Commodity module sorting method and device, electronic equipment and storage medium

Abstract

The invention discloses a commodity module sorting method and device, electronic equipment and a storage medium, when a commodity module sorting request sent by a user is received, user feature data and a commodity module candidate set are obtained, and for each candidate commodity module in the commodity module candidate set, the commodity module sorting request is sent to the user; estimating a commodity conversion rate corresponding to each commodity in a commodity set of the candidate commodity module based on the commodity module sorting model and the user feature data, and selecting a target commodity conversion rate with a high conversion rate to determine a corresponding target commodity; and estimating the module conversion rate of the candidate commodity module to which each target commodity belongs, selecting the target module conversion rate with high conversion rate to determine the corresponding target commodity module, and outputting each target commodity module and the target commodity contained in each target commodity module. According to the invention, the commodity module and the commodity row preferred by the user can be displayed at the front position, and the accuracy of obtaining the commodity preferred by the user is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of module sorting, and more specifically, to a commodity module sorting method, device, electronic device and storage medium. Background Art

[0002] On the mall page, there are usually multiple product modules displayed to users at the same time, such as limited-time purchase module, hardcore subsidy module, daily new product module, today's hot product module, etc. Different product modules will display products of different subcategories for users to choose from.

[0003] At present, the positions of different product modules in the mall page are generally fixed manually. However, different users have different preferences and usage frequencies for different product modules. For example, some users often use the limited-time flash sale module, while some users like to click on the products displayed in the sub-category of a certain product module.

[0004] Therefore, how to provide a product module sorting method that can display different product modules according to different user preferences, improve user experience, and increase the platform's click-through rate and final transaction amount to maximize the platform's revenue has become a technical problem that technical personnel in this field urgently need to solve. Summary of the invention

[0005] In view of this, the present invention discloses a commodity module sorting method, device, electronic device and storage medium, so as to display different commodity modules according to different user preferences, improve user experience, and increase platform click rate and final transaction amount, so as to maximize platform revenue.

[0006] A commodity module sorting method, comprising:

[0007] When receiving a commodity module sorting request sent by a user, obtaining user feature data and a commodity module candidate set, wherein the user feature data includes: a current module behavior sequence, a current commodity behavior sequence, and a current context feature including a user basic attribute feature;

[0008] For each candidate commodity module in the commodity module candidate set, estimating a commodity conversion rate corresponding to each commodity in the commodity set of the candidate commodity module based on a pre-trained commodity module ranking model and the user feature data;

[0009] Selecting a first preset number of target commodity conversion rates from all the commodity conversion rates, and determining a target commodity corresponding to each of the target commodity conversion rates, wherein each of the target commodity conversion rates is higher than a conversion rate of non-target commodities that have not been selected;

[0010] Based on the commodity module ranking model and the user characteristic data, estimating the module conversion rate of the candidate commodity module to which each target commodity belongs;

[0011] Selecting a second preset number of target module conversion rates from all the module conversion rates, and determining a target commodity module corresponding to each target module conversion rate, wherein each target module conversion rate is higher than a non-target module conversion rate that is not selected;

[0012] The target commodity modules and the target commodities contained in each target commodity module are output.

[0013] Optionally, for each candidate commodity module in the commodity module candidate set, estimating the commodity conversion rate corresponding to each commodity in the commodity set of the candidate commodity module based on a pre-trained commodity module ranking model and the user feature data includes:

[0014] Extracting corresponding module attribute features for each candidate commodity module in the commodity module candidate set;

[0015] Extracting corresponding commodity attribute features for each commodity in the commodity set of each candidate commodity module;

[0016] Each of the commodity attribute features and its corresponding module attribute features are combined with the user feature data and input into the commodity module ranking model to estimate the commodity conversion rate corresponding to each commodity in the commodity set of the candidate commodity module.

[0017] Optionally, the estimating the module conversion rate of the candidate commodity module to which each target commodity belongs based on the commodity module ranking model and the user characteristic data includes:

[0018] The product attribute features corresponding to each of the target products, the module attribute features of the candidate product module to which the target product belongs, and the user feature data are combined and input into the product module ranking model again to estimate the module conversion rate of the candidate product module to which each of the target products belongs.

[0019] Optionally, the training process of the commodity module ranking model includes:

[0020] Obtain user historical behavior log data within a preset time period;

[0021] Cleaning the user historical behavior log data to obtain a user behavior sequence acting on a commodity module and commodities in the commodity module, wherein the user behavior sequence includes: a module behavior sequence and a commodity behavior sequence;

[0022] Extracting module attribute features, product attribute features and context features from the user historical behavior log data;

[0023] Obtaining a product conversion rate and a module conversion rate based on the user's click behavior and conversion behavior in the user's historical behavior log data;

[0024] The module behavior sequence, the commodity behavior sequence, the module attribute features, the commodity attribute features and the context features are used as sample data, and the commodity conversion rate and the module conversion rate are used as sample labels to train the model network and obtain the commodity module ranking model.

[0025] Optionally, the module behavior sequence, the commodity behavior sequence, the module attribute feature, the commodity attribute feature and the context feature are used as sample data, the commodity conversion rate and the module conversion rate are used as sample labels, and the model network is trained to obtain the commodity module ranking model, including:

[0026] The module behavior sequence and the product behavior sequence are converted through the embedding layer of the model network to obtain embedding vector features;

[0027] Processing the product attribute features, the module attribute features and the context features through the hidden layer of the model network to obtain a feature set;

[0028] Concatenate the embedding vector feature and the feature set to obtain a target concatenated feature;

[0029] The target splicing features are processed by multiple activation layers and normalized exponential functions in sequence to obtain a model estimation output result, wherein the model estimation output result includes: module conversion rate and product conversion rate.

