrating the user car purchase intention prediction method provided in this application embodiment. Figure 7 ;

[0062] Figure 11 This is a schematic diagram of the user car purchase intention prediction device provided in an embodiment of this application;

[0063] Figure 12 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application.

[0064] Figure label:

[0065] 110 - Communication terminal; 120 - Backend server;

[0066] 210 - Smartphone; 220 - Data processing server;

[0067] 300 - User car purchase intention prediction device; 310 - User information acquisition module; 320 - Prediction probability determination module; 330 - Model weight determination module; 340 - Car purchase intention determination module;

[0068] 400 - Electronic device; 410 - Receiver; 420 - Transmitter; 430 - Processor; 440 - Memory; 450 - Bus. Detailed Implementation

[0069] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0070] In the embodiments of this application, the terms "first" and "second" are used to distinguish identical or similar items with substantially the same function and effect. Those skilled in the art will understand that the terms "first" and "second" do not limit the quantity or execution order, nor do they necessarily imply difference. It should be noted that in the embodiments of this application, words such as "exemplary" or "for example" are used to indicate examples, illustrations, or explanations. Any embodiment or design scheme described as "exemplary" or "for example" in this application should not be construed as being more preferred or advantageous than other embodiments or design schemes. Specifically, the use of words such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner. In the embodiments of this application, "at least one" refers to one or more, and "more than one" refers to two or more.

[0071] It should be noted that the phrase "at...time" in the embodiments of this application can refer to the instant at which a certain situation occurs, or to a period of time after the occurrence of a certain situation; the embodiments of this application do not specifically limit this. Furthermore, the user car purchase intention prediction method provided in the embodiments of this application is merely an example; user car purchase intention prediction methods may also include more or less content.

[0072] To facilitate a clear description of the technical solutions in the embodiments of this application, some terms and technologies involved in the embodiments of this application will be briefly introduced below:

[0073] Call information: refers to the objects transmitted and processed by the communication system.

[0074] Application (APP): A set of deployable software entities, including one or a set of containers or processes.

[0075] The Criteria Importance Through Intercrieria Correlation (CRITIC) method is a method that comprehensively measures the objective weight of indicators based on the comparative strength and conflict between indicators.

[0076] Long Short-Term Memory (LSTM) networks are a type of temporal recurrent neural network suitable for processing and predicting important events with relatively long intervals and delays in time series. LSTM has already found a variety of applications in the technology field. LSTM-based systems can perform tasks such as language learning and translation, robot control, image analysis, document summarization, speech and image recognition, handwriting recognition, chatbot control, disease prediction and click-through rate prediction, and music synthesis.

[0077] Recently, in the automotive industry, car sales have been growing slowly. Faced with the overall slow growth of the car market, competition in the car sales industry has become increasingly fierce. In the existing technology, data analysis is mainly based on users' browsing logs of car purchase vertical media and applications. A model of online browsing and car purchase intention is established through traditional BP neural networks to ultimately discover potential car buyers.

[0078] However, existing online car purchase intention models only incorporate behavioral characteristics from online browsing of automotive media and automotive applications, resulting in low model accuracy and an inability to accurately predict users' car purchase intentions based on these models.

[0079] Based on this, embodiments of this application provide a method, apparatus, device, and storage medium for predicting user car purchase intentions, which can be used in the field of big data and aims to solve the above-mentioned technical problems of the prior art.

[0080] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.

[0081] Figure 1 This is a schematic diagram of the system architecture for the user car purchase intention prediction method provided in this application embodiment. It should be noted that... Figure 1 The examples shown are merely examples of system architectures that can be applied to the embodiments of this application, in order to help those skilled in the art understand the technical content of this application, but do not mean that the embodiments of this application cannot be used in other devices, systems, environments or scenarios.

[0082] like Figure 1As shown, the system architecture of this method includes a communication terminal 110 and a backend server 120. The communication terminal 110 and the backend server 120 can be connected via wired or wireless communication. The wired communication connection can be made via Universal Serial Bus (USB), Fiber Optic, or High Definition Multimedia Interface (HDMI), etc. The wireless communication connection can be made via 5th Generation Mobile Communication Technology (5G), WiFi, or Bluetooth, etc.

[0083] In this embodiment, the communication terminal 110 can be various types of electronic devices, including but not limited to smartphones, tablets, and personal computers. Users can browse automotive media and automotive applications through the communication terminal 110, talk to car sales personnel through the communication terminal 110, and go to an Automobile Sales Service Shop 4S through the communication terminal 110.

