Product attribute decision method, device and electronic device based on artificial intelligence

Abstract

The present application provides a product attribute decision method, device, electronic device and computer-readable storage medium based on artificial intelligence; the method includes: calling a purchase probability prediction model based on user data to determine the purchase probability corresponding to the target product when the target product has different candidate attributes, wherein the different candidate attributes are included in a candidate attribute set; determining the estimated income based on the candidate attributes and the corresponding purchase probability; calling a violation probability prediction model based on the user's historical usage data to determine the violation probability of the user's illegal use of the target product, and determining the estimated loss based on the violation probability; determining the estimated expectation corresponding to the candidate attribute based on the estimated income and estimated loss; and taking the candidate attribute with the highest estimated expectation in the candidate attribute set as the attribute of the target product. Through the present application, the product attributes can be determined quickly and accurately.

Description

Technical Field

[0001] The present application relates to artificial intelligence technology, and in particular to an artificial intelligence-based product attribute decision method, device, electronic device and computer-readable storage medium. Background Art

[0002] Artificial Intelligence (AI) covers a wide range of fields and plays an increasingly important role. As a subset of AI technology, machine learning models have achieved breakthrough results in multiple application areas, especially in the financial field, such as determining product attributes based on user data, thereby providing direction for product updates and iterations.

[0003] However, the related art only determines product attributes based on single-source data. The product attributes determined by this method are often far from the actual needs, and multiple manual adjustments are required to determine the appropriate product attributes. Summary of the invention

[0004] The embodiments of the present application provide a product attribute decision method, device, electronic device and computer-readable storage medium based on artificial intelligence, which can quickly and accurately determine product attributes.

[0005] The technical solution of the embodiment of the present application is implemented as follows:

[0006] The present application embodiment provides a product attribute decision method based on artificial intelligence, including:

[0007] Calling a purchase probability prediction model based on user data to determine the purchase probabilities corresponding to the target product when the target product has different candidate attributes, wherein the different candidate attributes are included in the candidate attribute set;

[0008] Determining an estimated revenue based on the candidate attributes and the corresponding purchase probabilities;

[0009] Calling a violation probability prediction model based on the user's historical usage data to determine the violation probability of the user's illegal use of the target product, and determining an estimated loss based on the violation probability;

[0010] Determining an estimated expectation corresponding to the candidate attribute based on the estimated income and the estimated loss;

[0011] The candidate attribute with the highest estimated expectation in the candidate attribute set is used as the attribute of the target product.

[0012] The embodiment of the present application provides a product attribute decision device based on artificial intelligence, comprising:

[0013] A calling module, configured to call a purchase probability prediction model based on user data to determine the purchase probabilities corresponding to the target product when the target product has different candidate attributes, wherein the different candidate attributes are included in a candidate attribute set; and to call a violation probability prediction model based on the user's historical usage data to determine the violation probability of the user's illegal use of the target product, and determine an estimated loss based on the violation probability;

[0014] A determination module is used to determine the estimated revenue based on the candidate attributes and the corresponding purchase probability; and to determine the estimated expectation corresponding to the candidate attributes based on the estimated revenue and the estimated loss; and to select the candidate attribute with the highest estimated expectation in the candidate attribute set as the attribute of the target product.

[0015] In the above solution, the calling module is also used for:

[0016] Extracting user features and attribute features of the candidate attributes from the user data;

[0017] Performing feature conversion processing on the user features and the attribute features to obtain sparse intermediate features;

[0018] Classification processing is performed based on the intermediate features to obtain purchase probabilities corresponding to when the target product has different candidate attributes in the candidate attribute set.

[0019] In the above solution, the product attribute decision device based on artificial intelligence also includes a training module for:

[0020] determining a plurality of candidate users that meet the candidate user criteria;

[0021] Classifying each candidate user as a candidate positive sample or a candidate negative sample based on the purchase record of each candidate user;

[0022] Filtering out a plurality of filtered positive samples from the plurality of candidate positive samples, and filtering out a plurality of filtered negative samples from the plurality of candidate negative samples;

[0023] The purchase probability prediction model is trained based on the multiple filtered positive samples and the multiple filtered negative samples.

[0024] In the above solution, the determining module is further used for:

[0025] Based on the product of the price of the target product corresponding to the candidate attribute and the corresponding purchase probability, the corresponding estimated income is obtained.

[0026] In the above solution, the determining module is further used for:

[0027] The estimated loss is subtracted from the estimated income to obtain the estimated expectation corresponding to the candidate attribute.

[0028] In the above solution, the calling module is also used for:

[0029] extracting historical features from the historical usage data of the user;

[0030] Performing feature conversion processing on the historical features to obtain sparse historical features;

[0031] Classification processing is performed based on the sparse historical features to obtain the probability that the user uses the target product in violation of regulations as the violation probability.

[0032] In the above solution, the calling module is also used for:

[0033] Determine the average loss based on the historical loss data of the target product;

[0034] The product of the average loss and the violation probability is taken as the estimated loss.

[0035] In the above solution, the determining module is further used for:

[0036] When the highest estimated expectation is greater than the expected expectation of the target product, it is determined that the target product with the attribute will be launched.

[0037] In the above solution, the determining module is further used for:

[0038] adjusting the attributes of the target product based on the purchase feedback of the target product to determine actual purchase probabilities corresponding to different attributes;

[0039] adjusting the candidate attributes in the candidate attribute set based on the actual purchase probabilities corresponding to the different attributes;

[0040] The purchase probability prediction model is called based on the updated user data to determine the purchase probability corresponding to the adjusted candidate attribute, and the new attribute of the target product is determined based on the purchase probability corresponding to the adjusted candidate attribute and the violation probability of the user.

[0041] In the above solution, the product attribute decision device based on artificial intelligence also includes an evaluation module, which is used to:

[0042] When the attribute is greater than or equal to the historical attribute of the target product, determining a first purchase amount and a first non-purchase amount corresponding to the historical attribute; and

[0043] When the attribute is less than the historical attribute, determining a second purchase amount and a second non-purchase amount corresponding to the historical attribute;

[0044] The purchase probability prediction model and the violation probability prediction model are evaluated based on the first purchase amount, the first non-purchase amount, the second purchase amount and the second non-purchase amount.