[0030] Optionally, it also includes:

[0031] The loss function used in the commodity module ranking model is corrected based on a correction factor related to the commodity module size to obtain a commodity module ranking model after the loss function is corrected.

[0032] A commodity module sorting device, comprising:

[0033] An acquisition unit, configured to acquire user feature data and a candidate set of commodity modules when receiving a commodity module sorting request sent by a user, wherein the user feature data includes: a current module behavior sequence, a current commodity behavior sequence, and a current context feature including a user basic attribute feature;

[0034] a commodity conversion rate determination unit, configured to estimate, for each candidate commodity module in the commodity module candidate set, a commodity conversion rate corresponding to each commodity in the commodity set of the candidate commodity module based on a pre-trained commodity module ranking model and the user feature data;

[0035] a target commodity determination unit, configured to select a first preset number of target commodity conversion rates from all the commodity conversion rates, and determine a target commodity corresponding to each of the target commodity conversion rates, wherein each of the target commodity conversion rates is higher than a conversion rate of non-target commodities that have not been selected;

[0036] A module conversion rate determination unit, configured to estimate the module conversion rate of the candidate commodity module to which each target commodity belongs based on the commodity module ranking model and the user characteristic data;

[0037] a target commodity module determination unit, configured to select a second preset number of target module conversion rates from all the module conversion rates, and determine a target commodity module corresponding to each of the target module conversion rates, wherein each of the target module conversion rates is higher than a conversion rate of non-target modules that have not been selected;

[0038] The module output unit is used to output each of the target commodity modules and the target commodities contained in each of the target commodity modules.

[0039] Optionally, it also includes:

[0040] A model training module, used for training the commodity module sorting model;

[0041] The model training module is specifically used for:

[0042] Obtain user historical behavior log data within a preset time period;

[0043] Cleaning the user historical behavior log data to obtain a user behavior sequence acting on a commodity module and commodities in the commodity module, wherein the user behavior sequence includes: a module behavior sequence and a commodity behavior sequence;

[0044] Extracting module attribute features, product attribute features and context features from the user historical behavior log data;

[0045] Obtaining a product conversion rate and a module conversion rate based on the user's click behavior and conversion behavior in the user's historical behavior log data;

[0046] The module behavior sequence, the commodity behavior sequence, the module attribute features, the commodity attribute features and the context features are used as sample data, and the commodity conversion rate and the module conversion rate are used as sample labels to train the model network and obtain the commodity module ranking model.

[0047] A computer storage medium stores at least one instruction, and when the at least one instruction is executed by a processor, the above-mentioned commodity module sorting method is implemented.

[0048] An electronic device, comprising: a memory and a processor;

[0049] The memory is used to store at least one instruction;

[0050] The processor is used to execute the at least one instruction to implement the above-mentioned commodity module sorting method.

[0051] It can be seen from the above technical scheme that the present invention discloses a commodity module sorting method, device, electronic device and storage medium. When a commodity module sorting request sent by a user is received, user feature data and a commodity module candidate set are obtained. For each candidate commodity module in the commodity module candidate set, the commodity conversion rate corresponding to each commodity in the commodity set of the candidate commodity module is estimated based on a pre-trained commodity module sorting model and user feature data. A first preset number of high-conversion target commodity conversion rates are selected from all the commodity conversion rates, and the target commodity corresponding to each target commodity conversion rate is determined. Based on the commodity module sorting model and user feature data, the module conversion rate of the candidate commodity module to which each target commodity belongs is estimated, a second preset number of high-conversion target module conversion rates are selected from all the module conversion rates, and the target commodity module corresponding to each target module conversion rate is determined, and each target commodity module and the target commodity contained in each target commodity module are output. The present invention determines the commodity conversion rate and the module conversion rate based on the commodity module ranking model and the user characteristic data, firstly screens out each target commodity with a high commodity conversion rate, then screens out each target commodity module with a high module conversion rate from each commodity module to which the target commodity belongs, and outputs and displays each target commodity module and the target commodities contained therein, thereby realizing the display of different commodity modules according to the preferences of different users, that is, the commodity modules and commodities preferred by the users can be displayed in the front positions, thereby improving the accuracy rate of obtaining the commodities preferred by the users and the commodity purchase rate, thereby improving the click rate of the platform and the final transaction amount, and maximizing the platform revenue. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the disclosed drawings without paying creative work.

[0053] Figure 1 A flow chart of a commodity module sorting method disclosed in an embodiment of the present invention;

[0054] Figure 2 A schematic diagram of a user module and a product interest portrait generated based on a user behavior sequence disclosed in an embodiment of the present invention;

[0055] Figure 3 A schematic diagram of a model network structure disclosed in an embodiment of the present invention;

[0056] Figure 4 A schematic diagram of the structure of a commodity module sorting device disclosed in an embodiment of the present invention;

[0057] Figure 5 The present invention is a schematic diagram of the structure of an electronic device disclosed in an embodiment of the present invention. DETAILED DESCRIPTION

[0058] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0059] The embodiment of the present invention discloses a commodity module sorting method, device, electronic device and storage medium. The commodity conversion rate and module conversion rate are determined based on a commodity module sorting model and user characteristic data. First, each target commodity with a high commodity conversion rate is screened out. Then, from each commodity module to which the target commodity belongs, each target commodity module with a high module conversion rate is screened out. Each target commodity module and the target commodity contained therein are output and displayed, thereby realizing the display of different commodity modules according to different user preferences. That is, the commodity modules and commodities preferred by the user can be displayed in a front position, thereby improving the accuracy of obtaining the commodity preferred by the user and the commodity purchase rate, thereby improving the platform click rate and the final transaction amount, and maximizing the platform revenue.