[0084] The backend server 120 can be a backend management server that provides various service support (for example only). The backend server 120 can obtain terminal information such as application browsing information, call information, and location signaling information of the communication terminal 110, analyze and process the received terminal information, and feed back the results obtained from processing this terminal information (such as web pages, information, or data obtained or generated according to user requests) to the terminal device.

[0085] Figure 2 This is a framework diagram illustrating the application scenario of the user car purchase intention prediction method provided in the embodiments of this application. For example... Figure 2 As shown, the data processing server 220 (such as the backend server 120) is communicatively connected to multiple smartphones 210 (such as communication terminals 110). Users browse applications, make calls, and navigate through their respective smartphones 210. The data processing server 220 obtains terminal information such as application browsing information, call information, and location signaling information from each smartphone 210, and obtains the probability of the target user's car purchase intention based on this terminal information.

[0086] Figure 3 A flowchart illustrating the user car purchase intention prediction method provided in this application embodiment. Figure 1The backend server can process car purchase information data from multiple target users simultaneously. This data includes: first browsing information, second browsing information, call information, and access information. The method for processing the car purchase information data for each target user is the same; this embodiment describes in detail the method for processing the car purchase information data for one target user. Figure 3 As shown, the method includes:

[0087] S101. Obtain the target user's first browsing information, second browsing information, call information, and access information;

[0088] Specifically, the backend server identifies the target user to be tracked and obtains the target user's car purchase information data. The first browsing information indicates the target user's browsing history on automotive media platforms, including parameters such as car brands, models, prices, and energy consumption that the target user has viewed through these platforms. The second browsing information indicates the target user's browsing history on automotive applications, including the categories and names of the applications viewed. Call information indicates the target user's call records with car sales personnel, including communication activities between the target user and a specific car brand's sales staff. Access information indicates the target user's visit records to car dealerships and service centers, including information on the target user's visits to a specific brand's 4S store, mined from interest-based map data and signaling data.

[0089] S102. Based on the first browsing information, the second browsing information, the call information, and the access information, establish multiple prediction models through a long short-term memory network, and obtain the prediction probability output by each prediction model.

[0090] Specifically, the backend server inputs the first browsing information, second browsing information, call information, and access information into the existing LSTM model, constructs multiple prediction models for different behavioral stages for each target user, and finally outputs the car purchase prediction probability of each stage prediction model.

[0091] S103. Assign weights to each prediction model to obtain the model weights corresponding to each prediction model;

[0092] Specifically, the backend server assigns weights to each prediction model based on the car purchase information data in the prediction models of the target user at different behavioral stages. The weighting methods that can be used include, but are not limited to, equal weighting, machine learning methods, and swing weighting.

[0093] S104. Based on the predicted probability output by each prediction model and the model weight corresponding to each prediction model, obtain the target user's car purchase intention probability.

[0094] Specifically, the backend server weights and sums the predicted probabilities of the prediction models for different behavioral stages of the target user to obtain the car purchase intention probability for each target user.

[0095] This embodiment provides a method for predicting a user's car purchase intention. It acquires a target user's first browsing information, second browsing information, call information, and access information. The first browsing information indicates the target user's browsing history on automotive media; the second browsing information indicates the target user's browsing history on automotive applications; the call information indicates the target user's call history with car sales personnel; and the access information indicates the target user's visit to a car dealership. Based on this information, multiple prediction models are established using a Long Short-Term Memory (LSTM) network, and the prediction probability output by each model is obtained. Weights are then assigned to each prediction model to obtain its corresponding model weight. Finally, based on the prediction probability output by each model and its corresponding model weight, the target user's car purchase intention probability is obtained. The following technical effects were achieved: By incorporating call information between car sales representatives and target users, as well as visit information from target users' visits to car dealerships, the acquired information became more comprehensive, solving the problem of incomplete information leading to an inability to accurately determine the probability of a target user's car purchase intention. Furthermore, by inputting the target user's first browsing, second browsing, call, and visit information into an existing long short-term memory network model, multiple prediction models for different behavioral stages were constructed for each target user. This addressed the problem of a limited number of prediction models and the lack of model construction tailored to different behavioral stages, thus preventing further improvement in the accuracy of the prediction models. Finally, by assigning weights to each prediction model based on the car purchase information data from the prediction models at different behavioral stages, and by weighted summing of the prediction probabilities of the prediction models for different behavioral stages, the probability of each target user's car purchase intention was derived, resolving the issue of not being able to accurately determine the probability of a target user's car purchase intention.

