Information processing and model training method and device, and storage medium

Abstract

Embodiments of the invention disclose an information processing and model training method and apparatus, and a storage medium. The method comprises the steps of obtaining an object feature of a target object, multiple groups of resource niche features of multiple resource niche and multiple groups of business features of multiple businesses; inputting the object features, the multiple groups of resource niche features and the multiple groups of business features into a conversion rate prediction model to obtain a plurality of conversion rates of the target object under the multiple businesses and the multiple resource niches; and determining a target resource niche matched with the target object under the target business according to the plurality of conversion rates, the target resource niche being a part of the resource niche, and the target business being a part of the plurality of businesses.

Description

Technical Field

[0001] The present application relates to the field of e-commerce technology, and in particular to an information processing and model training method, device, and storage medium. Background Art

[0002] In order to facilitate users' shopping, a credit consumption product is set up on the shopping APP. When users use the credit consumption product, they will jump to multiple core pages. These core pages have locations where users can click to interact, which are resource locations, such as Figure 1 As shown in the figure, the user clicks on a resource position in the exposed page and completes the business process corresponding to the resource position, which means that the user completes the conversion.

[0003] In the related art, the priority order of resource positions is set, and the resource positions to be pushed to users are determined based on the users' conversion of the resource positions, and the resource positions are set on the core page. Since users have different demands for resource positions in different scenarios, this method of determining the resource positions in the core page is inaccurate, thereby reducing the user's conversion rate under the resource positions. Summary of the invention

[0004] In order to solve the above technical problems, the embodiments of the present application hope to provide an information processing and model training method, device, and storage medium that can improve the conversion rate of users in resource positions.

[0005] The technical solution of this application is implemented as follows:

[0006] The present invention provides an information processing method, which includes:

[0007] Acquire object characteristics of a target object, multiple groups of resource bit characteristics of multiple resource bits, and multiple groups of service characteristics of multiple services;

[0008] Inputting the object features, the multiple groups of resource position features and the multiple groups of business features into a conversion rate prediction model to obtain multiple conversion rates of the target object under the multiple businesses and the multiple resource positions respectively;

[0009] A target resource bit matching the target object under a target business is determined according to the multiple conversion rates, the target resource bit is part of the resource bits, and the target business is part of the multiple businesses.

[0010] The present application provides an information processing device, the device comprising:

[0011] A first acquisition unit, used to acquire an object feature of a target object, multiple groups of resource bit features of multiple resource bits, and multiple groups of service features of multiple services;

[0012] A first input unit is used to input the object feature, the multiple groups of resource position features and the multiple groups of business features into a conversion rate prediction model to obtain multiple conversion rates of the target object under the multiple businesses and the multiple resource positions respectively;

[0013] A first determining unit is used to determine target resource bits matching the target object under a target business according to the multiple conversion rates, wherein the target resource bits are part of the resource bits, and the target business is part of the multiple businesses.

[0014] The present application provides a model training method, which includes:

[0015] Acquire multiple groups of sample object features of multiple sample objects, multiple groups of sample resource bit features corresponding to multiple sample resource bits, multiple groups of sample service features corresponding to multiple sample services, and multiple sample labels;

[0016] Inputting the multiple groups of sample object features, the multiple groups of sample resource location features, and the multiple groups of sample business features into an initial conversion rate prediction model to obtain multiple sample output parameters;

[0017] Determining a loss value of the initial conversion rate prediction model according to the multiple sample output parameters and the multiple sample labels; the loss value includes a cross entropy loss, a reconstruction loss, and a regularization loss;

[0018] When the loss value is greater than a preset loss threshold, the initial conversion rate prediction model is continuously trained using the multiple groups of sample object features, the multiple groups of sample resource location features, the multiple groups of sample service features and the multiple sample labels to obtain a training model;

[0019] When the training model loss value corresponding to the training model is less than or equal to the preset loss threshold, the training model is determined as the conversion rate prediction model.

[0020] The present application provides a model training device, the device comprising:

[0021] A second acquisition unit, configured to acquire multiple groups of sample object features of multiple sample objects, multiple groups of sample resource bit features corresponding to multiple sample resource bits, multiple groups of sample service features corresponding to multiple sample services, and multiple sample labels;

[0022] A second input unit is used to input the multiple groups of sample object features, the multiple groups of sample resource location features, and the multiple groups of sample business features into an initial conversion rate prediction model to obtain multiple sample output parameters;

[0023] A second determination unit is used to determine a loss value of the initial conversion rate prediction model according to the multiple sample output parameters and the multiple sample labels; the loss value includes a cross entropy loss, a reconstruction loss and a regularization loss; when the training model loss value corresponding to the training model is less than or equal to the preset loss threshold, the training model is determined as the conversion rate prediction model;

[0024] A training unit is used to continue training the initial conversion rate prediction model using the multiple groups of sample object features, the multiple groups of sample resource location features, the multiple groups of sample business features and the multiple sample labels to obtain a training model when the loss value is greater than a preset loss threshold.

[0025] The present application provides an information processing device, the device comprising:

[0026] A first memory, a first processor and a first communication bus, wherein the first memory communicates with the first processor via the first communication bus, the first memory stores an information processing program executable by the first processor, and when the information processing program is executed, the above-mentioned information processing method is executed by the first processor.

[0027] The present application provides a model training device, the device comprising:

[0028] A second memory, a second processor and a second communication bus, wherein the second memory communicates with the second processor via the second communication bus, and the second memory stores an information processing program executable by the second processor. When the information processing program is executed, the above-mentioned model training method is executed by the second processor.

[0029] An embodiment of the present application provides a storage medium having a computer program stored thereon, which is applied to an information processing device and a model training device, and is characterized in that when the computer program is executed by a first processor, the above-mentioned information processing method is implemented, and when the computer program is executed by a second processor, the above-mentioned model training method is implemented.

[0030] The embodiment of the present application provides an information processing and model training method, device, and storage medium. The information processing method includes: obtaining object features of a target object, multiple resource bit features of multiple resource bits, and multiple business features of multiple services; inputting the object features, multiple resource bit features, and multiple business features into a conversion rate prediction model to obtain multiple conversion rates of the target object under multiple services and multiple resource bits; determining the target resource bits that match the target object under the target service according to the multiple conversion rates, the target resource bits are part of the resource bits, and the target service is part of the service among the multiple services. The above method is adopted to implement the scheme, and the information processing device obtains the object features of the target object, multiple resource bit features of multiple resource bits, and multiple business features of multiple services, and uses the conversion rate prediction model to determine multiple conversion rates of the target object under multiple services and multiple resource bits according to the object features, multiple resource bit features, and multiple business features; determining the target resource bit of the target object under the target service according to the multiple conversion rates, so that the target resource bit is the resource bit that the target object most expects to operate under the target service, thereby improving the accuracy of determining the resource bit in the core page, thereby improving the conversion rate of the user under the resource bit. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 An exemplary page schematic diagram including resource bits provided for an embodiment of the present application;

[0032] Figure 2 A flow chart of an information processing method provided in an embodiment of the present application;

[0033] Figure 3 A schematic diagram of an exemplary model structure provided for an embodiment of the present application;

[0034] Figure 4 A flow chart of a model training method provided in an embodiment of the present application;

[0035] Figure 5 A schematic diagram of the structure of an information processing device provided in an embodiment of the present application Figure 1 ;

[0036] Figure 6 A schematic diagram of the structure of an information processing device provided in an embodiment of the present application Figure 2

[0037] Figure 7 A schematic diagram of the structure of a model training device provided in an embodiment of the present application Figure 1 ;

[0038] Figure 8 A schematic diagram of the structure of a model training device provided in an embodiment of the present application Figure 2 . DETAILED DESCRIPTION

[0039] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. It should be understood that the specific embodiments described here are only used to explain the present application and are not used to limit the present application.