[0060] See also Figure 1 , a flowchart of a commodity module sorting method disclosed in an embodiment of the present invention, the method comprising:

[0061] Step S101: when receiving a commodity module sorting request sent by a user, obtaining user feature data and a commodity module candidate set.

[0062] The commodity module sorting request includes a commodity module candidate set, and the commodity module candidate set includes each commodity module to be sorted and commodities to be sorted contained in each commodity module.

[0063] The sorting of product modules determines user interests and usage habits. The platform database has accumulated a large amount of user behavior log data, which can obtain users' recent behaviors, including but not limited to exposure, clicks, purchases, interests, etc. User behavior acts on product modules or products on product modules. Combined with rich user behavior log data and the corresponding label system, user interests can be represented by behavior, including user interest in product modules and products under modules.

[0064] Based on this, when receiving a product module sorting request sent by a user, the application can obtain user feature data from the platform database according to the user's unique identifier. The user feature data in the application is a multi-dimensional vector, which may specifically include: current module behavior sequence, current product behavior sequence and current context feature.

[0065] The current module behavior sequence includes the module itself and the commodities on the module.

[0066] The current product behavior sequence includes the product itself and the module to which the product belongs.

[0067] The current context features include user basic attribute features, which mainly refer to attributes that describe the user's basic situation and background. These features are usually static, that is, they will not change in a short period of time, such as user interests, hobbies, usage habits, etc.

[0068] The current module behavior sequence combined with product information can effectively describe the product modules in which users like a certain product combination state, while the current product behavior sequence, including the module information, can better describe the product modules that users are interested in and the products under the product modules.

[0069] Step S102: for each candidate commodity module in the commodity module candidate set, estimate the commodity conversion rate corresponding to each commodity in the commodity set of the candidate commodity module based on the pre-trained commodity module ranking model and the user feature data.

[0070] The product module ranking model in the present application uses module behavior sequence, product behavior sequence, module attribute features, product attribute features and context features as sample data, and takes product conversion rate and module conversion rate as sample labels to train the model network.

[0071] Product conversion rate is a key e-commerce and marketing metric that measures the proportion of users who actually complete a purchase among users who visit a product page within a specific time period. This proportion is usually expressed as a percentage. The calculation formula is:

[0072] Product conversion rate = (number of customers who make purchases / number of visitors who reach the product page) × 100%.

[0073] The conversion rate of a product can directly reflect the attractiveness of the product, the effectiveness of the sales strategy, and the smoothness of the user's purchase process. A higher conversion rate usually means that the product is popular with users, the sales strategy is appropriate, and the purchase process is relatively simple, which can attract more users to complete the purchase.

[0074] The product module candidate set includes multiple candidate product modules, and the product combination of each candidate product module includes multiple products. This application estimates the product conversion rate corresponding to each product in the product set of each candidate product module based on the product module ranking model and user feature data.

[0075] Step S103: selecting a first preset number of target commodity conversion rates from all the commodity conversion rates, and determining a target commodity corresponding to each of the target commodity conversion rates.

[0076] The conversion rate of each target product is higher than the conversion rate of non-target products that are not selected.

[0077] In actual applications, after estimating the product conversion rate corresponding to each product under each candidate product module in the product module candidate set, all product conversion rates can be sorted from high to low, and the first preset number of target product conversion rates with the highest rankings are selected, and the products corresponding to each target product conversion rate are used as target products. The target products are also the products displayed in the product module, and the unselected products are defined as non-target products.

[0078] The value of the first preset number depends on actual needs and is not limited in the present invention.

[0079] Step S104: estimating the module conversion rate of the candidate commodity module to which each target commodity belongs based on the commodity module ranking model and the user characteristic data.

[0080] Module conversion rate is a term specific to e-commerce or online sales. It refers to the proportion of visitors converted into actual buyers in a certain product module of an e-commerce platform (such as product details page, product list page, etc.). This ratio is usually expressed as a percentage, reflecting the ability of the product module to attract visitors and prompt them to make purchases.

[0081] The calculation formula of module conversion rate is as follows:

[0082] Module conversion rate = (number of customers who completed purchases through the product module during a specific time period / number of visitors who visited the product module during the same time period) × 100%.

[0083] Step S105: selecting a second preset number of target module conversion rates from all the module conversion rates, and determining a target commodity module corresponding to each of the target module conversion rates.

[0084] The conversion rate of each target module is higher than the conversion rate of the non-target modules that are not selected.

[0085] In actual applications, after estimating the module conversion rate corresponding to each product module in the product candidate set, all module conversion rates can be sorted from high to low, and the second preset number of target module conversion rates with the top rankings can be selected, and the product modules corresponding to each target module conversion rate can be determined as target product modules, and the product modules that are not selected are defined as non-target product modules.

[0086] Step S106: output each of the target product modules and the target products contained in each of the target product modules.

[0087] Among them, each target commodity module output by this application and the target commodities contained in each target commodity module are also commodity modules recommended according to user preferences and the commodities contained in the commodity modules.