[0096] Figure 4 A flowchart illustrating the user car purchase intention prediction method provided in this application embodiment. Figure 2 This embodiment is in Figure 3 Based on the implementation examples, this paper details how to build multiple prediction models using LSTM based on first browsing information, second browsing information, call information, and access information. For example... Figure 4 As shown, the method includes:

[0097] S201. Obtain the target user's first browsing information, second browsing information, call information, and access information;

[0098] S201 is similar to S101, and will not be described again in this embodiment.

[0099] S202. Based on the first browsing information, obtain multiple first feature matrices;

[0100] Specifically, the first browsing information includes multiple first dates, and for each first date, the corresponding first browsing duration, first browsing frequency, and first browsing traffic. More specifically, if the marketed car brand or model is designated as A, then the vertical media browsing behavior information in the target user's first browsing information includes features such as the number of days, duration, frequency, and traffic of browsing A and its competitors. This information can characterize the characteristics of the target user's preferred car brand or model; therefore, the prediction model corresponding to the first feature matrix is ​​a preference model.

[0101] S203. Based on the second browsing information and call information, obtain multiple second feature matrices;

[0102] Specifically, the second browsing information includes multiple second dates, and the corresponding second browsing duration, number of second browsing visits, and second browsing traffic for each second date. Similarly, if the marketed car brand or model is set as A, then the target user's APP browsing behavior information is further subdivided into characteristics such as browsing days, browsing duration, number of visits, and browsing traffic for car transaction and car information apps. Call information includes multiple second dates, and the corresponding call duration and number of calls for each second date. Again, if the marketed car brand or model is set as A, then the target user's call behavior information with car sales staff of brand / model A is further subdivided into characteristics such as call days, call duration, and number of calls. In generating this information, target users typically leave information (contact information, preferred car models, etc.) in car transaction or car information apps (including lead generation pages), and also leave this information during calls with car sales staff. Therefore, the model corresponding to the second feature matrix constructed using the second browsing information and call information is called the lead generation model.

[0103] S204. Based on the access information, obtain multiple third feature matrices;

[0104] Specifically, the visit information includes multiple third dates, as well as the duration and number of visits for each third date. Similarly, if the marketed car brand or model is designated as A, the target user's behavior information regarding visits to A's 4S stores is further subdivided into features such as the number of days visited, the duration of visits, and the number of visits. This information covers the characteristics of target users' visits to car 4S stores; therefore, the model corresponding to the third feature matrix constructed using the visit information is called the store visit model.

[0105] S205. Input each feature matrix into the long short-term memory network to obtain the prediction probability corresponding to each feature matrix;

[0106] Specifically, the backend server inputs each feature matrix into the LSTM to build multiple prediction models and obtains the prediction probability output by each prediction model.

[0107] S206. Assign weights to each prediction model to obtain the model weights corresponding to each prediction model;

[0108] S207. Based on the predicted probability output by each prediction model and the model weight corresponding to each prediction model, obtain the target user's car purchase intention probability.

[0109] S206-S207 are similar to S103-S104, and will not be described again in this embodiment.

[0110] Figure 5 A flowchart illustrating the user car purchase intention prediction method provided in this application embodiment. Figure 3 In one possible example, such as Figure 5 As shown, in this embodiment... Figure 4 Based on the embodiments, a detailed explanation is provided on obtaining multiple feature matrices based on the first browsing information, the second browsing information, call information, and access information. S204. Based on the access information, multiple third feature matrices are obtained, including:

[0111] S301. Based on multiple third dates, obtain the third one-hot matrix corresponding to each third date;

[0112] Specifically, the backend server constructs a one-hot matrix based on the characteristic information of the target user's number of days spent in the store. For example... Figure 6 In the schematic diagram of the feature matrix shown, the one-hot matrix includes:

[0113] The twelve-dimensional month matrix M converts month information into a 12-dimensional one-hot vector, with the element corresponding to the month type being 1 and the rest being 0;

[0114] The seven-dimensional weekday matrix D converts weekday information into a seven-dimensional one-hot vector, with elements corresponding to the weekday type being 1 and the rest being 0;

[0115] And a one-dimensional holiday matrix H, which converts holiday information into a one-dimensional one-hot vector, where 1 indicates that the moment is a holiday and zero indicates otherwise.