[0045] In the above solution, the evaluation module is also used for:

[0046] Determine a first sum of the first unpurchased amount and the second unpurchased amount, and determine a first indicator based on a ratio of the second unpurchased amount to the first sum;

[0047] determining a second sum of the second purchased amount and the second unpurchased amount, and determining a second indicator based on a ratio of the second unpurchased amount to the second sum;

[0048] determining a third sum of the first purchase amount and the second purchase amount, and determining a third indicator based on a ratio of the first purchase amount to the third sum;

[0049] determining a relative difference between the attribute and the historical attribute, and determining a fourth indicator based on the relative difference;

[0050] Performing a weighted summation on the first indicator, the second indicator, the third indicator, and the fourth indicator to obtain an evaluation value;

[0051] When the evaluation value is greater than the evaluation threshold, it is determined that the purchase probability prediction model and the violation probability prediction model meet the requirements.

[0052] In the above solution, the training module is also used for:

[0053] Building an initial purchase probability prediction model through the first device;

[0054] Sending the initial purchase probability prediction model to multiple second devices, so that the multiple second devices train the initial purchase probability prediction model through different positive samples and negative samples stored in each of them, and encrypt and send the obtained training results to the first device;

[0055] The training results obtained by each of the plurality of second devices are aggregated through the first device to update the parameters of the initial purchase probability prediction model to obtain the trained purchase probability prediction model.

[0056] The present application also provides an electronic device, including:

[0057] A memory for storing executable instructions;

[0058] The processor is used to implement the product attribute decision method based on artificial intelligence provided in the embodiment of the present application when executing the executable instructions stored in the memory.

[0059] An embodiment of the present application provides a computer-readable storage medium storing executable instructions for causing a processor to execute and implement the artificial intelligence-based product attribute decision method provided in the embodiment of the present application.

[0060] The embodiments of the present application have the following beneficial effects:

[0061] The target product attributes are determined through machine learning models (purchase probability prediction model and violation probability prediction model), eliminating manual intervention. Therefore, the target product attributes are determined automatically and intelligently, improving the decision-making efficiency of product attributes. At the same time, when determining the purchase probability and violation probability, multi-source data such as user data and historical usage data are comprehensively considered. Therefore, the determined target product attributes are more accurate and more in line with user needs. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 is a schematic diagram of the architecture of the decision-making system 10 provided in an embodiment of the present application;

[0063] Figure 2 is a schematic diagram of the structure of the server 200 provided in an embodiment of the present application;

[0064] Figure 3 It is a flowchart of a product attribute decision method based on artificial intelligence provided in an embodiment of the present application;

[0065] Figure 4A This is a schematic diagram of the architecture of horizontal federated learning;

[0066] Figure 4B This is a schematic diagram of the architecture of vertical federated learning;

[0067] Figure 5 It is a schematic diagram of the training and prediction of the purchase probability prediction model provided in the embodiment of the present application;

[0068] Figure 6 It is a schematic diagram of the training and prediction of the violation probability prediction model provided in the embodiment of the present application;

[0069] Figure 7 It is a schematic diagram of pricing provided in an embodiment of the present application. DETAILED DESCRIPTION

[0070] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application will be further described in detail below in conjunction with the accompanying drawings. The described embodiments should not be regarded as limiting the present application. All other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of this application.

[0071] In the following description, reference is made to “some embodiments”, which describe a subset of all possible embodiments, but it will be understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0072] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application and are not intended to limit this application.

[0073] Before further describing the embodiments of the present application in detail, the nouns and terms involved in the embodiments of the present application are explained. The nouns and terms involved in the embodiments of the present application are subject to the following interpretations.

[0074] 1) Candidate attributes: attributes that a product may have, such as its price, appearance, and functions. The attributes of a product directly affect the purchase probability of the product.

[0075] 2) Estimated expectations: Before a product is put on the market, the revenue of the product is estimated and the estimated net profit of the product is obtained.

[0076] 3) Federated Learning: It is a distributed learning machine learning framework that conducts efficient machine learning between the computing devices of multiple participants while ensuring information security during data exchange, protecting terminal data and personal data privacy, and ensuring legality and compliance.

[0077] 4) B-end users: mainly enterprises, whose needs are mainly to meet the interests of the company and the needs of a group, which involves more group interests.

[0078] 5) Consumer surplus: Also known as the consumer's net benefit, it refers to the difference between the maximum total price a consumer is willing to pay and the total price actually paid when purchasing a certain amount of a certain commodity. Consumer surplus measures the additional benefits that consumers feel they have gained.

[0079] In the process of product updates and iterations, it is necessary to predict the user's acceptance of product attributes and willingness to purchase products, so as to adjust product attributes in time and increase revenue. The relevant technology establishes a single-input and single-output model, constantly changes the model's input, i.e., the attribute, to observe the change in the model's output, i.e., the predicted purchase probability, so as to determine whether the overall revenue has increased, and select the attribute with the highest overall revenue as the product's attribute. However, the data source type of this model is relatively single, with only one, and other factors that affect the user's willingness to purchase are not considered, so the predicted purchase probability is inaccurate. In addition, in order to determine the highest overall revenue, different attributes need to be input into the model multiple times to obtain different results. The process is complicated and redundant, inefficient, and requires human participation in decision-making.

[0080] The embodiment of the present application provides a product attribute decision method based on artificial intelligence, which can automatically and intelligently determine the attributes of a product, improve decision-making efficiency, and the determined attributes are more accurate and better meet user needs.

[0081] The product attribute decision method based on artificial intelligence provided in the embodiment of the present application can be implemented by various electronic devices, for example, it can be implemented by a terminal or a server alone, or it can be implemented by a server and a terminal in collaboration. For example, the server alone undertakes the product attribute decision method based on artificial intelligence described below, or the terminal sends an attribute determination request to the server, and the server executes the product attribute decision method based on artificial intelligence according to the received attribute determination request.

[0082] The electronic device for product attribute decision-making provided in the embodiments of the present application can be various types of terminal devices or servers, wherein the server can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms; the terminal can be a tablet computer, a laptop computer, a desktop computer, etc., but is not limited thereto. The terminal and the server can be directly or indirectly connected via wired or wireless communication, which is not limited in the embodiments of the present application.

[0083] Taking the server as an example, it can be a server cluster deployed in the cloud, opening artificial intelligence cloud services (AIaaS, AI as a Service) to users. The AIaaS platform will split several common AI services and provide independent or packaged services in the cloud. This service model is similar to an AI theme mall. All users can access one or more artificial intelligence services provided by the AIaaS platform through an application programming interface.

[0084] For example, one of the artificial intelligence cloud services may be a product attribute decision service, that is, a server in the cloud is encapsulated with a program for product attribute decision based on artificial intelligence provided by an embodiment of the present application. The terminal calls the product attribute decision service in the cloud service so that the server deployed in the cloud determines the attributes of the product based on user data and historical usage data, and sends the determined product attributes to the terminal.