[0088] In summary, the present invention discloses a commodity module sorting method. When a commodity module sorting request sent by a user is received, user feature data and a commodity module candidate set are obtained. For each candidate commodity module in the commodity module candidate set, a commodity conversion rate corresponding to each commodity in the commodity set of the candidate commodity module is estimated based on a pre-trained commodity module sorting model and user feature data. A first preset number of target commodity conversion rates with high conversion rates are selected from all the commodity conversion rates, and a target commodity corresponding to each target commodity conversion rate is determined. Based on the commodity module sorting model and user feature data, a module conversion rate of the candidate commodity module to which each target commodity belongs is estimated. A second preset number of target module conversion rates with high conversion rates are selected from all the module conversion rates, and a target commodity module corresponding to each target module conversion rate is determined. Each target commodity module and the target commodity contained in each target commodity module are output. The present invention determines the commodity conversion rate and the module conversion rate based on the commodity module ranking model and the user characteristic data, firstly screens out each target commodity with a high commodity conversion rate, then screens out each target commodity module with a high module conversion rate from each commodity module to which the target commodity belongs, and outputs and displays each target commodity module and the target commodities contained therein, thereby realizing the display of different commodity modules according to the preferences of different users, that is, the commodity modules and commodities preferred by the users can be displayed in the front positions, thereby improving the accuracy rate of obtaining the commodities preferred by the users and the commodity purchase rate, thereby improving the click rate of the platform and the final transaction amount, and maximizing the platform revenue.

[0089] In one embodiment, step S102 may specifically include:

[0090] (1) Extracting corresponding module attribute features for each candidate commodity module in the commodity module candidate set.

[0091] Module attribute features are key information that describe product characteristics. They are crucial for users to select products and for the management of e-commerce platforms. Product attributes usually include product name, product price, product brand, etc.

[0092] (2) Extracting corresponding product attribute features for each product in the product set of each candidate product module.

[0093] Product attribute characteristics are used to describe and distinguish product features, performance, functions, appearance, specifications, etc. They provide consumers with a basis for selection and comparison.

[0094] (3) Combining each of the product attribute features and its corresponding module attribute features with the user feature data and inputting them into the product module ranking model to estimate the product conversion rate corresponding to each product in the product set of the candidate product module.

[0095] This application combines the module attribute features of the candidate product module and the product attribute features corresponding to each product in the product set of the candidate product module one by one, and combines them with user feature data, and inputs them into the product module sorting model for processing, so as to obtain the product conversion rate corresponding to each product.

[0096] In one embodiment, step S104 may specifically include:

[0097] The commodity attribute features corresponding to each of the target commodities, the module attribute features of the candidate commodity module to which the target commodity belongs, and the user feature data are combined and input into the commodity module ranking model again to estimate the module conversion rate of the candidate commodity module to which each of the target commodities belongs.

[0098] Among them, when determining the module conversion rate, the product attribute features input into the product module ranking model are the product attribute features of each target product included in the candidate product module.

[0099] In one embodiment, the present application also discloses a training process of a commodity module ranking model, which is as follows:

[0100] (1) Obtain user historical behavior log data within a preset time period.

[0101] Among them, user historical behavior log data refers to the user's operation behavior on the product modules and products of the platform within a preset time period, which may include: exposure, clicks, add to cart, order completion, collection, etc.

[0102] The value of the preset time period is determined according to actual needs, such as one month, and the present invention is not limited here.

[0103] (2) Cleaning the user's historical behavior log data to obtain a user behavior sequence that acts on a product module and the products in the product module.

[0104] The user behavior sequence includes: a module behavior sequence and a commodity behavior sequence, and interest portraits of user modules and commodities can be generated according to the user behavior sequence.

[0105] This application cleans the user's historical behavior log data, so that the cleaned data can more accurately reflect the user's interests, preferences and behavior patterns, which helps to build a more detailed user portrait.

[0106] See also Figure 2 , a schematic diagram of a user module and a commodity interest portrait generated based on a user behavior sequence disclosed in an embodiment of the present invention, wherein the module behavior sequence includes the module itself and the commodities on the module, Figure 2In the example, the module behavior sequence involves three operation behaviors: exposure, click, and conversion. Product module 1 includes: product A1, product B1…product T1, product module 2 includes: product A2, product B2…product T2,…, and product module N includes: product AN, product BN…product TN.

[0107] The product behavior sequence includes five operations: exposure, click, add to cart, favorite and conversion. Products A and B belong to product module 1, product C belongs to product module 2, and product T belongs to product module N.

[0108] Among them, the module behavior sequence combined with the product information can effectively describe the product modules in the product combination state that the user likes, and the current product behavior sequence, including the module information, can better describe the product modules that the user is interested in and the products under the product module.

[0109] (3) Extracting module attribute features, product attribute features, and context features from the user's historical behavior log data.

[0110] Module attribute features are key information that describe product characteristics. They are crucial for users to select products and for the management of e-commerce platforms. Product attributes usually include product name, product price, product brand, etc.

[0111] Product attribute characteristics are used to describe and distinguish product features, performance, functions, appearance, specifications, etc. They provide consumers with a basis for selection and comparison.

[0112] Contextual features include user basic attribute features. User basic attribute features mainly refer to attributes that describe the user's basic situation and background. These features are usually static, that is, they will not change in a short period of time, such as user interests, hobbies, usage habits, etc.