[0116] S302. Based on multiple third dates, and the store visit duration and number of visits corresponding to each third date, obtain the third row matrix corresponding to each third unique hot matrix;

[0117] Specifically, because LSTM and other machine learning methods are sensitive to data scale, the backend server first normalizes the feature information of multiple visit days, visit duration, and visit frequency. The normalization method used is MinMax normalization, which transforms the data value range to [0, 1]. The calculation formula is as follows:

[0118]

[0119] Where, x norm This is the standardized data, where x is the original data. min and x max These are the minimum and maximum values ​​of the original data, respectively.

[0120] A three-dimensional matrix F is constructed using the normalized features of the number of days spent at the store, the duration of the visit, and the number of visits.

[0121] S303. Based on multiple third-independent-hot matrices and the third row matrix corresponding to each third-independent-hot matrix, obtain the third characteristic matrix corresponding to each third-independent-hot matrix.

[0122] Specifically, in the in-store model, the third one-hot matrix includes a twelve-dimensional month matrix M, a seven-dimensional weekday matrix D, and a one-dimensional holiday matrix H, which together form a twenty-dimensional one-hot matrix. Adding the three-dimensional third row matrix completes a twenty-three-dimensional third feature matrix. Of course, this embodiment only illustrates a specific scheme for constructing the feature matrix. Those skilled in the art can easily conceive of modifying or adding certain information to construct other forms of feature matrices, which are still within the scope of this invention.

[0123] Figure 6 This is a schematic diagram of the third feature matrix provided in an embodiment of this application. Figure 6 As shown, taking the in-store visit model as an example, to predict the probability of a target user visiting the store on a certain day, such as February 8, 2023 (Wednesday), if the daily step size K1 is set to 7, a daily relevant feature matrix can be constructed based on the target user's access information for the previous seven days (February 1-7, 2023); if the weekly step size K2 is set to 4, a weekly relevant feature matrix can be constructed based on the target user's access information for each Wednesday in the previous four weeks (January 11, January 18, January 25, and February 1, 2023).

[0124] Figure 7 A flowchart illustrating the user car purchase intention prediction method provided in this application embodiment. Figure 4 In one possible example, such as Figure 7 As shown, in this embodiment... Figure 4 Based on the embodiments, a detailed explanation is provided on obtaining multiple feature matrices based on first browsing information, second browsing information, call information, and access information. The second browsing information includes multiple second dates, and the second browsing duration, second browsing count, and second browsing traffic corresponding to each second date; the call information includes multiple second dates, and the call duration and call count corresponding to each second date. S202, Based on the first browsing information, multiple first feature matrices are obtained, including:

[0125] S401. Based on multiple first dates, obtain the first one-hot matrix corresponding to each first date;

[0126] S402. Based on multiple first dates, and the first browsing duration, first browsing count, and first browsing traffic corresponding to each first date, obtain the first row matrix corresponding to each first unique hot matrix;

[0127] S403. Based on multiple first one-hot matrices and the first row matrix corresponding to each first one-hot matrix, obtain the first feature matrix corresponding to each first one-hot matrix;

[0128] S401-S403 are similar to S301-S303, and will not be described again in this embodiment.

[0129] Figure 8 A flowchart illustrating the user car purchase intention prediction method provided in this application embodiment. Figure 5 In one possible example, such as Figure 8 As shown, in this embodiment... Figure 4 Based on the embodiments, a detailed explanation is provided on obtaining multiple feature matrices based on the first browsing information, second browsing information, call information, and access information. The access information includes multiple third dates, and the corresponding store visit duration and number of visits for each third date. S203, multiple second feature matrices are obtained based on the second browsing information and call information, including;

[0130] S501. Based on multiple second dates, obtain the second one-hot matrix corresponding to each second date;

[0131] S502. Based on multiple second dates, and the second browsing duration, second browsing count, second browsing traffic, call duration, and call count corresponding to each second date, obtain the second row matrix corresponding to each second one-hot matrix;

[0132] S503. Based on multiple second one-hot matrices and the second row matrix corresponding to each second one-hot matrix, obtain the second feature matrix corresponding to each second one-hot matrix;

[0133] S501-S503 are similar to S301-S303, and will not be described again in this embodiment.