[0085] In some embodiments, an exemplary decision system is described by taking the server and the terminal cooperating to implement the product attribute decision method based on artificial intelligence provided in the embodiments of the present application as an example. Figure 1 , Figure 1 1 is a schematic diagram of the architecture of the decision system 10 provided in an embodiment of the present application. The terminal 400 is connected to the server 200 via the network 300. The network 300 may be a wide area network or a local area network, or a combination of the two. The terminal sends an attribute determination request to the server, and the server performs an artificial intelligence-based product attribute decision according to the received attribute determination request.

[0086] In some embodiments, after the server 200 has trained the purchase probability prediction model and the violation probability prediction model, the server 200 receives an attribute determination request from the terminal 400, determines the purchase probability through the purchase probability prediction model, and determines the violation probability through the violation probability prediction model, determines the attributes of the product based on the purchase probability and the violation probability, and sends the determined product attributes to the terminal 400.

[0087] It can be foreseen that the above product decision method can have multiple uses, for example, to determine the price of a product, determine the appearance of a product, the function of a product, etc. It can also provide users with products with different attribute combinations to meet the needs of different users as much as possible.

[0088] The electronic device for product attribute decision making implementing the embodiment of the present application is Figure 1 The server 200 is shown as an example to illustrate the structure of the electronic device provided by the embodiment of the present application. Figure 2 , Figure 2 is a schematic diagram of the structure of the server 200 provided in an embodiment of the present application, Figure 2 The server 200 shown includes: at least one processor 410, a memory 440, and at least one network interface 420. The various components in the server 200 are coupled together via a bus system 430. It is understood that the bus system 430 is used to achieve connection and communication between these components. In addition to the data bus, the bus system 430 also includes a power bus, a control bus, and a status signal bus. However, for the sake of clarity, the bus system 430 is not described in detail. Figure 2 Various buses are labeled as bus system 430 .

[0089] Processor 410 can be an integrated circuit chip with signal processing capabilities, such as a general-purpose processor, a digital signal processor (DSP), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc., where the general-purpose processor can be a microprocessor or any conventional processor, etc.

[0090] The memory 440 may be removable, non-removable, or a combination thereof. Exemplary hardware devices include solid-state memory, hard drives, optical drives, etc. The memory 440 may optionally include one or more storage devices that are physically remote from the processor 410.

[0091] The memory 440 includes a volatile memory or a non-volatile memory, and may also include both volatile and non-volatile memories. The non-volatile memory may be a read-only memory (ROM), and the volatile memory may be a random access memory (RAM). The memory 440 described in the embodiments of the present application is intended to include any suitable type of memory.

[0092] In some embodiments, memory 440 can store data to support various operations, examples of which include programs, modules, and data structures, or a subset or superset thereof, as exemplarily described below.

[0093] Operating system 441, including system programs for processing various basic system services and performing hardware-related tasks, such as a framework layer, a core library layer, a driver layer, etc., for implementing various basic services and processing hardware-based tasks;

[0094] A network communication module 442, for reaching other computing devices via one or more (wired or wireless) network interfaces 420, exemplary network interfaces 420 include: Bluetooth, Wireless Compatibility Authentication (WiFi), and Universal Serial Bus (USB);

[0095] In some embodiments, the product attribute decision device based on artificial intelligence provided in the embodiments of the present application can be implemented in software. Figure 2 The product attribute decision device 443 based on artificial intelligence stored in the memory 440 is shown, which can be software in the form of a program and a plug-in, etc., including the following software modules: a calling module 4431, a determining module 4432, a training module 4433 and an evaluation module 4434. These modules are logical, so they can be arbitrarily combined or further split according to the functions implemented. The functions of each module will be explained below.

[0096] The following will describe the product attribute decision method based on artificial intelligence provided by the embodiment of the present application in conjunction with the accompanying drawings. The following description is based on the server as the execution subject of the product attribute decision based on artificial intelligence, which can be implemented by the server by running the various computer programs mentioned above; of course, based on the understanding of the following, it is not difficult to see that the product attribute decision method based on artificial intelligence provided by the embodiment of the present application can also be implemented by the terminal and the server in collaboration.

[0097] See also Figure 3 , Figure 3 is a flowchart of a product attribute decision method based on artificial intelligence provided by an embodiment of the present application, which will be combined with Figure 3 The steps shown are explained.

[0098] In step 101, a purchase probability prediction model is called based on user data to determine the purchase probabilities corresponding to the target product when the target product has different candidate attributes, wherein the different candidate attributes are included in a candidate attribute set.

[0099] In some embodiments, the target product can be an actual item (such as food), a service (such as car rental, housekeeping), or a virtual item (such as game props). The attributes of the target product can be price, appearance, function, gifts or rewards, etc. User data includes a candidate attribute set, user real-time order data, user historical behavior data, and user social data. Among them, the candidate attribute set includes multiple attributes of the target product in the same dimension, for example, it may include multiple prices, multiple colors, multiple functions, etc. User historical behavior data includes purchase records, historical violation records, and user basic attributes (gender, age, hobbies, etc.). User social data includes social application data (such as number of logins and visit duration, etc.), trajectory data, and browsing data.

[0100] In some embodiments, the purchase probability prediction model is a machine learning model, which may be a model composed of a gradient boosting decision tree (GBDT) + logistic regression (LR), random forest (RF) + LR, gradient boosting tree (GBT) + LR, extreme gradient boosting (XGBoost) + LR, etc. The following takes GBDT + LR as an example to illustrate the process of predicting the purchase probability by the purchase probability prediction model, wherein GBDT is an integrated decision tree, which includes multiple unrelated decision trees.

[0101] First, user features and attribute features of candidate attributes are extracted from user data through GBDT.

[0102] Then, the user features and attribute features are transformed according to the positions of the leaf nodes where they fall, and sparse intermediate features are obtained. The intermediate features are the output features obtained after GBDT performs feature transformation on the input features (user features and attribute features), and the output features are high-dimensional sparse features.