[0040] Figure 1 Specifically, it is an interactive schematic diagram under a page in the prior art, and specifically, it is an interactive schematic diagram under each page that may be exposed to the user when using a credit consumption product. Figure 1 The left page in the is the billing page. Figure 1 The right page is the repayment success page. Figure 1 The middle page in the diagram is the repayment page; these pages contain locations where users can click to interact, also known as resource locations.

[0041] The present application provides an information processing method, which is applied to an information processing device. Figure 2 A flow chart of an information processing method provided in an embodiment of the present application, such as Figure 2 As shown, the information processing method may include:

[0042] S101, obtaining object characteristics of a target object, multiple groups of resource bit characteristics of multiple resource bits, and multiple groups of service characteristics of multiple services.

[0043] An information processing method provided in an embodiment of the present application is applicable to a scenario in which a target resource location matching a target object under a target service is determined.

[0044] In the embodiments of the present application, the information processing device can be implemented in various forms. For example, the information processing device described in the present application may include devices such as mobile phones, cameras, tablet computers, laptop computers, PDAs, portable media players (PMPs), navigation devices, wearable devices, smart bracelets, pedometers, and devices such as digital TVs, desktop computers, and servers.

[0045] In the embodiment of the present application, multiple resource bits correspond one-to-one to multiple groups of resource bit features, that is, one resource bit corresponds to one group of resource bit features.

[0046] In the embodiment of the present application, multiple services correspond one-to-one to multiple groups of service features, that is, one service corresponds to one group of service features.

[0047] In the embodiment of the present application, the resource bit characteristics include the location of the resource bit on the core display page, the size (dimensions) of the resource bit, the type of the resource bit, the number of parallel resource bits, etc.

[0048] In the embodiment of the present application, the service feature may be a service identifier related to the service, etc.

[0049] In an embodiment of the present application, the information processing device can obtain object characteristics, multiple groups of resource bit characteristics and multiple groups of business characteristics of the target object in a database; it can also obtain object characteristics, multiple groups of resource bit characteristics and multiple groups of business characteristics of the target object from other devices; the specific manner in which the information processing device obtains the object characteristics, multiple groups of resource bit characteristics and multiple groups of business characteristics of the target object can be determined according to actual conditions, and the embodiment of the present application does not limit this.

[0050] In an embodiment of the present application, the process of obtaining the object characteristics of the target object includes: obtaining the historical operation behavior information of the target object in the target mall and the product usage information when the target object uses the credit consumption product; and determining the historical operation behavior information and the product usage information as the object characteristics.

[0051] It should be noted that the target mall is a mall application that displays multiple resource locations; and the credit consumption product is a product in the mall application that is associated with multiple resource locations.

[0052] In the embodiment of the present application, the historical operation behavior information may include behavior characteristics such as browsing, adding to cart, and ordering on the target mall's app. The product usage information may be the number of times a user has used a credit consumption product, the number of times a user has made repayments, and other usage characteristics of a credit consumption product.

[0053] For example, the product usage information may be usage characteristics of the credit consumption product, such as the number of times the user uses the credit consumption product and the number of times the user repays the product.

[0054] In the embodiment of the present application, object characteristics, multiple groups of resource position characteristics and multiple groups of business characteristics affect the determination result of the target resource position. The reasons include: First, the user's behavior on the mall reflects the user's stickiness to the products in the mall. The higher the frequency of using the mall, the higher the user's stickiness to the mall app, and the stronger the conversion tendency to the mall's products; secondly, the transaction behavior characteristics of the user on credit consumption products reflect the strength of the user's financial attributes. The more frequently the user trades on credit consumption products, the stronger the financial attributes, and the higher the user's conversion tendency to cash loans, consumer loans and other products (i.e., credit consumption products) under the mall. Finally, the closer the exposure position of the resource position is to the middle and the larger the display position of the resource position is, the more likely the user is to pay attention and click to convert. Therefore, both user characteristics and resource position characteristics are factors that affect whether the user converts, and there is a correlation.

[0055] S102: Inputting object features, multiple groups of resource position features, and multiple groups of business features into a conversion rate prediction model to obtain multiple conversion rates of the target object under multiple businesses and multiple resource positions, respectively.

[0056] In an embodiment of the present application, after the information processing device obtains the object characteristics of the target object, multiple groups of resource position characteristics of multiple resource positions, and multiple groups of business characteristics of multiple businesses, the object characteristics, multiple groups of resource position characteristics, and multiple groups of business characteristics can be input into the conversion rate prediction model to obtain multiple conversion rates of the target object under multiple businesses and multiple resource positions respectively.

[0057] In an embodiment of the present application, an information processing device inputs object features, multiple groups of resource position features, and multiple groups of business features into a conversion rate prediction model to obtain multiple conversion rates of a target object under multiple businesses and multiple resource positions, respectively, including: inputting object features into a multi-layer first neural network of the conversion rate prediction model to obtain object feature mapping parameters; inputting multiple groups of resource position features into a representation network of the conversion rate prediction model to obtain multiple groups of resource position feature mapping parameters; splicing the object feature mapping parameters, multiple groups of resource position feature mapping parameters, and multiple groups of business features to obtain multiple spliced ​​features; and inputting multiple spliced ​​features into a multi-layer second neural network of the conversion rate prediction model to obtain multiple conversion rates.

[0058] In an embodiment of the present application, the conversion rate prediction model includes a multi-layer first neural network, a representation network, a concat layer (i.e., a splicing layer), and an output layer, wherein the output layer also includes a multi-layer second neural network.

[0059] In an embodiment of the present application, the concat layer splices the object feature mapping parameters, the multiple groups of resource bit feature mapping parameters and the multiple groups of business features to obtain multiple spliced ​​features. Specifically, the concat layer first splices the object feature mapping parameters and the multiple groups of resource bit feature mapping parameters to obtain multiple groups of initial splicing parameters, and then splices the multiple groups of initial splicing parameters with the multiple groups of business features to obtain multiple splicing features.

[0060] It should be noted that the multiple groups of initial splicing parameters are in matrix representation form, and the concat layer can convert the multiple groups of business features into matrix representation form, and then splice the multiple groups of initial splicing parameters in matrix representation form and the multiple groups of business features in matrix representation form to obtain multiple splicing features.

[0061] In the embodiment of the present application, a plurality of splicing features and a plurality of conversion rates are one-to-one, that is, one splicing feature corresponds to one conversion rate.

[0062] In an embodiment of the present application, the information processing device inputs multiple splicing features into the multi-layer second neural network of the conversion rate prediction model to obtain multiple conversion rates, including: using multiple second neural networks to determine multiple initial conversion parameters corresponding to the multiple splicing features; normalizing the multiple initial conversion parameters to obtain multiple conversion rates.