[0134] Figure 9 A flowchart illustrating the user car purchase intention prediction method provided in this application embodiment. Figure 6 .like Figure 9 As shown, in this embodiment... Figure 4 Based on the embodiments, this paper details how to input each feature matrix into a Long Short-Term Memory (LSTM) network to obtain the prediction probability corresponding to each feature matrix. Each feature matrix includes a daily correlation feature matrix and a weekly correlation feature matrix. The method includes:

[0135] S601. Obtain the target user's first browsing information, second browsing information, call information, and access information;

[0136] S602. Based on the first browsing information, obtain multiple first feature matrices;

[0137] S603. Based on the second browsing information and call information, obtain multiple second feature matrices;

[0138] S604. Based on the access information, obtain multiple third feature matrices;

[0139] S601-S604 are similar to S201-S204, and will not be described again in this embodiment.

[0140] S605. Input each daily correlation feature matrix and the corresponding weekly correlation feature matrix into the long short-term memory network to obtain the prediction probability corresponding to each daily correlation feature matrix.

[0141] Specifically, each daily correlation feature matrix and its corresponding weekly correlation feature matrix are simultaneously input into an LSTM. The outputs of the two parallel LSTMs are connected to an output layer, and the output of the output layer is the car purchase probability value to be predicted.

[0142] During the training of LSTM, an automated model framework and adaptive parameter adjustment techniques can be used. The network model framework can be encapsulated, and hyperparameters such as the LSTM step size, number of training epochs, and learning rate can be set as input variables in the form of value ranges to perform automated training and determine the optimal combination of hyperparameters.

[0143] S606. Assign weights to each prediction model to obtain the model weights corresponding to each prediction model;

[0144] S607. Based on the predicted probability output by each prediction model and the model weight corresponding to each prediction model, obtain the target user's car purchase intention probability.

[0145] S606-S607 are similar to S103-S104, and will not be described again in this embodiment.

[0146] Figure 10 A flowchart illustrating the user car purchase intention prediction method provided in this application embodiment. Figure 7 .like Figure 10 As shown, in this embodiment... Figure 3 Based on the examples, this paper provides a detailed explanation of how to assign weights to each prediction model to obtain the corresponding model weights. The method includes:

[0147] S701, Obtain the target user's first browsing information, second browsing information, call information, and access information;

[0148] S702. Based on the first browsing information, the second browsing information, the call information, and the access information, establish multiple prediction models through a long short-term memory network, and obtain the prediction probability output by each prediction model.

[0149] S701-S702 are similar to S101-S102, and will not be described again in this embodiment.

[0150] S703. Assign weights to each prediction model using the objective weighting method to obtain the model weights corresponding to each prediction model.

[0151] Specifically, to more accurately calculate the final purchase intention probability of the target user, the backend server assigns weights to each prediction model to obtain the model weights for each model. Based on the predicted probability output by each prediction model and its corresponding model weight, the purchase intention probability of the target user is obtained. Here, the objective weighting method (Criteria Importance Though Intercrieria Correlation, CRITIC method) is used to assign weights to each prediction model. The CRITIC method is a method that comprehensively measures the objective weights of indicators based on the comparative strength and conflict between indicators.

[0152] S704. Based on the predicted probability output by each prediction model and the model weight corresponding to each prediction model, obtain the target user's car purchase intention probability.

[0153] S704 is similar to S104, and will not be described again in this embodiment.

[0154] S705: Based on the probability of car purchase intention, determine the target marketing plan from a variety of pre-stored marketing plans, and push the target marketing plan to the target users;

[0155] Specifically, by deeply analyzing the target users' automotive media browsing logs and operator internet data, the backend server can uncover parameters such as the target users' preferred car brands, models, price ranges, and energy types. This allows the marketing plan to incorporate a matching process between the marketing brand and model parameters and the user's car purchase preferences, selecting target users who are interested in the marketing brand and model, and then distributing the marketing plan based on the predicted probability of purchase intention.