[0103] Each decision tree of GBDT includes 1 root node and multiple leaf nodes. After traversing GBDT, each dimension of user features and attribute features will eventually fall into different leaf nodes of different decision trees. For example, GBDT has 2 decision trees, the first decision tree has 2 leaf nodes, and the second decision tree has 3 leaf nodes. The user feature is 1-dimensional feature a, and the attribute feature is 1-dimensional feature b. After traversing GBDT, 1-dimensional feature a eventually falls into the second leaf node of the first decision tree, and 1-dimensional feature b eventually falls into the third leaf node of the second decision tree. After feature conversion processing, the corresponding code of the first tree is [0, 1], and the corresponding code of the second tree is [0, 0, 1]. The corresponding intermediate features are obtained by splicing the codes corresponding to the two trees, which is [0, 1, 0, 0, 1]. It can be seen that after feature conversion processing, sparse intermediate features can be obtained. In this way, the stability of the model can be improved, and in the subsequent LR processing process, because the intermediate features in the input LR are linearly separable, the processing speed of LR can be improved, and it is not easy to overfit. While performing feature conversion processing, GBDT also removes input features with high correlation, thereby obtaining output features with greater discrimination.

[0104] Finally, the intermediate features, i.e., the output features of GBDT, are input into LR, and LR performs classification processing based on the intermediate features to obtain the purchase probabilities corresponding to different candidate attributes in the candidate attribute set when the target product has different candidate attributes. In some possible examples, the intermediate features are multiplied by the corresponding weights in LR to obtain a product vector composed of multiple products, and the product vector is processed by the Sigmoid classification function to obtain multiple probabilities, i.e., the purchase probabilities corresponding to different candidate attributes.

[0105] In some embodiments, before calling the purchase probability prediction model to predict the purchase probability, it is also necessary to determine multiple candidate users who meet the candidate user criteria. Among them, the candidate users who meet the candidate user criteria may be users who have purchased the target product, consulted the target product, or followed / collected the target product. Then, each candidate user is divided into a candidate positive sample or a candidate negative sample based on the purchase record of each candidate user. Among them, the candidate positive sample is a user who has purchased the target product, and the candidate negative sample is a user who has not purchased the target product. After that, multiple filtered positive samples are filtered out from multiple candidate positive samples, and multiple filtered negative samples are filtered out from multiple candidate negative samples. By filtering biased samples, the validity of the sample is improved. For example, filter out positive samples and negative samples in special periods (such as economic depressions), or filter out positive samples and negative samples from B-end users. Finally, train the purchase probability prediction model based on multiple filtered positive samples and multiple filtered negative samples.

[0106] In some embodiments, the purchase probability prediction model may be trained by a federated learning method, wherein the federated learning includes vertical federated learning and horizontal federated learning.

[0107] See also Figure 4A , Figure 4A It is a schematic diagram of the architecture of horizontal federated learning. In the scenario of horizontal federation, the data of multiple devices (a first device and multiple second devices (the second devices 401, 402 and 403 are shown as examples)) are jointly modeled without leaving the local area. Horizontal federation is equivalent to adding sample data of each device to improve the model effect, that is, the sample data between different devices do not completely overlap (the users or transactions corresponding to the sample data have overlapping parts and different parts), but the characteristics of the sample data are the same. In the modeling process of the purchase probability prediction model, different devices can be regarded as separate clients for joint training, thereby helping to improve the results of all parties. During the training process, in the first step, an initial purchase probability prediction model is constructed through the first device; in the second step, the first device sends the initial purchase probability prediction model to multiple second devices; in the third step, multiple second devices train the initial purchase probability prediction model through different positive samples and negative samples stored in each device, and encrypt and send the training results obtained by each device (such as model parameters obtained after the training, intermediate results (such as gradients) during the training process) to the first device; in the fourth step, the first device aggregates the training results obtained by each of the multiple second devices to update the parameters of the initial purchase probability prediction model and obtain the trained purchase probability prediction model. In the fifth step, the first device sends the trained purchase probability prediction model to multiple second devices. Among them, the first device and the second device in the embodiment of the present application are both servers.

[0108] See also Figure 4B , Figure 4B This is a schematic diagram of the architecture of vertical federated learning. In the scenario of vertical federation, device 1 has user data; device 2 has both user data and label data. Vertical federation essentially improves the model effect by expanding the feature dimension of device 2's training data. Device 1 has a large amount of user data, while device 2 has a small amount of user data. The user data of device 1 can enrich the feature dimension of device 2's user data. During the training process, in the first step, collaborative party 3 sends a public key to device 1 and device 2 to encrypt the user data to be transmitted; in the second step, device 1 and device 2 respectively calculate the intermediate results of the features related to themselves, and encrypt the interaction to obtain their respective gradients and losses; in the third step, device 1 and device 2 respectively calculate their own encrypted gradients and add masks to send to collaborative party 3, while device 2 calculates the encrypted loss and sends it to collaborative party 3; in the fourth step, collaborative party 3 decrypts the gradient and loss and sends them back to device 1 and device 2, and device 1 and device 2 remove the mask and update the purchase probability prediction model.

[0109] It can be seen that federated learning can effectively solve the problem of data silos while ensuring user privacy protection and data security, allowing different devices to jointly model without sharing data, technically breaking down data silos and realizing AI collaboration, thereby improving the accuracy of the trained purchase probability prediction model.

[0110] In step 102, estimated revenue is determined based on the candidate attributes and the corresponding purchase probabilities.

[0111] In some embodiments, the estimated income is determined based on the candidate attributes and the corresponding purchase probabilities, that is, the corresponding estimated income is obtained based on the product of the price of the target product corresponding to the candidate attributes and the corresponding purchase probability.

[0112] When the candidate attribute is the price of the target product, the product of the price and the corresponding purchase probability is calculated to obtain the corresponding estimated income. When the candidate attribute is not the price of the target product, for example, the candidate attribute is the color of the target product, then the price of the target product corresponding to the candidate attribute is determined. For example, for the same mobile phone, different colors correspond to different prices, and the price of the mobile phone corresponding to the candidate attribute (such as silver) is determined. Then, the product of the price and the corresponding purchase probability is calculated to obtain the corresponding estimated income.

[0113] In step 103, a violation probability prediction model is called based on the user's historical usage data to determine the violation probability of the user's illegal use of the target product.

[0114] In some embodiments, the user's historical usage data includes the user's real-time order data, the user's historical behavior data, and the user's social data. The user's illegal use of the target product, such as when the target product is a car rental product, the user's illegal behavior while driving; when the target product is a mobile phone, the mobile phone is kept in a high temperature state for a long time, causing the mobile phone to be damaged. The violation probability prediction model is a machine learning model.

[0115] In some embodiments, a violation probability prediction model is called based on the user's historical usage data to determine the violation probability of the user's illegal use of the target product. This can be achieved by sampling the following methods: extracting historical features from the user's historical usage data; performing feature conversion processing on the historical features to obtain sparse historical features; performing classification processing based on the sparse historical features to obtain the probability of the user's illegal use of the target product as the violation probability.