[0063] In the embodiment of the present application, the conversion rate prediction model also includes a sigmoid function, and the sigmoid function can be used to normalize multiple initial conversion parameters to obtain multiple conversion rates.

[0064] In an embodiment of the present application, the information processing device may also normalize multiple initial conversion parameters in other ways to obtain multiple conversion rates. The specific method of normalizing multiple initial conversion parameters to obtain multiple conversion rates can be determined based on actual conditions, and the embodiment of the present application does not limit this.

[0065] For example, Figure 3 As shown: the conversion rate prediction model includes a multi-layer neural network (a multi-layer first neural network), a representation network, a concat layer and an output layer (including a multi-layer second neural network). The object features and resource bit features input to the conversion rate prediction model can be expressed as X = (x u ,x d )(where X represents resource bit characteristics and object characteristics), specifically, is a p-dimensional user feature (where is the first dimension feature of the target object, is the second dimension feature of the target object, is the p-th dimension feature of the target object), that is, the object feature or multiple sets of sample object features, It can be expressed as a q-dimensional resource bit feature (where is the first dimension resource bit feature, is the second dimension resource bit feature, is the third-dimensional resource bit feature), that is, multiple sets of resource bit features. A multi-layer neural network can be composed of a multi-layer network structure. The specific structure of the network is not limited. In general, it is expressed as a nonlinear mapping relationship, expressed as f(·). Representation learning network (i.e. Figure 3 The representation network in the above example focuses on the mapping of the original feature space, which is represented by g(·). User features (user features can be represented by x u ) After a multi-layer neural network, the mapping x is obtained 1 =f 1 (x u ), the resource bit feature is mapped to x through the representation learning network 2 =g(x d), the two are concatenated in the middle of the network (concat layer) to obtain the intermediate layer feature x post =[x 1 ,x 2 ]. The intermediate layer features enter the output layer to predict the label and get the output: y pred =f 2 (x post ), that is, the output result of the output layer is obtained, which is also the output result of the conversion rate prediction model.

[0066] It should be noted that f 1 is a multi-layer neural network; x 1 For the multi-layer neural network x u The result obtained after processing. d is the resource bit feature; g is the representation network; x 2 To characterize the network pair x d The result obtained after processing.

[0067] It should be noted that p and q are natural numbers.

[0068] It should be noted that the concat layer converts x 1 and x 2 Put it together and you get x post , the conversion rate prediction model is based on x post Perform conversion rate prediction and finally obtain the output result of the conversion rate prediction model, that is, y pred Among them, f 2 For the conversion rate prediction model, based on x post The process of making conversion rate predictions.

[0069] S103: Determine target resource bits that match the target object under the target business according to the multiple conversion rates, where the target resource bits are part of the resource bits, and the target business is part of the multiple businesses.

[0070] In an embodiment of the present application, the information processing device inputs object characteristics, multiple groups of resource position characteristics and multiple groups of business characteristics into the conversion rate prediction model, and after obtaining multiple conversion rates of the target object under multiple businesses and multiple resource positions, the target resource position that matches the target object under the target business can be determined based on the multiple conversion rates.

[0071] In the embodiment of the present application, the target business may be a part of multiple businesses, or may be multiple businesses. The specific number may be determined based on actual conditions, and the example of the present application does not limit this.

[0072] In an embodiment of the present application, the process of an information processing device determining a target resource bit that matches a target object under a target business based on multiple conversion rates includes: determining a target conversion rate with the largest conversion rate value under the target business from multiple conversion rates; and determining the resource bit corresponding to the target conversion rate as the target resource bit that matches the target object under the target business.

[0073] In an embodiment of the present application, the process of an information processing device determining a target resource bit that matches a target object under a target business based on multiple conversion rates includes: obtaining multiple object conversion coefficients of the target object under multiple businesses; determining multiple conversion value parameters of the target object for multiple resource bits under multiple businesses respectively based on the multiple object conversion coefficients and the multiple conversion rates; determining a target conversion value parameter with the highest parameter value from the multiple conversion value parameters; and determining the resource bit corresponding to the target conversion value parameter as the target resource bit that matches the target object under the target business.

[0074] In the embodiment of the present application, the multiple object conversion coefficients may be conversion coefficients configured for target objects under multiple services, and the conversion coefficients may be adjusted at different stages.

[0075] In an embodiment of the present application, the information processing device can obtain multiple object conversion coefficients from information input by a user (i.e., the target object), or can obtain multiple object conversion coefficients from an object information storage center, or can obtain multiple object conversion coefficients through other methods. The specific method in which the information processing device obtains multiple object conversion coefficients can be determined based on actual conditions, and the embodiment of the present application does not limit this.

[0076] In the embodiment of the present application, multiple services correspond to multiple object conversion coefficients one by one, that is, one service corresponds to one object conversion coefficient.

[0077] In an embodiment of the present application, the process of determining multiple conversion value parameters of the target object for multiple resource positions under multiple services respectively according to multiple object conversion coefficients and multiple conversion rates can be as follows: determining a first group of conversion rates under a first service among multiple services from multiple conversion rates, determining the product of the first group of conversion rates and the first object conversion coefficient under the first service to obtain a first group of conversion value parameters; determining a second group of conversion rates under a second service among multiple services from multiple conversion rates, determining the product of the second group of conversion rates and the second object conversion coefficient under the second service to obtain a second group of conversion value parameters;…; determining a last group of conversion rates under a last service among multiple services from multiple conversion rates, determining the product of the last group of conversion rates and the last object conversion coefficient under the last service to obtain a last group of conversion value parameters; determining the first group of conversion value parameters, the second group of conversion value parameters,…, the last group of conversion value parameters as multiple conversion value parameters.

[0078] It should be noted that the multiple resource bits include a first group of resource bits, a second group of resource bits, ..., and a last group of resource bits. The multiple object conversion coefficients include a first object conversion coefficient, a second object conversion coefficient, ..., and a last object conversion coefficient.

[0079] In the embodiment of the present application, for the recommended business line set, the conversion rate prediction model is used to traverse and obtain the conversion scores (conversion rates) of the user (target object) under n business lines, respectively: score 0 ,score 1 ,…,score n , the corresponding prediction result obtained by using the conversion rate prediction model is P(y=1|x,v 0 ),P(y=1|x,v 1 ),…,P(y=1|x,v n ), and combined with the LTV value (object conversion coefficient) provided by the business to determine the user value conversion score (conversion value parameter) as shown in formula (1):

[0080] weigh_score i =ltv i *score i (1)

[0081] Among them, ltv i The user value that a certain user can bring to the business line (target business) for a given conversion, that is, the conversion coefficient of the user under the target business, also known as the object conversion coefficient. i is the object conversion coefficient under the i-th business; score i is the conversion rate of the user under the i-th business; weigh_score i is the conversion value parameter of the user under the i-th business.

[0082] It should be noted that for a single user, the business line with the highest value conversion score is recommended, and the relevant business text link is displayed in the designated resource position.

[0083] In an embodiment of the present application, after the information processing device determines the target resource bit that matches the target object under the target business based on multiple conversion rates, it will also determine the target group resource bit characteristics corresponding to the target resource bit among multiple groups of resource bit characteristics; and display the target resource bit to the target object based on the target group resource bit characteristics.