[0156] This embodiment provides a method for predicting a user's car purchase intention. It acquires a target user's first browsing information, second browsing information, call information, and access information. The first browsing information indicates the target user's browsing history on automotive media; the second browsing information indicates the target user's browsing history on automotive applications; the call information indicates the target user's call history with car sales personnel; and the access information indicates the target user's visit to a car dealership. Based on this information, multiple prediction models are established using a Long Short-Term Memory (LSTM) network, and the prediction probability output by each model is obtained. Weights are then assigned to each prediction model to obtain its corresponding model weight. Finally, based on the prediction probability output by each model and its corresponding model weight, the target user's car purchase intention probability is obtained. The following technical effects were achieved: By incorporating call information between car sales representatives and target users, as well as visit information from target users' visits to car dealerships, the acquired information became more comprehensive, solving the problem of incomplete information leading to an inability to accurately determine the probability of a target user's purchase intention. By inputting the target user's first browsing information, second browsing information, call information, and visit information into an existing long short-term memory network model, multiple prediction models for different behavioral stages were constructed for each target user. This addressed the problem of a limited number of prediction models and the lack of model construction based on the different behavioral stages of the target user, thus resolving the issue of limited accuracy in further improving the prediction models. Furthermore, by assigning weights to each prediction model based on the car purchase information data from the prediction models at different behavioral stages, and considering the different stages of the target user's behavior, the system effectively addressed the issue of limited accuracy in the prediction models. The prediction probabilities of the behavioral stage prediction model are weighted and summed to obtain the purchase intention probability of each target user, solving the problem of not being able to accurately obtain the purchase intention probability of target users. By constructing a daily relevant feature matrix and a corresponding weekly relevant feature matrix for each prediction model, and simultaneously inputting the daily and weekly relevant feature matrices into an LSTM to calculate the purchase intention prediction probability of the target user, the final prediction probability after averaging the probabilities is output at the output layer of the LSTM. This solves the problem of how to simultaneously construct two matrices and input them into the LSTM to calculate the purchase intention probability of the target user, thereby improving the accuracy of each prediction model. Based on the purchase intention probability, a target marketing plan is determined from a variety of pre-stored marketing plans and pushed to the target user, solving the problem of how to reduce marketing costs and improve marketing accuracy.

[0157] In this embodiment of the invention, electronic devices or main control devices can be divided into functional modules according to the above method examples. For example, each function can be divided into its own functional modules, or two or more functions can be integrated into one processing unit. The integrated unit can be implemented in hardware or as a software functional module. It should be noted that the module division in this embodiment of the invention is illustrative and only represents one logical functional division; other division methods may be used in actual implementation.

[0158] Figure 11 This is a schematic diagram of the user car purchase intention prediction device provided in an embodiment of this application. Figure 11 As shown, the user's car purchase intention prediction device 300 includes:

[0159] User information acquisition module 310, prediction probability determination module 320, model weight determination module 330, and car purchase intention determination module 340;

[0160] The user information acquisition module 310 is used to acquire the target user's first browsing information, second browsing information, call information and access information. The first browsing information is used to instruct the target user to browse the browsing history of automotive vertical media, the second browsing information is used to instruct the target user to browse the browsing history of automotive applications, the call information is used to instruct the target user to communicate with automotive sales personnel, and the access information is used to instruct the target user to visit automotive sales and service stores.

[0161] The prediction probability determination module 320 is used to establish multiple prediction models through a long short-term memory network based on the first browsing information, the second browsing information, the call information, and the access information, and to obtain the prediction probability output by each prediction model.

[0162] The model weight determination module 330 is used to assign weights to each prediction model and obtain the model weights corresponding to each prediction model.

[0163] The car purchase intention determination module 340 is used to obtain the car purchase intention probability of the target user based on the prediction probability output by each prediction model and the model weight corresponding to each prediction model.

[0164] In one possible design, the prediction probability determination module 320 includes: a feature matrix construction module and a feature matrix input module;

[0165] The feature matrix construction module is used to obtain multiple feature matrices based on the first browsing information, the second browsing information, the call information, and the access information;

[0166] The feature matrix input module is used to input each feature matrix into the long short-term memory network to obtain the prediction probability corresponding to each feature matrix.

[0167] In one possible design, the feature matrix construction module includes: a first feature matrix construction module, a second feature matrix construction module, and a third feature matrix construction module;

[0168] The first feature matrix construction module is used to obtain multiple first feature matrices based on the first browsing information;

[0169] The second feature matrix construction module is used to obtain multiple second feature matrices based on the second browsing information and call information;

[0170] The third feature matrix construction module is used to obtain multiple third feature matrices based on the access information.