[0113] (4) Obtaining a product conversion rate and a module conversion rate based on the user's click behavior and conversion behavior in the user's historical behavior log data.

[0114] The calculation formula for the module conversion rate in this application is as follows:

[0115] Module conversion rate = (number of customers who completed purchases through the product module during a specific time period / number of visitors who visited the product module during the same time period) × 100%.

[0116] The formula for calculating product conversion rate is as follows:

[0117] Product conversion rate = (number of customers who make purchases / number of visitors who reach the product page) × 100%.

[0118] (5) Taking the module behavior sequence, the product behavior sequence, the module attribute features, the product attribute features and the context features as sample data, and taking the product conversion rate and the module conversion rate as sample labels, the model network is trained to obtain the product module ranking model.

[0119] For details, see Figure 3 The model network structure diagram is shown in the figure. The training process of the model network is as follows:

[0120] (1) The module behavior sequence and product behavior sequence are converted through the embedding layer of the model network to obtain the embedding vector features.

[0121] As an important component in deep learning models, the embedding layer has multiple functions such as dimensionality reduction, expression of word relationships, introduction of additional information, and optimization effects. In the word vector space, the embedding layer can represent each word as a vector of fixed dimension. This vector not only contains the semantic information of the word, but also reflects the relationship between words. This relationship is learned by the neural network during the training process, making the positions of words with similar semantics closer in the vector space.

[0122] (2) The product attribute features, the module attribute features, and the context features are all processed through the hidden layer of the model network to obtain a feature set.

[0123] The hidden layer in the model network refers to the middle layer between the input layer and the output layer. The hidden layer provides the neural network with powerful expression and learning capabilities through features extraction and transformation, nonlinear mapping, learning complex patterns, information synthesis and global analysis. At the same time, the number of hidden layers and the number of neurons in each layer are also one of the key factors affecting the performance of the model.

[0124] In this application, product attribute features, module attribute features and context features are all processed through a hidden layer to obtain a feature set.

[0125] (3) concatenating the embedding vector feature and the feature set to obtain a target concatenated feature;

[0126] (4) The target splicing features are processed by multiple activation layers and normalized exponential functions in sequence to obtain the model estimation output result.

[0127] The model estimation output results include module conversion rate and product conversion rate.

[0128] Multi-layer activation layers solve the problem of insufficient expression ability of linear models by introducing nonlinear activation functions. Each activation layer of the multi-layer activation layer can be regarded as a nonlinear transformation of the feature space. This transformation can transform the original linear inseparable problem into a linear separable problem, thereby greatly increasing the model's ability to solve complex problems.

[0129] The normalized exponential function can "compress" a K-dimensional vector z containing any real number into another K-dimensional real vector σ(z), so that the range of each element is between (0, 1) and the sum of all elements is 1. This is actually a gradient logarithmic normalization of a finite discrete probability distribution.

[0130] The present application obtains the model estimation output results, including module conversion rate and product conversion rate, by sequentially processing the target splicing features through multiple activation layers and normalized exponential functions.

[0131] It is understandable that different product modules differ in product form and display size. Larger modules have certain advantages in conversion efficiency, but they consume more display resources. Therefore, in order to ensure fairness in module conversion calculations and avoid the model being biased towards larger modules, the product module sorting model needs to be corrected. During the model correction process, the module size is introduced to correct the conversion rate. For modules with larger sizes and larger product images, a certain degree of demotion is given to ensure that the exposure of each module is fair and reasonable.

[0132] In one embodiment, the commodity module sorting method may further include:

[0133] The loss function used in the commodity module ranking model is corrected based on a correction factor related to the commodity module size, so as to obtain a commodity module ranking model after the loss function is corrected.

[0134] The present application corrects the loss function used in the commodity module ranking model based on a modified positive factor related to the commodity model size, and can obtain a commodity module ranking model after the loss function is corrected.

[0135] In practice, this goal is achieved by improving the neural network loss function. On the basis of the original loss function, the product of the estimated conversion rate minus the actual conversion rate and the module size term minus function is added. For example, the neural network in the present invention adopts the focal loss loss function, plus the bias term, and the final loss function is as follows:

[0136] Loss = -α(1-p) γ ylog(p)-(1-α)p γ (1-y)log(1-p)+[y(py)] / (Mxa *My b );

[0137] In the formula, α, γ, a, and b are all common parameters, and the optimal value is obtained by adjusting the parameters. y is the label value in the sample, the positive sample is 1, and the negative sample is 0. p is the model's estimated click rate, Mx is the module length, and My is the module width. The larger the area, the larger the loss function. When the positive sample estimate is too small, the second term of the loss function will decrease, achieving the effect of adjusting the module area bias.

[0138] Corresponding to the above method embodiment, the present invention also discloses a commodity module sorting device.

[0139] See also Figure 4 , a schematic diagram of the structure of a commodity module sorting device disclosed in an embodiment of the present invention, the device may include:

[0140] The acquisition unit 201 is used to acquire user feature data and a commodity module candidate set when receiving a commodity module sorting request sent by a user.

[0141] The commodity module sorting request includes a commodity module candidate set, and the commodity module candidate set includes each commodity module to be sorted and commodities to be sorted contained in each commodity module.

[0142] The sorting of product modules determines user interests and usage habits. The platform database has accumulated a large amount of user behavior log data, which can obtain users' recent behaviors, including but not limited to exposure, clicks, purchases, interests, etc. User behavior acts on product modules or products on product modules. Combined with rich user behavior log data and the corresponding label system, user interests can be represented by behavior, including user interest in product modules and products under modules.