[0084] In the embodiment of the present application, the target resource bits are some of the resource bits among the multiple resource bits.

[0085] It can be understood that the information processing device obtains the object characteristics of the target object, multiple groups of resource position characteristics of multiple resource positions, and multiple groups of business characteristics of multiple businesses, and uses the conversion rate prediction model to determine multiple conversion rates of the target object under multiple businesses and multiple resource positions respectively according to the object characteristics, multiple groups of resource position characteristics, and multiple groups of business characteristics; determines the target resource position of the target object under the target business according to the multiple conversion rates, so that the target resource position is the resource position that the target object most expects to operate under the target business, thereby improving the accuracy of determining the resource position in the core page, thereby improving the conversion rate of users under the resource position.

[0086] The present application embodiment provides a model training method, a model training method is applied to a model training device, Figure 4 A flow chart of a model training method provided in an embodiment of the present application, such as Figure 4 As shown, the model training method may include:

[0087] S201. Acquire multiple groups of sample object features of multiple sample objects, multiple groups of sample resource bit features corresponding to multiple sample resource bits, multiple groups of sample service features corresponding to multiple sample services, and multiple sample labels.

[0088] A model training method provided in an embodiment of the present application is suitable for the scenario of training a conversion rate prediction model.

[0089] In the embodiments of the present application, the model training device can be implemented in various forms. For example, the model training device described in the present application can include devices such as mobile phones, cameras, tablet computers, laptop computers, PDAs, portable media players (PMPs), navigation devices, wearable devices, smart bracelets, pedometers, and devices such as digital TVs, desktop computers, and servers.

[0090] In the embodiment of the present application, multiple sample objects correspond to multiple groups of sample object features one-to-one, that is, one sample object corresponds to one group of sample object features. Multiple sample resource bits correspond to multiple groups of sample resource bit features one-to-one, that is, one sample resource bit corresponds to one group of sample resource bit features. Multiple sample services correspond to multiple groups of sample service features one-to-one, that is, one sample service corresponds to one sample service feature.

[0091] In the embodiment of the present application, the sample resource bit characteristics include the location of the sample resource bit on the core display page, the size (dimensions) of the sample resource bit, the type of the sample resource bit, the number of parallel sample resource bits, etc.

[0092] In the embodiment of the present application, the sample service feature may be a service identifier related to the sample service, etc.

[0093] In an embodiment of the present application, the information processing device can obtain multiple groups of sample object characteristics, multiple groups of sample resource bit characteristics, multiple groups of sample business characteristics and multiple sample tags in a database; it can also obtain multiple groups of sample object characteristics, multiple groups of sample resource bit characteristics, multiple groups of sample business characteristics and multiple sample tags from other devices; the specific information processing device obtains multiple groups of sample object characteristics, multiple groups of sample resource bit characteristics, multiple groups of sample business characteristics and multiple sample tags according to actual conditions, and the embodiment of the present application is not limited to this.

[0094] In an embodiment of the present application, the process of obtaining multiple groups of sample object features of multiple sample objects includes: obtaining multiple groups of historical operation behavior information of multiple sample objects in a target mall and multiple groups of product usage information when multiple sample objects use credit consumption products; determining the multiple groups of historical operation behavior information and the multiple groups of product usage information as multiple groups of sample object features.

[0095] In the embodiment of the present application, each input sample of the model (initial conversion rate prediction model) is a combination of a user (sample object) and a resource position (sample resource position), the feature is the total feature composed of user features (sample object features) and resource position features (sample resource position features), and the label is whether the user has been converted or the user has not been converted. For example, if the resource position displayed by the user is a banner showing the words "Log in to the financial app to get a discount of 10.99", then if the user clicks the banner and logs in to the financial app, the label is 1, otherwise it is 0. Divide the training set and test set from the historical data set, and train and evaluate the model.

[0096] S202: Input multiple groups of sample object features, multiple groups of sample resource location features, and multiple groups of sample business features into an initial conversion rate prediction model to obtain multiple sample output parameters.

[0097] In an embodiment of the present application, after the model training device obtains multiple groups of sample object features of multiple sample objects, multiple groups of sample resource bit features corresponding to multiple sample resource bits, multiple groups of sample business features corresponding to multiple sample businesses, and multiple sample labels, the multiple groups of sample object features, the multiple groups of sample resource bit features, and the multiple groups of sample business features are input into the initial conversion rate prediction model to obtain multiple sample output parameters.

[0098] In an embodiment of the present application, a model training device inputs multiple groups of sample object features, multiple groups of sample resource bit features, and multiple groups of sample business features into an initial conversion rate prediction model to obtain multiple sample output parameters, including: inputting multiple groups of sample object features into a multi-layer first initial neural network of the initial conversion rate prediction model to obtain multiple groups of sample object feature mapping parameters; inputting multiple groups of sample resource bit features into an initial representation network of the initial conversion rate prediction model to obtain multiple groups of output resource bit feature mapping parameters; inputting multiple groups of sample object feature mapping parameters, multiple groups of output resource bit feature mapping parameters, and multiple groups of sample business features into a multi-layer second initial neural network of the initial conversion rate prediction model to obtain multiple output conversion rates; and using multiple output conversion rates and multiple groups of output resource bit feature mapping parameters as multiple sample output parameters.

[0099] In the embodiment of the present application, multiple groups of sample object features correspond to multiple groups of sample object feature mapping parameters one by one, that is, one group of sample object features corresponds to one group of sample object feature mapping parameters. Multiple groups of sample resource bit features correspond to multiple groups of output resource bit feature mapping parameters one by one, that is, one group of sample resource bit features corresponds to one group of output resource bit feature mapping parameters.

[0100] In the embodiment of the present application, the multi-layer first initial neural network is composed of a multi-layer network structure (the specific structure of the network can be determined according to actual conditions, and the embodiment of the present application is not limited to this), which is generally expressed as a nonlinear mapping relationship.

[0101] In the embodiment of the present application, the initial representation network may be a trained representation network or a representation network to be trained. The specific representation network may be determined according to actual conditions, and the embodiment of the present application does not limit this.

[0102] In the embodiment of the present application, the initial characterization network may be replaced by other networks having the functions of the initial characterization network, or the functions of the initial characterization network may be implemented by other methods.

[0103] In an embodiment of the present application, a model training device inputs multiple sets of sample object feature mapping parameters, multiple sets of output resource bit feature mapping parameters, and multiple sets of sample business features into a multi-layer second initial neural network of an initial conversion rate prediction model to obtain multiple output conversion rates, including: splicing multiple sets of sample object feature mapping parameters, multiple sets of output resource bit feature mapping parameters, and multiple sets of sample business features to obtain multiple sample splicing features; inputting multiple sample splicing features into the multi-layer second initial neural network of the initial conversion rate prediction model to obtain multiple output conversion rates.

[0104] In an embodiment of the present application, the initial conversion rate prediction model includes a multi-layer first initial neural network, an initial representation network, an initial concat layer, and an initial output layer, wherein the initial output layer also includes a multi-layer second initial neural network.