[0116] It should be noted that the process of determining the violation probability through the violation probability prediction model is similar to the process of determining the purchase probability through the purchase probability prediction model in step 101. The specific process can be found in step 101 and will not be repeated here.

[0117] In step 104, an estimated loss is determined based on the probability of violation.

[0118] In some embodiments, the average loss can be determined based on the historical loss data of the target product, and the product of the average loss and the violation probability is used as the estimated loss. For example, the historical loss data of 100 users is obtained, of which 50 users have violated the regulations when using the target product, and the average loss caused by the violation is 500. Then, the violation probability corresponding to each of the 50 users is multiplied by 500 to obtain the estimated loss corresponding to each of the 50 users.

[0119] Among them, when the user has illegal use behavior and suffers losses due to illegal use behavior, the product provider shall bear the losses or compensate for the losses. For example, the target product is an insurance product for a certain mobile phone. When the user's mobile phone is damaged due to personal reasons and pays a certain amount of repair fees, the insurance product provider will compensate the user for the repair fees. For another example, the target product is a car rental rights product. If the user violates the regulations while driving and causes deductions (such as illegal deductions + handling fees), the car rental rights product provider will bear the deductions.

[0120] In step 105, the estimated expectation corresponding to the candidate attribute is determined based on the estimated income and the estimated loss.

[0121] In some embodiments, the estimated expectation corresponding to the candidate attribute is determined based on the estimated income and the estimated loss, that is, the estimated loss is subtracted from the estimated income to obtain the estimated expectation corresponding to the candidate attribute.

[0122] In step 106 , the candidate attribute with the highest estimated expectation in the candidate attribute set is used as the attribute of the target product.

[0123] In some embodiments, because there are multiple candidate attributes and corresponding multiple estimated revenues, the estimated expectations obtained in step 105 are also multiple, and each estimated expectation corresponds to a different candidate attribute. The highest estimated expectation can be selected from the multiple estimated expectations, and the candidate attribute corresponding to the highest estimated expectation is used as the attribute of the target product. In this way, the loss caused by the possible illegal use behavior of the user can be minimized and the estimated expectation can be maximized.

[0124] After determining the attributes of the target product, it is also necessary to determine whether to launch the target product. In some possible examples, when the highest estimated expectation is greater than the expected expectation of the target product, it is determined that the target product with the attribute corresponding to the highest estimated expectation will be launched. When the highest estimated expectation is not greater than the expected expectation of the target product, the target product will not be launched.

[0125] In some embodiments, after the target product is launched, the attributes of the target product are adjusted based on the purchase feedback of the target product to determine the actual purchase probabilities corresponding to different attributes. For example, the price of the target product is adjusted to determine the actual purchase probabilities corresponding to different prices. Then, the candidate attributes in the candidate attribute set are adjusted based on the actual purchase probabilities corresponding to different attributes. For example, in the process of adjusting the price of the target product, it is determined that the actual purchase probabilities of price a and price b are higher, and the candidate attribute set of the previous purchase probability prediction model does not include price a and price b, then price a and price b are added to the candidate attribute set, and the user data is updated. The purchase probability prediction model is called, and predictions are made based on the updated user data to determine the purchase probability when the target product has each attribute in the new candidate attribute set, and the estimated income is determined based on the purchase probability and the corresponding candidate attributes, and the new attributes of the target product corresponding to the highest estimated expectation are determined based on the estimated income and estimated loss.

[0126] It can be seen that the artificial intelligence-based product attribute decision method provided in the embodiment of the present application can not only quickly and accurately determine the attributes of the target product, but also adjust it according to actual conditions. During the adjustment process, as the amount of collected user data increases, the purchase probability prediction model and the violation probability prediction model can be incrementally trained to improve the accuracy of the model, thereby obtaining more accurate attributes of the target product.

[0127] In some embodiments, after determining the attributes of the target product, the quality (such as the accuracy of the prediction) of the purchase probability prediction model and the violation probability prediction model can also be evaluated based on the attributes of the target product. When the attribute of the target product is greater than or equal to the historical attribute of the target product, the first purchase amount and the first non-purchased amount corresponding to the historical attribute are determined; when the attribute of the target product is less than the historical attribute, the second purchase amount and the second non-purchased amount corresponding to the historical attribute are determined; the purchase probability prediction model and the violation probability prediction model are evaluated based on the first purchase amount, the first non-purchased amount, the second purchase amount, and the second non-purchased amount.

[0128] In some possible examples, the evaluation process proceeds as follows.

[0129] Determine the first sum (B+D) of the first unpurchased amount (such as B) and the second unpurchased amount (such as D), and determine the first indicator (D / (B+D)) based on the ratio of the second unpurchased amount to the first sum. The higher the first indicator, the greater the corresponding purchase probability, which can reduce consumer losses.

[0130] A second sum (C+D) of the second purchase amount (such as C) and the second non-purchased amount (such as D) is determined, and a second indicator (D / (C+D)) is determined based on the ratio of the second non-purchased amount (such as D) to the second sum. The smaller the second indicator is, the more it can avoid the situation where users who have purchased the target product are unwilling to purchase products with current attributes again.

[0131] Determine the third sum (C+D) of the first purchase amount (such as A) and the second purchase amount (such as C), and determine the third indicator (A / (A+C)) based on the ratio of the first purchase amount to the third sum. The higher the third indicator, the more likely it is to increase consumer surplus and improve estimated expectations.

[0132] Determine the properties of the target product (such as P i ) and historical attributes (such as Psug i ) relative difference ((P i -Psug i ) / P i ), and determine the fourth index (Median(Max1(0, ​​(P1-Psug1) / P1), Max2(0, (P2-Psug2) / P2)…, Max n (0, (Pn-Psug n ) / P n ))). Among them, P i represents the attribute of the target product for the i-th person, Psug irepresents the historical attributes of the target product for the i-th person. Max(,) means taking the larger of two numbers, and Median(,,,) means taking the median of multiple values.

[0133] According to actual needs, different weights are assigned to the first indicator, the second indicator, the third indicator and the fourth indicator, and the first indicator, the second indicator, the third indicator and the fourth indicator are weighted and summed to obtain the evaluation value. When the evaluation value is greater than the evaluation threshold, it is determined that the purchase probability prediction model and the violation probability prediction model meet the requirements. When the evaluation value is not greater than the evaluation threshold, it is determined which indicator has the greatest impact on the evaluation value, and the above two models are adjusted according to the indicator.

[0134] Through such an evaluation process, the parameters of the above two models can be adjusted again to improve the prediction accuracy of the above two models.