[0171] In one possible design, the user information acquisition module 310 acquires first browsing information including multiple first dates, and the first browsing duration, first browsing count and first browsing traffic corresponding to each first date;

[0172] The second browsing information includes multiple second dates, as well as the second browsing duration, second browsing count, and second browsing traffic corresponding to each second date;

[0173] The call information includes multiple secondary dates, as well as the call duration and number of calls corresponding to each secondary date;

[0174] The visit information includes multiple third dates, as well as the duration and number of visits for each third date;

[0175] The first feature matrix construction module includes: a first one-hot matrix construction module and a first row matrix construction module;

[0176] The first one-hot matrix construction module is used to obtain the first one-hot matrix corresponding to each first date based on multiple first dates;

[0177] The first row matrix construction module is used to obtain the first row matrix corresponding to each first unique hot matrix based on multiple first dates and the first browsing duration, first browsing count and first browsing traffic corresponding to each first date;

[0178] The first feature matrix construction module is also used to obtain the first feature matrix corresponding to each first one-hot matrix based on multiple first one-hot matrices and the first row matrix corresponding to each first one-hot matrix;

[0179] The second feature matrix construction module includes a second one-hot matrix construction module and a second row matrix construction module;

[0180] The second one-hot matrix construction module obtains the second one-hot matrix corresponding to each second date based on multiple second dates;

[0181] The second row matrix construction module is used to obtain the second row matrix corresponding to each second one-hot matrix based on multiple second dates and the second browsing duration, second browsing count, second browsing traffic, call duration and call count corresponding to each second date.

[0182] The second feature matrix construction module is also used to obtain the second feature matrix corresponding to each second one-hot matrix based on multiple second one-hot matrices and the second row matrix corresponding to each second one-hot matrix;

[0183] The third feature matrix construction module includes a third one-hot matrix construction module and a third row matrix construction module;

[0184] The third one-hot matrix construction module is used to obtain the third one-hot matrix corresponding to each third date based on multiple third dates;

[0185] The third row matrix construction module is used to obtain the third row matrix corresponding to each third unique hot matrix based on multiple third dates, as well as the store visit duration and number of store visits corresponding to each third date.

[0186] The third feature matrix construction module is also used to obtain the third feature matrix corresponding to each third one-hot matrix based on multiple third one-hot matrices and the third row matrix corresponding to each third one-hot matrix.

[0187] In one possible design, the feature matrix input module is also used to input each daily correlation feature matrix and the weekly correlation feature matrix corresponding to each daily correlation feature matrix into the long short-term memory network to obtain the prediction probability corresponding to each daily correlation feature matrix.

[0188] In one possible design, the model weight determination module 330 is also used to assign weights to each prediction model using an objective weighting method to obtain the model weights corresponding to each prediction model.

[0189] In one possible design, the user's car purchase intention prediction device 300 is also used to determine a target marketing plan from a variety of pre-stored marketing plans based on the probability of car purchase intention, and push the target marketing plan to the target user.

[0190] In the specific implementation of the aforementioned method for predicting user car purchase intention, each module can be implemented as a processor. The processor can execute computer execution instructions stored in the memory, thereby enabling the processor to execute the aforementioned method for predicting user car purchase intention.

[0191] Figure 12 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Figure 12As shown, this application provides an electronic device 400, which includes at least one processor 430 and a memory 440. The electronic device 400 also includes a receiver 410 and a transmitter 420. The processor 430, memory 440, receiver 410, and transmitter 420 are connected via a bus 450. Specifically:

[0192] Receiver 410 is used to receive instructions and data;

[0193] Transmitter 420 is used to send commands and data;

[0194] Memory 440 is used to store instructions executed by the computer;

[0195] The processor 430 is configured to execute computer execution instructions stored in the memory 440 to implement the various steps of the user car purchase intention prediction method in the above embodiments. For details, please refer to the relevant description in the embodiments of the user car purchase intention prediction method.

[0196] Alternatively, the memory 440 can be either standalone or integrated with the processor 430.

[0197] When the memory 440 is set up independently, the electronic device also includes a bus for connecting the memory 440 and the processor 430.

[0198] In the above embodiments, it should be understood that the processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this invention can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor.

[0199] The memory may include high-speed RAM, and may also include non-volatile storage (NVM), such as at least one disk storage.

[0200] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of illustration, the buses shown in the accompanying drawings are not limited to a single bus or a single type of bus.

[0201] The above description of the functions implemented by electronic devices and main control devices has introduced the solutions provided by the embodiments of the present invention. It is understood that, in order to implement the above functions, the electronic device or main control device includes hardware structures and / or software modules corresponding to the execution of each function. By combining the units and algorithm steps of the various examples described in the embodiments of the present invention, the embodiments of the present invention can be implemented in hardware or a combination of hardware and computer software. Whether a function is executed by hardware or by computer software driving hardware depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the technical solutions of the embodiments of the present invention.

[0202] This application also provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the above-mentioned method for predicting user car purchase intentions.

[0203] The aforementioned readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The readable storage medium can be any available medium accessible to a general-purpose or special-purpose computer.