[0105] In an embodiment of the present application, the initial concat layer splices multiple groups of sample object feature mapping parameters, multiple groups of output resource bit feature mapping parameters, and multiple groups of sample business features to obtain multiple sample splicing features. Specifically, the initial concat layer first splices multiple groups of sample object feature mapping parameters with multiple groups of output resource bit feature mapping parameters to obtain multiple groups of initial sample splicing parameters; then splices multiple groups of initial sample splicing parameters with multiple groups of sample business features to obtain multiple sample splicing features.

[0106] In an embodiment of the present application, the initial output layer also includes a sigmoid function, and multiple sample splicing features can be input into the multi-layer second initial neural network of the initial conversion rate prediction model to obtain multiple initial output conversion rates, and then the multiple initial output conversion rates are normalized using the sigmoid function to obtain multiple output conversion rates.

[0107] S203, determining a loss value of the initial conversion rate prediction model according to multiple sample output parameters and multiple sample labels; the loss value includes a cross entropy loss, a reconstruction loss, and a regularization loss;

[0108] In an embodiment of the present application, the model training device inputs multiple groups of sample object features, multiple groups of sample resource location features, and multiple groups of sample business features into the initial conversion rate prediction model, and after obtaining multiple sample output parameters, determines the loss value of the initial conversion rate prediction model based on the multiple sample output parameters and the multiple sample labels.

[0109] In an embodiment of the present application, a model training device determines a loss value of an initial conversion rate prediction model based on multiple sample output parameters and multiple sample labels, including: obtaining a gradient update parameter of the initial conversion rate prediction model; determining a regularization loss of the initial conversion rate prediction model based on the gradient update parameter; obtaining multiple groups of sample resource bit feature mapping parameters and multiple sample conversion rates from multiple sample labels; determining a cross entropy loss of the initial conversion rate prediction model based on multiple sample conversion rates and multiple output conversion rates; determining a reconstruction loss of the initial characterization network based on multiple groups of output resource bit feature mapping parameters and multiple groups of sample resource bit features; and determining a loss value based on the regularization loss, the cross entropy loss, and the reconstruction loss.

[0110] In an embodiment of the present application, the model training device determines the process of the loss value according to the regularization loss, the cross entropy loss and the reconstruction loss, including: obtaining the first weight coefficient of the reconstruction loss participating in the model update and the second weight coefficient of the regularization loss participating in the model update; using the first weight coefficient to update the reconstruction loss to obtain the updated reconstruction loss; using the second weight coefficient to update the regularization loss to obtain the updated regularization loss; fusing the updated reconstruction loss, the updated regularization loss and the cross entropy loss to obtain the loss value.

[0111] In an embodiment of the present application, the first weight coefficient may be a coefficient configured in the model training device, or a coefficient transmitted to the model training device from other devices, or a coefficient obtained by the model training device in other ways. The specific way in which the model training device obtains the first weight coefficient may be determined based on actual conditions, and the embodiment of the present application does not limit this.

[0112] In an embodiment of the present application, the second weight coefficient may be a coefficient configured in the model training device, or a coefficient transmitted to the model training device from other devices, or a coefficient obtained by the model training device in other ways. The specific way in which the model training device obtains the second weight coefficient may be determined based on actual conditions, and the embodiment of the present application does not limit this.

[0113] In an embodiment of the present application, the reconstruction loss is updated using the first weight coefficient to obtain the updated reconstruction loss by determining the product of the first weight coefficient and the reconstruction loss to obtain the updated reconstruction loss.

[0114] In an embodiment of the present application, the regularized loss is updated using the second weight coefficient to obtain an updated regularized loss. The updated regularized loss can be obtained by determining the product of the second weight coefficient and the regularized loss.

[0115] In an embodiment of the present application, the updated reconstruction loss, the updated regularization loss and the cross entropy loss are fused to obtain the loss value. The loss value can be obtained by determining the sum of the updated reconstruction loss, the updated regularization loss and the cross entropy loss.

[0116] It can be understood that the cross entropy that characterizes the distribution before and after the network is added as a loss to the total loss function of the model, which improves the stability of the distribution of resource positions before and after the network.

[0117] In the embodiment of the present application, it is assumed that there are n types of business lines that can be diverted (multiple groups of sample business features), represented by v 1 ,v 2 ,v 3 ,…,v n For the i-th divertible business line (i-th sample business), the corresponding sample business feature vi The value of (sample business feature) is embedded and then concatenated with the intermediate layer features (the initial sample concatenation parameters obtained by concatenating the sample object feature mapping parameters and the output resource bit feature mapping parameters) to obtain the output layer (initial output layer) input x input =[x post ,v i ], the output layer input passes through a multi-layer neural network (multi-layer second initial neural network) f 3 Get the output x 3 =f 3 (x input ), and then through the sigmoid function, we get That is, P(y=1|x,v i ).

[0118] It should be noted that x post is the initial sample splicing parameter, v i is the sample business characteristic; x input The sample splicing features are obtained by splicing the initial sample splicing parameters and the sample business features.

[0119] It should be noted that f 3 is the second initial neural network with multiple layers; x 3 is the multi-layer second initial neural network for x input The result obtained after processing.

[0120] It should be noted that is the sigmoid function on x 3 The result obtained after processing. P is the probability, x represents the sample object characteristics and sample resource bit characteristics, v i is the sample service feature of the i-th sample service, y=1 indicates that the sample object transforms the sample resource bit, P(y=1|x,v i ) is the probability of the sample object transforming the sample resource bit under the conditions of the sample object characteristics, the sample resource bit characteristics, and the sample service characteristics.

[0121] The loss function of the initial conversion rate prediction model consists of three parts, as shown in formula (2), including the cross entropy loss (loss 0 ), reconstruction loss 1 ) and regularization loss 2 ).

[0122] L = loss 0 + α loss 1 + βloss 2 (2)

[0123] Among them, α and β are control loss respectively. 1 、loss 2 The weights involved in updating the overall parameters of the model (i.e., the first weight coefficient and the second weight coefficient); the loss function is L.

[0124] It should be noted that loss 0 It is the sum of the cross entropy between the model's predicted conversion probability and the sample's true label on n samples. This loss mainly measures the model's overall learning of the label, as shown in formula (3):

[0125]

[0126] in, is the model's predicted value for the ith sample (i.e., the output conversion rate corresponding to the ith sample resource position determined by the model), is the true label value of the sample (i.e., sample conversion rate), ln is a logarithmic function (log) with e as the base. Wherein, k refers to the task, if K=1, there is only one task, if K is greater than or equal to 2, there are multiple tasks.

[0127] It should be noted that loss 1 The loss is the reconstruction loss of the representation network part in the model, which mainly measures the original input x of the representation network d With reconstructed output The similarity of the distribution of is as shown in formula (4):

[0128]

[0129] Among them, ||.|| is the Euclidean distance. The smaller the reconstruction error, the more likely it is that the decoder can successfully restore the representation features extracted by the encoder, that is, the representation network can well restore the representation features extracted by the encoder.

[0130] It should be noted that loss 2 The loss is the regularization loss, which is the sum of the squares of all trainable parameters in the model. The regularization loss can effectively prevent the model from overfitting, as shown in formula (5);

[0131]

[0132] Among them, β i is the i-th parameter in the model that is updated with gradients, and N is the total number of parameters in the model that can be updated with gradients.