[0135] It can be seen that the embodiment of the present application uses machine learning models (purchase probability prediction model and violation probability prediction model) to help determine the attributes of the target product, eliminating manual intervention, and thus automatically and intelligently determining the attributes of the target product, improving the decision-making efficiency of product attributes. At the same time, when determining the purchase probability and violation probability, multi-source data such as user data and historical usage data are comprehensively considered, so the determined attributes of the target product are more accurate and more in line with user needs.

[0136] The following will illustrate an exemplary application of an embodiment of the present application when the product is a car rental rights product and the product attribute is price. When the product is a car rental rights product, if the user purchases the product and there is a violation during the use of the car rental, the car rental rights product party will bear the violation costs and handling fees. When the product attribute is price, it is necessary to determine the product price when the profit is maximized. The product price is determined by the purchase probability prediction model and the violation probability prediction model. Before this, it is necessary to first train the purchase probability prediction model and the violation probability prediction model. The two models are described below.

[0137] See also Figure 5 , Figure 5It is a schematic diagram of the training and prediction of the purchase probability prediction model provided by the embodiment of the present application. User data is obtained through a user portrait database or a user inquiry database or a third party, and the user data is used as a sample for training the purchase probability prediction model. Among them, the sample includes positive samples and negative samples. The positive sample is a record obtained by removing special records (such as purchase records during the economic depression, purchase records of B-end users) from the purchase record; the negative sample is a record of no purchase, including no purchase of products after inquiry, no purchase of products after collection, no purchase of products after adding to the shopping cart, and no payment within the scheduled time after placing an order. Both the positive and negative samples include user basic attributes (gender, age, hobbies, etc.), historical purchase records (when no purchase, the historical purchase record is empty), etc. The purchase probability prediction model belongs to a two-category model. After the purchase probability prediction model is trained, it can be used to predict the user's purchase probability of the product.

[0138] See also Figure 6 , Figure 6 Schematic diagram of the training and prediction of the violation probability prediction model provided by the embodiment of the present application. The user's historical usage data (such as historical violation records, car rental locations, car rental days, etc.) and user portraits, purchase records, and violation records are used as samples for training the violation probability prediction model. Among them, the samples include positive samples and negative samples. The positive sample is a historical violation record, and the negative sample is a historical non-violation record. The violation probability prediction model belongs to a two-classification model. After the violation probability prediction model is trained, it can be used to predict the probability of a user violating the rules during the car rental process.

[0139] See also Figure 7 , Figure 7It is a schematic diagram of pricing provided by an embodiment of the present application. First, possible prices are estimated according to user traffic to form a candidate price set. The price range of the candidate price set can be determined based on personal experience or based on historical data analysis. Then the candidate price set, user real-time order data (including rental days, rental cities, pick-up time, etc.), user historical behavior data (including purchase records, historical violation records, user basic attributes, etc.) and user social data (including social application data (such as login times, access duration, etc.), trajectory data and browsing data, etc.) are input into the purchase estimation model to predict the purchase probability of users corresponding to different prices in the candidate price set. At the same time, the user real-time order data, user historical behavior data and user social data are input into the violation probability prediction model to predict the user's violation probability. Afterwards, the estimated income is obtained by calculating the product of the user's purchase probability and the corresponding price, the estimated loss is obtained by calculating the product of the user's violation probability and the average loss, and the estimated loss is subtracted from the estimated income to obtain the estimated expectation. Because there are multiple prices, there are also multiple corresponding purchase probabilities, and multiple estimated expectations can be calculated, and the price with the highest estimated expectation is selected as the pricing of the product.

[0140] according to Figure 7 It can be seen that after determining the product pricing, it is also necessary to determine whether to launch the product. After the product is launched, the product pricing needs to be adaptively adjusted according to the product purchase feedback to determine the actual purchase probability at different pricing. The price in the candidate price set of the purchase estimation model is adjusted based on the demand curve feedback composed of the pricing and the actual purchase probability. After that, the purchase probability is predicted again through the purchase probability prediction model, and the corresponding product pricing is determined.

[0141] Estimated product pricing Purchase quantity Not purchased Psug>=P A B Psug<P C D

[0142] Table 1 Evaluation scheme

[0143] Among them, after determining the product pricing, the accuracy of the purchase probability prediction model and the violation probability prediction model can also be evaluated. The evaluation criteria include: a) For historical orders that have been purchased, the current estimated product pricing (Psug) should be greater than or equal to the historical pricing (P); b) For historical orders that have not been purchased, the current estimated product pricing (Psug) should be less than the historical pricing (P). The evaluation scheme can be seen in Table 1. A and B in Table 1 are the historical purchase volume and historical non-purchase volume when Psug>=P, respectively, and C and D are the historical purchase volume and historical non-purchase volume when Psug<P, respectively.

[0144] The quality of the purchase probability prediction model and the violation probability prediction model can be evaluated based on the following four indicators.

[0145] ① Price Decrease Recall (PDR), PDR = D / (B+D). The higher the PDR, the more likely the estimated product pricing is to facilitate a transaction and reduce consumer losses (corresponding to criterion b).

[0146] ② Price Decrease Precision (PDP), PDP = D / (C+D). The smaller the PDP, the less likely it is to mistakenly kill a transaction that has already been completed. This avoids the situation where a user purchases a product when the product is priced at the historical price in the historical order, but no longer purchases the product when the product is priced at the estimated product price.

[0147] ③ Price Increase Recall (PIR), PIR = A / (A+C), the higher the PIR, the more likely it is to increase consumer surplus (corresponding to criterion a).

[0148] ④ Purchase Regret (PR), PR = Median (Max1 (0, (P1-Psug1) / P1), Max2 (0, (P2-Psug2) / P2)…, Max n (0, (P n -Psug n ) / P n )), the lower the PR, the closer the estimated price is to the actual purchase price, and the more accurate the estimated price is.

[0149] In some possible examples, different weights may be assigned to the above four indicators according to the characteristics of the business, and then weighted summed up, and the quality of the purchase probability prediction model and the violation probability prediction model may be determined according to the size of the weighted sum.

[0150] The following is a description of an exemplary structure of the product attribute decision device 443 based on artificial intelligence provided in the present application as a software module. Figure 2 As shown, the software modules stored in the artificial intelligence-based product attribute decision device 443 of the memory 440 may include: a calling module 4431, a determining module 4432, a training module 4433, and an evaluation module 4434.