[0143] Based on this, when receiving a product module sorting request sent by a user, the application can obtain user feature data from the platform database according to the user's unique identifier. The user feature data in the application is a multi-dimensional vector, which may specifically include: current module behavior sequence, current product behavior sequence and current context feature.

[0144] The current module behavior sequence includes the module itself and the commodities on the module.

[0145] The current product behavior sequence includes the product itself and the module to which the product belongs.

[0146] The current context features include user basic attribute features, which mainly refer to attributes that describe the user's basic situation and background. These features are usually static, that is, they will not change in a short period of time, such as user interests, hobbies, usage habits, etc.

[0147] The current module behavior sequence combined with product information can effectively describe the product modules in which users like a certain product combination state, while the current product behavior sequence, including the module information, can better describe the product modules that users are interested in and the products under the product modules.

[0148] A commodity conversion rate determination unit 202 is used to estimate, for each candidate commodity module in the commodity module candidate set, a commodity conversion rate corresponding to each commodity in the commodity set of the candidate commodity module based on a pre-trained commodity module ranking model and the user feature data;

[0149] The product module ranking model in the present application uses module behavior sequence, product behavior sequence, module attribute features, product attribute features and context features as sample data, and takes product conversion rate and module conversion rate as sample labels to train the model network.

[0150] Product conversion rate is a key e-commerce and marketing metric that measures the proportion of users who actually complete a purchase among users who visit a product page within a specific time period. This proportion is usually expressed as a percentage. The calculation formula is:

[0151] Product conversion rate = (number of customers who make purchases / number of visitors who reach the product page) × 100%.

[0152] The conversion rate of a product can directly reflect the attractiveness of the product, the effectiveness of the sales strategy, and the smoothness of the user's purchase process. A higher conversion rate usually means that the product is popular with users, the sales strategy is appropriate, and the purchase process is relatively simple, which can attract more users to complete the purchase.

[0153] The product module candidate set includes multiple candidate product modules, and the product combination of each candidate product module includes multiple products. This application estimates the product conversion rate corresponding to each product in the product set of each candidate product module based on the product module ranking model and user feature data.

[0154] The target product determination unit 203 is used to select a first preset number of target product conversion rates from all the product conversion rates, and determine the target product corresponding to each of the target product conversion rates, wherein each of the target product conversion rates is higher than the conversion rate of the non-target products that are not selected;

[0155] In actual applications, after estimating the product conversion rate corresponding to each product under each candidate product module in the product module candidate set, all product conversion rates can be sorted from high to low, and the first preset number of target product conversion rates with the highest rankings are selected, and the products corresponding to each target product conversion rate are used as target products. The target products are also the products displayed in the product module, and the unselected products are defined as non-target products.

[0156] The value of the first preset number depends on actual needs and is not limited in the present invention.

[0157] A module conversion rate determination unit 204, configured to estimate the module conversion rate of the candidate commodity module to which each target commodity belongs based on the commodity module ranking model and the user characteristic data;

[0158] Module conversion rate is a term specific to e-commerce or online sales. It refers to the proportion of visitors converted into actual buyers in a certain product module of an e-commerce platform (such as product details page, product list page, etc.). This ratio is usually expressed as a percentage, reflecting the ability of the product module to attract visitors and prompt them to make purchases.

[0159] The calculation formula of module conversion rate is as follows:

[0160] Module conversion rate = (number of customers who completed purchases through the product module during a specific time period / number of visitors who visited the product module during the same time period) × 100%.

[0161] The target commodity module determination unit 205 is used to select a second preset number of target module conversion rates from all the module conversion rates, and determine the target commodity module corresponding to each of the target module conversion rates, wherein each of the target module conversion rates is higher than the conversion rate of the non-target modules that are not selected;

[0162] In actual applications, after estimating the module conversion rate corresponding to each product module in the product candidate set, all module conversion rates can be sorted from high to low, and the second preset number of target module conversion rates with the top rankings can be selected, and the product modules corresponding to each target module conversion rate can be determined as target product modules, and the product modules that are not selected are defined as non-target product modules.

[0163] The module output unit 206 is used to output each of the target commodity modules and the target commodities included in each of the target commodity modules.

[0164] Among them, each target commodity module output by this application and the target commodities contained in each target commodity module are also commodity modules recommended according to user preferences and the commodities contained in the commodity modules.

[0165] In summary, the present invention discloses a commodity module sorting device. When a commodity module sorting request sent by a user is received, user feature data and a commodity module candidate set are obtained. For each candidate commodity module in the commodity module candidate set, the commodity conversion rate corresponding to each commodity in the commodity set of the candidate commodity module is estimated based on a pre-trained commodity module sorting model and user feature data. A first preset number of target commodity conversion rates with high conversion rates are selected from all the commodity conversion rates, and a target commodity corresponding to each target commodity conversion rate is determined. Based on the commodity module sorting model and user feature data, the module conversion rate of the candidate commodity module to which each target commodity belongs is estimated. A second preset number of target module conversion rates with high conversion rates are selected from all the module conversion rates, and a target commodity module corresponding to each target module conversion rate is determined. Each target commodity module and the target commodity contained in each target commodity module are output. The present invention determines the commodity conversion rate and the module conversion rate based on the commodity module ranking model and the user characteristic data, firstly screens out each target commodity with a high commodity conversion rate, then screens out each target commodity module with a high module conversion rate from each commodity module to which the target commodity belongs, and outputs and displays each target commodity module and the target commodities contained therein, thereby realizing the display of different commodity modules according to the preferences of different users, that is, the commodity modules and commodities preferred by the users can be displayed in the front positions, thereby improving the accuracy rate of obtaining the commodities preferred by the users and the commodity purchase rate, thereby improving the click rate of the platform and the final transaction amount, and maximizing the platform revenue.