[0204] An exemplary readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can reside in an Application Specific Integrated Circuit (ASIC). Alternatively, the processor and the readable storage medium can exist as discrete components in an electronic device or a host device.

[0205] The technical solutions of this application have been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it is readily understood by those skilled in the art that the scope of protection of this application is obviously not limited to these specific embodiments. The above embodiments are only used to illustrate the technical solutions of this application and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. These modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

Claims

1. A method for predicting user car purchase intention, characterized in that, include: The system acquires the target user's first browsing information, second browsing information, call information, and access information. The first browsing information is used to indicate the target user's browsing history of automotive vertical media, the second browsing information is used to indicate the target user's browsing history of automotive applications, the call information is used to indicate the call history between the target user and automotive sales personnel, and the access information is used to indicate the target user's access history of visiting automotive sales and service stores. Based on the first browsing information, the second browsing information, the call information, and the access information, multiple prediction models are established through a long short-term memory network, and the prediction probability output by each prediction model is obtained, including: obtaining multiple feature matrices based on the first browsing information, the second browsing information, the call information, and the access information; and inputting each feature matrix into the long short-term memory network to obtain the prediction probability corresponding to each feature matrix. Assign weights to each prediction model to obtain the model weights corresponding to each prediction model; Based on the predicted probability output by each prediction model and the model weight corresponding to each prediction model, the purchase intention probability of the target user is obtained. The first browsing information includes multiple first dates, and the first browsing duration, first browsing count, and first browsing traffic corresponding to each first date; The second browsing information includes multiple second dates, and for each second date, the second browsing duration, the second browsing count, and the second browsing traffic. The call information includes the plurality of second dates, and the call duration and number of calls corresponding to each second date; The access information includes multiple third dates, as well as the duration and number of visits corresponding to each third date; Based on the first browsing information, multiple first feature matrices are obtained, including: Based on the plurality of first dates, a first one-hot matrix corresponding to each first date is obtained; Based on the plurality of first dates, and the first browsing duration, first browsing count and first browsing traffic corresponding to each first date, a first row matrix corresponding to each first unique hot matrix is ​​obtained; Based on multiple first one-hot matrices and the first row matrix corresponding to each first one-hot matrix, the first feature matrix corresponding to each first one-hot matrix is ​​obtained; Based on the second browsing information and the call information, multiple second feature matrices are obtained, including: Based on the plurality of second dates, a second unique-hot matrix corresponding to each second date is obtained; Based on the plurality of second dates, and the second browsing duration, second browsing count, second browsing traffic, call duration and call count corresponding to each second date, a second row matrix corresponding to each second one-hot matrix is ​​obtained; Based on multiple second one-hot matrices and the second row matrix corresponding to each second one-hot matrix, the second feature matrix corresponding to each second one-hot matrix is ​​obtained; Based on the access information, multiple third feature matrices are obtained, including: Based on the multiple third dates, obtain the third unique hot matrix corresponding to each third date; Based on the multiple third dates, and the in-store duration and number of visits corresponding to each third date, a third row matrix corresponding to each third unique hot matrix is ​​obtained; Based on multiple third unique heat matrices and the third row matrix corresponding to each third unique heat matrix, the third feature matrix corresponding to each third unique heat matrix is ​​obtained.

2. The method according to claim 1, characterized in that, Each feature matrix includes a daily correlation feature matrix and a weekly correlation feature matrix; The step of inputting each feature matrix into a long short-term memory network to obtain the prediction probability corresponding to each feature matrix includes: Each daily correlation feature matrix and the corresponding weekly correlation feature matrix are input into a long short-term memory network to obtain the prediction probability corresponding to each daily correlation feature matrix.

3. The method according to claim 1, characterized in that, The step of assigning weights to each prediction model to obtain the model weights corresponding to each prediction model includes: Each prediction model is assigned a weight using an objective weighting method to obtain the model weight corresponding to each prediction model.

4. The method according to claim 1, characterized in that, After obtaining the purchase intention probability of the target user based on the predicted probability output by each prediction model and the model weight corresponding to each prediction model, the method further includes: Based on the probability of car purchase intention, a target marketing plan is determined from a variety of pre-stored marketing plans, and the target marketing plan is pushed to the target user.

5. An electronic device, characterized in that, include: A processor, and a memory communicatively connected to the processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory to implement a user car purchase intention prediction method as described in any one of claims 1 to 4.

6. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement a user car purchase intention prediction method as described in any one of claims 1 to 4.

Images