[0133] S204, when the loss value is greater than a preset loss threshold, continue to train the initial conversion rate prediction model using multiple sets of sample object features, multiple sets of sample resource location features, multiple sets of sample business features, and multiple sample labels to obtain a training model;

[0134] In an embodiment of the present application, after the model training device determines the loss value of the initial conversion rate prediction model based on multiple sample output parameters and multiple sample labels, when the loss value is greater than a preset loss threshold, the initial conversion rate prediction model is continued to be trained using multiple groups of sample object features, multiple groups of sample resource location features, multiple groups of sample business features and multiple sample labels to obtain a training model.

[0135] In an embodiment of the present application, the preset loss threshold may be a threshold configured in the model training device, or a threshold transmitted to the model training device by other devices, or a threshold obtained by the model training device in other ways. The specific way in which the model training device obtains the preset loss threshold may be determined based on actual conditions, and the embodiment of the present application does not limit this.

[0136] S205: When the training model loss value corresponding to the training model is less than or equal to a preset loss threshold, the training model is determined as a conversion rate prediction model.

[0137] In an embodiment of the present application, the model training device continues to train the initial conversion rate prediction model with multiple groups of sample object features, multiple groups of sample resource location features, multiple groups of sample business features and multiple sample labels. After obtaining the training model, when the training model loss value corresponding to the training model is less than or equal to the preset loss threshold, the training model is determined as the conversion rate prediction model.

[0138] It can be understood that the model training device obtains multiple groups of sample object features, multiple groups of sample resource position features, multiple groups of sample business features and multiple sample labels, and uses multiple groups of sample object features, multiple groups of sample resource position features, multiple groups of sample business features and multiple sample labels to train the initial conversion rate prediction model, and calculates the loss value of the model, and the loss value is a loss value including cross entropy loss, reconstruction loss and regularization loss, uses cross entropy loss to adjust the overall learning of the model on the label, uses reconstruction loss to measure the similarity of the distribution between the original input and the reconstructed output of the representation network, and uses regularization term loss to prevent model overfitting, thereby improving the accuracy of the final conversion rate prediction model, and using the high-accuracy conversion rate prediction model to determine multiple conversion rates of the target object under multiple businesses and multiple resource positions according to object features, multiple groups of resource position features and multiple groups of business features; determines the target resource position of the target object under the target business according to the multiple conversion rates, so that the target resource position is the resource position that the target object most expects to operate under the target business, thereby improving the accuracy of determining the resource position in the core page, thereby improving the conversion rate of the user under the resource position.

[0139] Based on the same inventive concept as the above-mentioned information processing method, the embodiment of the present application provides an information processing device 1 corresponding to an information processing method; Figure 5 A schematic diagram of the structure of an information processing device provided in an embodiment of the present application Figure 1 , the information processing device 1 may include:

[0140] A first acquisition unit 11 is used to acquire object characteristics of a target object, multiple groups of resource bit characteristics of multiple resource bits, and multiple groups of service characteristics of multiple services;

[0141] The first input unit 12 is used to input the object feature, the multiple groups of resource position features and the multiple groups of business features into a conversion rate prediction model to obtain multiple conversion rates of the target object under the multiple businesses and the multiple resource positions respectively;

[0142] The first determination unit 13 is used to determine target resource bits matching the target object under the target business according to the multiple conversion rates, the target resource bits are part of the resource bits, and the target business is part of the multiple businesses.

[0143] In some embodiments of the present application, the device further comprises a splicing unit;

[0144] The first input unit 12 is used to input the object feature into the multi-layer first neural network of the conversion rate prediction model to obtain object feature mapping parameters; input the multiple sets of resource bit features into the representation network of the conversion rate prediction model to obtain multiple sets of resource bit feature mapping parameters; input the multiple splicing features into the multi-layer second neural network of the conversion rate prediction model to obtain the multiple conversion rates;

[0145] The splicing unit is used to splice the object feature mapping parameters, the multiple groups of resource bit feature mapping parameters and the multiple groups of service features to obtain multiple splicing features.

[0146] In some embodiments of the present application, the device further comprises a processing unit;

[0147] The first determining unit 13 is used to determine a plurality of initial transformation parameters corresponding to the plurality of splicing features using the plurality of second neural networks;

[0148] The processing unit is used to perform normalization processing on the multiple initial conversion parameters to obtain the multiple conversion rates.

[0149] In some embodiments of the present application, the first acquisition unit 11 is used to acquire multiple object conversion coefficients of the target object under the multiple services;

[0150] The first determination unit 13 is used to determine multiple conversion value parameters of the target object for the multiple resource positions under the multiple services respectively according to the multiple object conversion coefficients and the multiple conversion rates; determine the target conversion value parameter with the highest parameter value from the multiple conversion value parameters; and determine the resource position corresponding to the target conversion value parameter as the target resource position matching the target object under the target service.

[0151] In some embodiments of the present application, the first determination unit 13 is used to determine a target conversion rate having the largest conversion rate value under the target business from the multiple conversion rates; and determine the resource bit corresponding to the target conversion rate as the target resource bit matching the target object under the target business.

[0152] In some embodiments of the present application, the device further comprises a display unit;

[0153] The first determining unit 13 is used to determine the target group resource bit feature corresponding to the target resource bit from the multiple groups of resource bit features;

[0154] The display unit is used to display the target resource location to the target object according to the resource location characteristics of the target group.

[0155] It should be noted that, in actual applications, the above-mentioned first acquisition unit 11, the first input unit 12 and the first determination unit 13 can be implemented by the first processor 14 on the information processing device 1, specifically a CPU (Central Processing Unit), an MPU (Microprocessor Unit), a DSP (Digital Signal Processing) or a field programmable gate array (FPGA); the above-mentioned data storage can be implemented by the first memory 15 on the information processing device 1.

[0156] The present application also provides an information processing device 1, such as Figure 6 As shown, the information processing device 1 includes: a first processor 14, a first memory 15 and a first communication bus 16. The first memory 15 communicates with the first processor 14 through the first communication bus 16. The first memory 15 stores a program executable by the first processor 14. When the program is executed, the information processing method described above is executed by the first processor 14.

[0157] In practical applications, the first memory 15 may be a volatile memory, such as a random access memory (RAM); or a non-volatile memory, such as a read-only memory (ROM), a flash memory, a hard disk (HDD) or a solid-state drive (SSD); or a combination of the above types of memories, and provide instructions and data to the first processor 14.

[0158] An embodiment of the present application provides a computer-readable storage medium having a computer program thereon, and when the program is executed by the first processor 14, the information processing method as described above is implemented.

[0159] It can be understood that the information processing device obtains the object characteristics of the target object, multiple groups of resource position characteristics of multiple resource positions, and multiple groups of business characteristics of multiple businesses, and uses the conversion rate prediction model to determine multiple conversion rates of the target object under multiple businesses and multiple resource positions respectively according to the object characteristics, multiple groups of resource position characteristics, and multiple groups of business characteristics; determines the target resource position of the target object under the target business according to the multiple conversion rates, so that the target resource position is the resource position that the target object most expects to operate under the target business, thereby improving the accuracy of determining the resource position in the core page, thereby improving the conversion rate of users under the resource position.