[0166] In one embodiment, the commodity conversion rate determination unit 202 may be specifically used to:

[0167] Extracting corresponding module attribute features for each candidate commodity module in the commodity module candidate set;

[0168] Extracting corresponding commodity attribute features for each commodity in the commodity set of each candidate commodity module;

[0169] Each of the commodity attribute features and its corresponding module attribute features are combined with the user feature data and input into the commodity module ranking model to estimate the commodity conversion rate corresponding to each commodity in the commodity set of the candidate commodity module.

[0170] In one embodiment, the module conversion rate determination unit 204 may be specifically used to:

[0171] The product attribute features corresponding to each of the target products, the module attribute features of the candidate product module to which the target product belongs, and the user feature data are combined and input into the product module ranking model again to estimate the module conversion rate of the candidate product module to which each of the target products belongs.

[0172] In one embodiment, the commodity module sorting device may further include:

[0173] The model training module is used to train the commodity module sorting model.

[0174] The model training module can be specifically used for:

[0175] Obtain user historical behavior log data within a preset time period;

[0176] Cleaning the user historical behavior log data to obtain a user behavior sequence acting on a commodity module and commodities in the commodity module, wherein the user behavior sequence includes: a module behavior sequence and a commodity behavior sequence;

[0177] Extracting module attribute features, product attribute features and context features from the user historical behavior log data;

[0178] Obtaining a product conversion rate and a module conversion rate based on the user's click behavior and conversion behavior in the user's historical behavior log data;

[0179] The module behavior sequence, the commodity behavior sequence, the module attribute features, the commodity attribute features and the context features are used as sample data, and the commodity conversion rate and the module conversion rate are used as sample labels to train the model network and obtain the commodity module ranking model.

[0180] In one embodiment, the model training module may also be used to:

[0181] The module behavior sequence and the product behavior sequence are converted through the embedding layer of the model network to obtain embedding vector features;

[0182] Processing the product attribute features, the module attribute features and the context features through the hidden layer of the model network to obtain a feature set;

[0183] Concatenate the embedding vector feature and the feature set to obtain a target concatenated feature;

[0184] The target splicing features are processed by multiple activation layers and normalized exponential functions in sequence to obtain a model estimation output result, wherein the model estimation output result includes: module conversion rate and product conversion rate.

[0185] In one embodiment, the model training module may also be used to:

[0186] The loss function used in the commodity module ranking model is corrected based on a correction factor related to the commodity module size to obtain a commodity module ranking model after the loss function is corrected.

[0187] It should be noted that, for the specific working principles of each component in the device embodiment, please refer to the corresponding part of the method embodiment, which will not be repeated here.

[0188] Corresponding to the above embodiment, the present invention further discloses a computer storage medium, which stores at least one instruction. When the at least one instruction is executed by a processor, the steps shown in the embodiment of the commodity module sorting method are implemented.

[0189] Computer storage media may be tangible media that may contain or store programs for use by or in conjunction with an instruction execution system, device, or apparatus. Computer storage media may be machine-readable signal media or machine-readable storage media. Computer storage media may include, but are not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media may include electrical connections based on one or more lines, portable computer disks, hard disks, random access memories (RAM), read-only memories (ROM), erasable programmable read-only memories (EPROM or flash memory), optical fibers, portable compact disk read-only memories (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0190] Corresponding to the above embodiment, Figure 5 As shown, the present invention also provides an electronic device, which may include: a processor 1 and a memory 2;

[0191] The processor 1 and the memory 2 communicate with each other via a communication bus 3;

[0192] Processor 1, configured to execute at least one instruction;

[0193] Memory 2, used to store at least one instruction;

[0194] The processor 1 may be a central processing unit (CPU), or an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of the present invention.

[0195] The memory 2 may include a high-speed RAM memory, and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory.

[0196] The processor executes at least one instruction to implement the steps shown in the embodiment of the commodity module sorting method.

[0197] Finally, it should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0198] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0199] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A commodity module sorting method, characterized in that: include: When receiving a commodity module sorting request sent by a user, obtaining user feature data and a commodity module candidate set, wherein the user feature data includes: a current module behavior sequence, a current commodity behavior sequence, and a current context feature including a user basic attribute feature; For each candidate commodity module in the commodity module candidate set, estimating a commodity conversion rate corresponding to each commodity in the commodity set of the candidate commodity module based on a pre-trained commodity module ranking model and the user feature data; Selecting a first preset number of target commodity conversion rates from all the commodity conversion rates, and determining a target commodity corresponding to each of the target commodity conversion rates, wherein each of the target commodity conversion rates is higher than a conversion rate of non-target commodities that have not been selected; Based on the commodity module ranking model and the user characteristic data, estimating the module conversion rate of the candidate commodity module to which each target commodity belongs; Selecting a second preset number of target module conversion rates from all the module conversion rates, and determining a target commodity module corresponding to each target module conversion rate, wherein each target module conversion rate is higher than a non-target module conversion rate that is not selected; The target commodity modules and the target commodities contained in each target commodity module are output.