[0151] Calling module 4431 is used to call a purchase probability prediction model based on user data to determine the purchase probabilities corresponding to the target product when the target product has different candidate attributes, wherein the different candidate attributes are included in a candidate attribute set; and to call a violation probability prediction model based on the user's historical usage data to determine the violation probability of the user's illegal use of the target product, and to determine the estimated loss based on the violation probability; determining module 4432 is used to determine the estimated income based on the candidate attributes and the corresponding purchase probabilities; and to determine the estimated expectations corresponding to the candidate attributes based on the estimated income and the estimated losses; and to use the candidate attribute with the highest estimated expectation in the candidate attribute set as the attribute of the target product.

[0152] In some embodiments, calling module 4431 is also used to extract user characteristics and attribute characteristics of candidate attributes from user data; perform feature conversion processing on user characteristics and attribute characteristics to obtain sparse intermediate features; perform classification processing based on the intermediate features to obtain the purchase probabilities corresponding to when the target product has different candidate attributes in the candidate attribute set.

[0153] In some embodiments, the artificial intelligence-based product attribute decision device also includes a training module 4433, which is used to determine multiple candidate users who meet the candidate user criteria; divide each candidate user into a candidate positive sample or a candidate negative sample based on the purchase record of each candidate user; filter out multiple filtered positive samples from multiple candidate positive samples, and filter out multiple filtered negative samples from multiple candidate negative samples; train a purchase probability prediction model based on multiple filtered positive samples and multiple filtered negative samples.

[0154] In some embodiments, the determination module 4432 is further used to obtain the corresponding estimated income based on the product of the price of the target product corresponding to the candidate attribute and the corresponding purchase probability.

[0155] In some embodiments, the determination module 4432 is further used to subtract the estimated loss from the estimated income to obtain the estimated expectation corresponding to the candidate attribute.

[0156] In some embodiments, calling module 4431 is also used to extract historical features from the user's historical usage data; perform feature conversion processing on the historical features to obtain sparse historical features; perform classification processing based on the sparse historical features to obtain the probability of the user's illegal use of the target product as the violation probability.

[0157] In some embodiments, calling module 4431 is also used to determine the average loss based on the historical loss data of the target product; and the product of the average loss and the probability of violation is used as the estimated loss.

[0158] In some embodiments, the determination module 4432 is further used to determine that a target product with attributes will be launched when the highest estimated expectation is greater than the expected expectation of the target product.

[0159] In some embodiments, the determination module 4432 is also used to adjust the attributes of the target product based on the purchase feedback of the target product to determine the actual purchase probabilities corresponding to different attributes; adjust the candidate attributes in the candidate attribute set based on the actual purchase probabilities corresponding to different attributes; call the purchase probability prediction model based on the updated user data to determine the purchase probability corresponding to the adjusted candidate attributes, and determine the new attributes of the target product based on the purchase probability corresponding to the adjusted candidate attributes and the user's violation probability.

[0160] In some embodiments, the artificial intelligence-based product attribute decision device also includes an evaluation module 4434, which is used to determine a first purchase quantity and a first non-purchased quantity corresponding to the historical attribute when the attribute is greater than or equal to the historical attribute of the target product; and when the attribute is less than the historical attribute, determine a second purchase quantity and a second non-purchased quantity corresponding to the historical attribute; and evaluate and process the purchase probability prediction model and the violation probability prediction model based on the first purchase quantity, the first non-purchased quantity, the second purchase quantity and the second non-purchased quantity.

[0161] In some embodiments, the evaluation module 4434 is also used to determine a first sum of a first unpurchased amount and a second unpurchased amount, and determine a first indicator based on the ratio of the second unpurchased amount to the first sum; determine a second sum of a second purchased amount and a second unpurchased amount, and determine a second indicator based on the ratio of the second unpurchased amount to the second sum; determine a third sum of a first purchased amount and a second purchased amount, and determine a third indicator based on the ratio of the first purchased amount to the third sum; determine a relative difference between an attribute and a historical attribute, and determine a fourth indicator based on the relative difference; perform weighted summation of the first indicator, the second indicator, the third indicator and the fourth indicator to obtain an evaluation value; when the evaluation value is greater than the evaluation threshold, determine that the purchase probability prediction model and the violation probability prediction model meet the requirements.

[0162] In some embodiments, the training module 4433 is also used to build an initial purchase probability prediction model through the first device; send the initial purchase probability prediction model to multiple second devices, so that the multiple second devices train the initial purchase probability prediction model through different positive samples and negative samples stored in each of them, and encrypt the training results obtained by each of them and send them to the first device; aggregate the training results obtained by each of the multiple second devices through the first device to update the parameters of the initial purchase probability prediction model to obtain the trained purchase probability prediction model.

[0163] The present application embodiment provides a computer-readable storage medium storing executable instructions, wherein the executable instructions are stored. When the executable instructions are executed by a processor, the processor will execute the product attribute decision method based on artificial intelligence provided by the present application embodiment, for example, Figure 3 The artificial intelligence-based product attribute decision-making method is shown.

[0164] In some embodiments, the computer-readable storage medium may be a memory such as FRAM, ROM, PROM, EPROM, EEPROM, flash memory, magnetic surface storage, optical disk, or CD-ROM; or it may be various devices including one or any combination of the above memories.

[0165] In some embodiments, executable instructions may be in the form of a program, software, software module, script or code, written in any form of programming language (including compiled or interpreted languages, or declarative or procedural languages), and may be deployed in any form, including as a stand-alone program or as a module, component, subroutine or other unit suitable for use in a computing environment.

[0166] As an example, executable instructions may, but need not, correspond to a file in a file system, may be stored as part of a file that stores other programs or data, such as in one or more scripts in a HyperText Markup Language (HTML) document, in a single file dedicated to the program in question, or in multiple coordinated files (e.g., files storing one or more modules, subroutines, or code portions).

[0167] By way of example, executable instructions may be deployed to be executed on one computing device, or on multiple computing devices located at one site, or on multiple computing devices distributed across multiple sites and interconnected by a communication network.

[0168] In summary, the embodiment of the present application determines the attributes of the target product through a machine learning model (purchase probability prediction model and violation probability prediction model), eliminating manual intervention, and thus automatically and intelligently determines the attributes of the target product, thereby improving the decision-making efficiency of the product attributes. At the same time, when determining the purchase probability and violation probability, multi-source data such as user data and historical usage data are comprehensively considered, so the determined attributes of the target product are more accurate and more in line with user needs.

[0169] The above is only an embodiment of the present application and is not intended to limit the protection scope of the present application. Any modifications, equivalent substitutions and improvements made within the spirit and scope of the present application are included in the protection scope of the present application.