[0160] Based on the same inventive concept as the above-mentioned model training method, the embodiment of the present application provides a model training device 2, corresponding to a model training method; Figure 7 A schematic diagram of the structure of a model training device provided in an embodiment of the present application Figure 1 , the model training device 2 may include:

[0161] A second acquisition unit 21 is used to acquire multiple groups of sample object features of multiple sample objects, multiple groups of sample resource bit features corresponding to multiple sample resource bits, multiple groups of sample service features corresponding to multiple sample services, and multiple sample labels;

[0162] The second input unit 22 is used to input the multiple groups of sample object features, the multiple groups of sample resource location features, and the multiple groups of sample business features into the initial conversion rate prediction model to obtain multiple sample output parameters;

[0163] A second determining unit 23 is used to determine a loss value of the initial conversion rate prediction model according to the multiple sample output parameters and the multiple sample labels; the loss value includes a cross entropy loss, a reconstruction loss and a regularization loss; when the training model loss value corresponding to the training model is less than or equal to the preset loss threshold, the training model is determined as the conversion rate prediction model;

[0164] The training unit 24 is used to continue training the initial conversion rate prediction model using the multiple groups of sample object features, the multiple groups of sample resource location features, the multiple groups of sample business features and the multiple sample labels to obtain a training model when the loss value is greater than a preset loss threshold.

[0165] In some embodiments of the present application, the second input unit 22 is used to input the multiple groups of sample object features into the multi-layer first initial neural network of the initial conversion rate prediction model to obtain multiple groups of sample object feature mapping parameters; input the multiple groups of sample resource bit features into the initial characterization network of the initial conversion rate prediction model to obtain multiple groups of output resource bit feature mapping parameters; input the multiple groups of sample object feature mapping parameters, the multiple groups of output resource bit feature mapping parameters and the multiple groups of sample business features into the multi-layer second initial neural network of the initial conversion rate prediction model to obtain the multiple output conversion rates;

[0166] The second determining unit 23 is configured to use the multiple output conversion rates and the multiple groups of output resource bit feature mapping parameters as the multiple sample output parameters.

[0167] In some embodiments of the present application, the second acquisition unit 21 is used to acquire the gradient update parameters of the initial conversion rate prediction model; acquire multiple groups of sample resource bit feature mapping parameters and multiple sample conversion rates from the multiple sample labels;

[0168] The second determination unit 23 is used to determine the regularization loss of the initial conversion rate prediction model according to the gradient update parameter; determine the cross entropy loss of the initial conversion rate prediction model according to the multiple sample conversion rates and the multiple output conversion rates; determine the reconstruction loss of the initial characterization network according to the multiple groups of output resource bit feature mapping parameters and the multiple groups of sample resource bit features; determine the loss value according to the regularization loss, the cross entropy loss and the reconstruction loss.

[0169] In some embodiments of the present application, the apparatus further includes an updating unit and a fusion unit;

[0170] The second acquisition unit 21 is used to acquire a first weight coefficient of the reconstruction loss participating in the model update and a second weight coefficient of the regularization loss participating in the model update;

[0171] The updating unit is used to update the reconstruction loss using the first weight coefficient to obtain an updated reconstruction loss; and update the regularization loss using the second weight coefficient to obtain an updated regularization loss;

[0172] The fusion unit is used to fuse the updated reconstruction loss, the updated regularization loss and the cross entropy loss to obtain the loss value.

[0173] It should be noted that, in actual applications, the above-mentioned second acquisition unit 21, second input unit 22, second determination unit 23 and training unit 24 can be implemented by the second processor 25 on the model training device 2, specifically a CPU (Central Processing Unit), MPU (Microprocessor Unit), DSP (Digital Signal Processing) or a field programmable gate array (FPGA), etc.; the above-mentioned data storage can be implemented by the second memory 26 on the model training device 2.

[0174] The present application embodiment also provides a model training device 2, such as Figure 8 As shown, the model training device 2 includes: a second processor 25, a second memory 26 and a second communication bus 27. The second memory 26 communicates with the second processor 25 via the second communication bus 27. The second memory 26 stores a program executable by the second processor 25. When the program is executed, the model training method described above is executed by the second processor 25.

[0175] In practical applications, the second memory 26 may be a volatile memory, such as a random access memory (RAM); or a non-volatile memory, such as a read-only memory (ROM), a flash memory, a hard disk drive (HDD) or a solid-state drive (SSD); or a combination of the above types of memories, and provide instructions and data to the second processor 25.

[0176] An embodiment of the present application provides a computer-readable storage medium having a computer program thereon, which, when executed by the second processor 25, implements the model training method as described above.

[0177] It can be understood that the model training device obtains multiple groups of sample object features, multiple groups of sample resource position features, multiple groups of sample business features and multiple sample labels, and uses multiple groups of sample object features, multiple groups of sample resource position features, multiple groups of sample business features and multiple sample labels to train the initial conversion rate prediction model, and calculates the loss value of the model, and the loss value is a loss value including cross entropy loss, reconstruction loss and regularization loss, uses cross entropy loss to adjust the overall learning of the model on the label, uses reconstruction loss to measure the similarity of the distribution between the original input and the reconstructed output of the representation network, and uses regularization term loss to prevent the model from overfitting, thereby improving the accuracy of the final conversion rate prediction model, and uses the high-accuracy conversion rate prediction model to determine multiple conversion rates of the target object under multiple businesses and multiple resource positions according to object features, multiple groups of resource position features and multiple groups of business features; determines the target resource position of the target object under the target business according to the multiple conversion rates, so that the target resource position is the resource position that the target object most expects to operate under the target business, thereby improving the accuracy of determining the resource position in the core page, thereby improving the conversion rate of the user under the resource position.

[0178] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of hardware embodiments, software embodiments, or embodiments in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage and optical storage, etc.) that contain computer-usable program code.

[0179] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0180] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1A function specified in one or more boxes.

[0181] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0182] The above description is only a preferred embodiment of the present application and is not intended to limit the protection scope of the present application.

Claims

1. An information processing method, It is characterized in that The method comprises: Acquire object characteristics of a target object, multiple groups of resource bit characteristics of multiple resource bits, and multiple groups of service characteristics of multiple services; Inputting the object features, the multiple groups of resource position features and the multiple groups of business features into a conversion rate prediction model to obtain multiple conversion rates of the target object under the multiple businesses and the multiple resource positions respectively; A target resource bit matching the target object under a target business is determined according to the multiple conversion rates, the target resource bit is part of the resource bits, and the target business is part of the multiple businesses.

2. The method according to claim 1, It is characterized in that The step of inputting the object feature, the multiple groups of resource position features and the multiple groups of business features into a conversion rate prediction model to obtain multiple conversion rates of the target object under the multiple businesses and the multiple resource positions respectively includes: Inputting the object features into the multi-layer first neural network of the conversion rate prediction model to obtain object feature mapping parameters; Inputting the multiple sets of resource bit features into the characterization network of the conversion rate prediction model to obtain multiple sets of resource bit feature mapping parameters; Splicing the object feature mapping parameters, the multiple groups of resource bit feature mapping parameters and the multiple groups of service features to obtain multiple splicing features; The multiple splicing features are input into the multi-layer second neural network of the conversion rate prediction model to obtain the multiple conversion rates.