2. The commodity module sorting method according to claim 1, characterized in that: The estimating, for each candidate commodity module in the commodity module candidate set, a commodity conversion rate corresponding to each commodity in the commodity set of the candidate commodity module based on a pre-trained commodity module ranking model and the user feature data, includes: Extracting corresponding module attribute features for each candidate commodity module in the commodity module candidate set; Extracting corresponding commodity attribute features for each commodity in the commodity set of each candidate commodity module; Each of the commodity attribute features and its corresponding module attribute features are combined with the user feature data and input into the commodity module ranking model to estimate the commodity conversion rate corresponding to each commodity in the commodity set of the candidate commodity module.

3. The commodity module sorting method according to claim 1, characterized in that: The estimating the module conversion rate of the candidate commodity module to which each target commodity belongs based on the commodity module ranking model and the user characteristic data includes: The commodity attribute features corresponding to each of the target commodities, the module attribute features of the candidate commodity module to which the target commodity belongs, and the user feature data are combined and input into the commodity module ranking model again to estimate the module conversion rate of the candidate commodity module to which each of the target commodities belongs.

4. The commodity module sorting method according to any one of claims 1 to 3, characterized in that: The training process of the commodity module sorting model includes: Obtain user historical behavior log data within a preset time period; Cleaning the user historical behavior log data to obtain a user behavior sequence acting on a commodity module and commodities in the commodity module, wherein the user behavior sequence includes: a module behavior sequence and a commodity behavior sequence; Extracting module attribute features, product attribute features and context features from the user historical behavior log data; Obtaining a product conversion rate and a module conversion rate based on the user's click behavior and conversion behavior in the user's historical behavior log data; The module behavior sequence, the commodity behavior sequence, the module attribute features, the commodity attribute features and the context features are used as sample data, and the commodity conversion rate and the module conversion rate are used as sample labels to train the model network and obtain the commodity module ranking model.

5. The commodity module sorting method according to claim 4, characterized in that: The module behavior sequence, the commodity behavior sequence, the module attribute feature, the commodity attribute feature and the context feature are used as sample data, the commodity conversion rate and the module conversion rate are used as sample labels, and the model network is trained to obtain the commodity module ranking model, including: The module behavior sequence and the product behavior sequence are converted through the embedding layer of the model network to obtain embedding vector features; Processing the product attribute features, the module attribute features and the context features through the hidden layer of the model network to obtain a feature set; Concatenate the embedding vector feature and the feature set to obtain a target concatenated feature; The target splicing features are processed by multiple activation layers and normalized exponential functions in sequence to obtain a model estimation output result, wherein the model estimation output result includes: module conversion rate and product conversion rate.

6. The commodity module sorting method according to claim 4, characterized in that: Also includes: The loss function used in the commodity module ranking model is corrected based on a correction factor related to the commodity module size to obtain a commodity module ranking model after the loss function is corrected.

7. A commodity module sorting device, characterized in that: include: An acquisition unit, configured to acquire user feature data and a candidate set of commodity modules when receiving a commodity module sorting request sent by a user, wherein the user feature data includes: a current module behavior sequence, a current commodity behavior sequence, and a current context feature including a user basic attribute feature; a commodity conversion rate determination unit, configured to estimate, for each candidate commodity module in the commodity module candidate set, a commodity conversion rate corresponding to each commodity in the commodity set of the candidate commodity module based on a pre-trained commodity module ranking model and the user feature data; a target commodity determination unit, configured to select a first preset number of target commodity conversion rates from all the commodity conversion rates, and determine a target commodity corresponding to each of the target commodity conversion rates, wherein each of the target commodity conversion rates is higher than a conversion rate of non-target commodities that have not been selected; A module conversion rate determination unit, configured to estimate the module conversion rate of the candidate commodity module to which each target commodity belongs based on the commodity module ranking model and the user characteristic data; a target commodity module determination unit, configured to select a second preset number of target module conversion rates from all the module conversion rates, and determine a target commodity module corresponding to each of the target module conversion rates, wherein each of the target module conversion rates is higher than a conversion rate of non-target modules that have not been selected; The module output unit is used to output each of the target commodity modules and the target commodities contained in each of the target commodity modules.

8. The commodity module sorting device according to claim 7, characterized in that: Also includes: A model training module, used for training the commodity module sorting model; The model training module is specifically used for: Obtain user historical behavior log data within a preset time period; Cleaning the user historical behavior log data to obtain a user behavior sequence acting on a commodity module and commodities in the commodity module, wherein the user behavior sequence includes: a module behavior sequence and a commodity behavior sequence; Extracting module attribute features, product attribute features and context features from the user historical behavior log data; Obtaining a product conversion rate and a module conversion rate based on the user's click behavior and conversion behavior in the user's historical behavior log data; The module behavior sequence, the commodity behavior sequence, the module attribute features, the commodity attribute features and the context features are used as sample data, and the commodity conversion rate and the module conversion rate are used as sample labels to train the model network and obtain the commodity module ranking model.

9. A computer storage medium, characterized in that: The computer storage medium stores at least one instruction, and when the at least one instruction is executed by the processor, the commodity module sorting method according to any one of claims 1 to 6 is implemented.

10. An electronic device, characterized in that: The electronic device comprises: a memory and a processor; The memory is used to store at least one instruction; The processor is used to execute the at least one instruction to implement the commodity module sorting method according to any one of claims 1 to 6.

Images