Claims

1. A product attribute decision method based on artificial intelligence, characterized in that: The method comprises: Calling a purchase probability prediction model based on user data to determine the purchase probabilities corresponding to the target product when the target product has different candidate attributes, wherein the different candidate attributes are included in the candidate attribute set; Obtaining a corresponding estimated income based on the product of the price of the target product corresponding to the candidate attribute and the corresponding purchase probability; Calling a violation probability prediction model based on the user's historical usage data to determine the violation probability of the user's illegal use of the target product, and determining the average loss based on the historical loss data of the target product; The product of the average loss and the violation probability is taken as the estimated loss; Subtracting the estimated loss from the estimated income to obtain an estimated expectation corresponding to the candidate attribute; Taking the candidate attribute with the highest estimated expectation in the candidate attribute set as the attribute of the target product; When the highest estimated expectation is greater than the expected expectation of the target product, determining to launch the target product with the attribute; The calling of the purchase probability prediction model based on the user data to determine the purchase probabilities corresponding to the target products when the target products have different candidate attributes respectively includes: Extracting user features and attribute features of the candidate attributes from the user data; Performing feature conversion processing on the user features and the attribute features to obtain sparse intermediate features; Performing classification processing based on the intermediate features to obtain purchase probabilities corresponding to different candidate attributes in the candidate attribute set when the target product has different candidate attributes; The calling of the violation probability prediction model based on the historical usage data of the user to determine the violation probability of the user's illegal use of the target product includes: extracting historical features from the historical usage data of the user; Performing feature conversion processing on the historical features to obtain sparse historical features; Classification processing is performed based on the sparse historical features to obtain the probability that the user uses the target product in violation of regulations as the violation probability.

2. The method according to claim 1, characterized in that Before calling the purchase probability prediction model based on user data, the method further includes: determining a plurality of candidate users that meet the candidate user criteria; Classifying each candidate user as a candidate positive sample or a candidate negative sample based on the purchase record of each candidate user; Filtering out a plurality of filtered positive samples from the plurality of candidate positive samples, and filtering out a plurality of filtered negative samples from the plurality of candidate negative samples; The purchase probability prediction model is trained based on the multiple filtered positive samples and the multiple filtered negative samples.

3. The method according to claim 1, characterized in that After selecting the candidate attribute with the highest expected value in the candidate attribute set as the attribute of the target product, the method further includes: adjusting the attributes of the target product based on the purchase feedback of the target product to determine actual purchase probabilities corresponding to different attributes; adjusting the candidate attributes in the candidate attribute set based on the actual purchase probabilities corresponding to the different attributes; The purchase probability prediction model is called based on the updated user data to determine the purchase probability corresponding to the adjusted candidate attribute, and the new attribute of the target product is determined based on the purchase probability corresponding to the adjusted candidate attribute and the violation probability of the user.

4. The method according to claim 1, characterized in that: After selecting the candidate attribute with the highest expected value in the candidate attribute set as the attribute of the target product, the method further includes: When the attribute is greater than or equal to the historical attribute of the target product, determining a first purchase amount and a first non-purchase amount corresponding to the historical attribute; and When the attribute is less than the historical attribute, determining a second purchase amount and a second non-purchase amount corresponding to the historical attribute; The purchase probability prediction model and the violation probability prediction model are evaluated based on the first purchase amount, the first non-purchase amount, the second purchase amount and the second non-purchase amount.

5. The method according to claim 4, characterized in that The evaluating the purchase probability prediction model and the violation probability prediction model based on the first purchase amount, the first non-purchase amount, the second purchase amount, and the second non-purchase amount includes: Determine a first sum of the first unpurchased amount and the second unpurchased amount, and determine a first indicator based on a ratio of the second unpurchased amount to the first sum; determining a second sum of the second purchased amount and the second unpurchased amount, and determining a second indicator based on a ratio of the second unpurchased amount to the second sum; determining a third sum of the first purchase amount and the second purchase amount, and determining a third indicator based on a ratio of the first purchase amount to the third sum; determining a relative difference between the attribute and the historical attribute, and determining a fourth indicator based on the relative difference; Performing a weighted summation on the first indicator, the second indicator, the third indicator, and the fourth indicator to obtain an evaluation value; When the evaluation value is greater than the evaluation threshold, it is determined that the purchase probability prediction model and the violation probability prediction model meet the requirements.

6. The method according to claim 1, characterized in that Before calling the purchase probability prediction model based on user data, the method further includes: Building an initial purchase probability prediction model through the first device; Sending the initial purchase probability prediction model to multiple second devices, so that the multiple second devices train the initial purchase probability prediction model through different positive samples and negative samples stored in each of them, and encrypt and send the obtained training results to the first device; The training results obtained by each of the plurality of second devices are aggregated through the first device to update the parameters of the initial purchase probability prediction model to obtain the trained purchase probability prediction model.

7. A product attribute decision-making device based on artificial intelligence, characterized in that: include: A calling module is used to call a purchase probability prediction model based on user data to determine the purchase probabilities corresponding to the target product when the target product has different candidate attributes, wherein the different candidate attributes are included in a candidate attribute set; and to call a violation probability prediction model based on the user's historical usage data to determine the violation probability of the user's illegal use of the target product, and determine the estimated loss based on the violation probability; the calling module is also used to extract user features and attribute features of the candidate attributes from the user data; perform feature conversion processing on the user features and attribute features to obtain sparse intermediate features; perform classification processing based on the intermediate features to obtain the purchase probabilities corresponding to the target product when it has different candidate attributes in the candidate attribute set; the calling module is also used to extract historical features from the user's historical usage data; perform feature conversion processing on the historical features to obtain sparse historical features; perform classification processing based on the sparse historical features to obtain the probability of the user's illegal use of the target product as the violation probability; the calling module is also used to determine the average loss based on the historical loss data of the target product; and take the product of the average loss and the violation probability as the estimated loss; A determination module is used to obtain the corresponding estimated income based on the product of the price of the target product corresponding to the candidate attribute and the corresponding purchase probability; and to subtract the estimated loss from the estimated income to obtain the estimated expectation corresponding to the candidate attribute; and to use the candidate attribute with the highest estimated expectation in the candidate attribute set as the attribute of the target product. The determination module is also used to determine that the target product with the attribute will be launched when the highest estimated expectation is greater than the expected expectation of the target product.

8. An electronic device, characterized in that: include: A memory for storing executable instructions; A processor, used to implement the artificial intelligence-based product attribute decision method described in any one of claims 1 to 6 when executing the executable instructions stored in the memory.

9. A computer-readable storage medium, characterized in that: Executable instructions are stored, which are used to cause the processor to execute and implement the product attribute decision method based on artificial intelligence as described in any one of claims 1 to 6.

Images