3. The method according to claim 2, It is characterized in that The step of inputting the plurality of splicing features into the multi-layer second neural network of the conversion rate prediction model to obtain the plurality of conversion rates comprises: Determining a plurality of initial transformation parameters corresponding to the plurality of splicing features using the plurality of second neural networks; The multiple initial conversion parameters are normalized to obtain the multiple conversion rates.

4. The method according to claim 1, It is characterized in that The determining, according to the multiple conversion rates, a target resource position matching the target object under the target business includes: Acquire multiple object conversion coefficients of the target object under the multiple services; Determining, according to the multiple object conversion coefficients and the multiple conversion rates, multiple conversion value parameters of the target object for the multiple resource locations under the multiple services respectively; Determine a target conversion value parameter with the highest parameter value from the multiple conversion value parameters; The resource bit corresponding to the target conversion value parameter is determined as the target resource bit matching the target object under the target business.

5. The method according to claim 1, It is characterized in that The determining, according to the multiple conversion rates, a target resource position matching the target object under the target business includes: Determining a target conversion rate having a maximum conversion rate value under the target business from the multiple conversion rates; The resource bit corresponding to the target conversion rate is determined as the target resource bit matching the target object under the target business.

6. The method according to claim 1, It is characterized in that After determining the target resource position matching the target object under the target business according to the multiple conversion rates, the method further includes: Determine, among the multiple groups of resource bit characteristics, a target group resource bit characteristic corresponding to the target resource bit; The target resource bits are displayed to the target object according to the target group resource bit characteristics.

7. A model training method, It is characterized in that The method comprises: Acquire multiple groups of sample object features of multiple sample objects, multiple groups of sample resource bit features corresponding to multiple sample resource bits, multiple groups of sample service features corresponding to multiple sample services, and multiple sample labels; Inputting the multiple groups of sample object features, the multiple groups of sample resource location features, and the multiple groups of sample business features into an initial conversion rate prediction model to obtain multiple sample output parameters; Determining a loss value of the initial conversion rate prediction model according to the multiple sample output parameters and the multiple sample labels; the loss value includes a cross entropy loss, a reconstruction loss, and a regularization loss; When the loss value is greater than a preset loss threshold, the initial conversion rate prediction model is continuously trained using the multiple groups of sample object features, the multiple groups of sample resource location features, the multiple groups of sample service features and the multiple sample labels to obtain a training model; When the training model loss value corresponding to the training model is less than or equal to the preset loss threshold, the training model is determined as the conversion rate prediction model.

8. The method according to claim 7, It is characterized in that The inputting the multiple groups of sample object features, the multiple groups of sample resource location features, and the multiple groups of sample business features into the initial conversion rate prediction model to obtain multiple sample output parameters includes: Inputting the plurality of groups of sample object features into the multi-layer first initial neural network of the initial conversion rate prediction model to obtain a plurality of groups of sample object feature mapping parameters; Inputting the plurality of groups of sample resource bit features into the initial characterization network of the initial conversion rate prediction model to obtain a plurality of groups of output resource bit feature mapping parameters; Inputting the multiple groups of sample object feature mapping parameters, the multiple groups of output resource bit feature mapping parameters and the multiple groups of sample business features into the multi-layer second initial neural network of the initial conversion rate prediction model to obtain the multiple output conversion rates; The multiple output conversion rates and the multiple groups of output resource bit feature mapping parameters are used as the multiple sample output parameters.

9. The method according to claim 8, It is characterized in that The step of determining the loss value of the initial conversion rate prediction model according to the multiple sample output parameters and the multiple sample labels includes: Obtaining a gradient update parameter of the initial conversion rate prediction model; Determining the regularization loss of the initial conversion rate prediction model according to the gradient update parameter; Acquire multiple groups of sample resource bit feature mapping parameters and multiple sample conversion rates from the multiple sample tags; Determining a cross entropy loss of the initial conversion rate prediction model according to the multiple sample conversion rates and the multiple output conversion rates; Determining a reconstruction loss of the initial characterization network according to the multiple sets of output resource bit feature mapping parameters and the multiple sets of sample resource bit features; The loss value is determined according to the regularization loss, the cross entropy loss and the reconstruction loss.

10. The method according to claim 9, It is characterized in that The determining the loss value according to the regularization loss, the cross entropy loss and the reconstruction loss includes: Obtaining a first weight coefficient of the reconstruction loss participating in the model update and a second weight coefficient of the regularization loss participating in the model update; Updating the reconstruction loss using the first weight coefficient to obtain an updated reconstruction loss; Updating the regularized loss using the second weight coefficient to obtain an updated regularized loss; The updated reconstruction loss, the updated regularization loss and the cross entropy loss are fused to obtain the loss value.

11. An information processing device, It is characterized in that The device comprises: A first acquisition unit, used to acquire an object feature of a target object, multiple groups of resource bit features of multiple resource bits, and multiple groups of service features of multiple services; A first input unit is used to input the object feature, the multiple groups of resource position features and the multiple groups of business features into a conversion rate prediction model to obtain multiple conversion rates of the target object under the multiple businesses and the multiple resource positions respectively; A first determining unit is used to determine target resource bits matching the target object under a target business according to the multiple conversion rates, wherein the target resource bits are part of the resource bits, and the target business is part of the multiple businesses.

12. An information processing device, It is characterized in that The device comprises: A first memory, a first processor and a first communication bus, wherein the first memory communicates with the first processor via the first communication bus, the first memory stores an information processing program executable by the first processor, and when the information processing program is executed, the method described in any one of claims 1 to 6 is performed by the first processor.

13. A model training device, It is characterized in that The device comprises: A second acquisition unit, configured to acquire multiple groups of sample object features of multiple sample objects, multiple groups of sample resource bit features corresponding to multiple sample resource bits, multiple groups of sample service features corresponding to multiple sample services, and multiple sample labels; A second input unit is used to input the multiple groups of sample object features, the multiple groups of sample resource location features, and the multiple groups of sample business features into an initial conversion rate prediction model to obtain multiple sample output parameters; A second determination unit is used to determine a loss value of the initial conversion rate prediction model according to the multiple sample output parameters and the multiple sample labels; the loss value includes a cross entropy loss, a reconstruction loss and a regularization loss; when the training model loss value corresponding to the training model is less than or equal to the preset loss threshold, the training model is determined as the conversion rate prediction model; A training unit is used to continue training the initial conversion rate prediction model using the multiple groups of sample object features, the multiple groups of sample resource location features, the multiple groups of sample business features and the multiple sample labels to obtain a training model when the loss value is greater than a preset loss threshold.

14. A model training device, It is characterized in that The device comprises: A second memory, a second processor and a second communication bus, wherein the second memory communicates with the second processor via the second communication bus, and the second memory stores information processing programs executable by the second processor. When the information processing program is executed, the method described in any one of claims 7 to 10 is performed by the second processor.

15. A storage medium having a computer program stored thereon, applied to an information processing device and a model training device, It is characterized in that When the computer program is executed by a first processor, the method described in any one of claims 1 to 6 is implemented; when the computer program is executed by a second processor, the method described in any one of claims 7 to 10 is implemented.

Images