A commodity module sorting method and device, electronic equipment and storage medium

Abstract

The application discloses a commodity module sorting method and device, electronic equipment and storage medium. When receiving a commodity module sorting request sent by a user, 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 commodity module sorting model and the user feature data, and a target commodity with a high conversion rate is selected to determine the corresponding target commodity. Based on the commodity module sorting model and the user feature data, the module conversion rate of each target commodity belonging to the candidate commodity module is estimated, and a target module conversion rate with a high conversion rate is selected to determine the corresponding target commodity module. The target commodity module and the target commodity included in each target commodity module are output. The application can make the commodity module and commodity preferred by the user be displayed in a position at the front, and improve the accuracy of obtaining the user's favorite commodity.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of module sorting, and more particularly to a commodity module sorting method and device, an electronic device and a storage medium. BACKGROUND

[0002] There are usually multiple commodity modules on a shopping mall page to display to users, such as a limited-time purchase module, a hardcore subsidy module, a daily new product module, a today's hot product module, etc. Different commodity modules display different sub-category commodities for users to select.

[0003] At present, the positions of different commodity modules in the shopping mall page are generally fixed by humans. However, different users have different preferences and use frequencies for different commodity modules, for example, some users often use the limited-time purchase module, and some users like to click on the commodities displayed by a sub-category of a commodity module.

[0004] Therefore, how to provide a commodity module sorting method that can display different commodity modules according to different user preferences, improve user experience, and increase platform click-through rate and final transaction volume to maximize platform revenue has become a technical problem that technicians in the field need to solve. SUMMARY

[0005] Therefore, the present application discloses a commodity module sorting method, device, electronic device and storage medium to display different commodity modules according to different user preferences, improve user experience, and increase platform click-through rate and final transaction volume to maximize platform revenue.

[0006] A commodity module sorting method comprises:

[0007] When a commodity module sorting request sent by a user is received, user feature data and a commodity module candidate set are obtained, wherein the user feature data comprises a current module behavior sequence, a current commodity behavior sequence, and a current context feature comprising user basic attribute features;

[0008] For each candidate commodity module in the commodity module candidate set, the conversion rate of each commodity in the commodity set of the candidate commodity module is estimated based on a pre-trained commodity module sorting model and the user feature data;

[0009] A first predetermined number of target commodity conversion rates are selected from all the commodity conversion rates, and a target commodity corresponding to each target commodity conversion rate is determined, wherein each target commodity conversion rate is higher than a non-target commodity conversion rate that is not selected;

[0010] estimate a module conversion rate of a candidate product module to which each of the target products belongs, based on the product module ranking model and the user feature data;

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

[0012] output each of the target product modules and the target products included in each of the target product modules.

[0013] Optionally, for each of the candidate product modules in the candidate product module set, a product conversion rate corresponding to each product in a product set of the candidate product module is estimated based on a pre-trained product module ranking model and the user feature data, including:

[0014] extracting a corresponding module attribute feature for each of the candidate product modules in the candidate product module set;

[0015] extracting a corresponding product attribute feature for each product in the product set of each of the candidate product modules;

[0016] combining each of the product attribute features and the corresponding module attribute features with the user feature data, and inputting them into the product module ranking model to obtain the product conversion rate corresponding to each product in the product set of the candidate product module.

[0017] Optionally, the estimation of the module conversion rate of the candidate product module to which each of the target products belongs based on the product module ranking model and the user feature data includes:

[0018] combining the product attribute feature corresponding to each of the target products, the module attribute feature of the candidate product module to which the target product belongs, and the user feature data, and inputting them into the product module ranking model again to obtain the module conversion rate of the candidate product module to which each of the target products belongs.

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

[0020] obtaining user historical behavior log data in a preset time period;

[0021] cleaning the user historical behavior log data to obtain a user behavior sequence acting on a product module and a product in the product module, wherein the user behavior sequence includes a module behavior sequence and a product behavior sequence;

[0022] extracting a module attribute feature, a commodity attribute feature and a context feature from the user historical behavior log data;

[0023] obtaining a commodity conversion rate and a module conversion rate according to a click behavior and a conversion behavior of the user in the user historical behavior log data;

[0024] training a model network by taking the module behavior sequence, the commodity behavior sequence, the module attribute feature, the commodity attribute feature and the context feature as sample data and taking the commodity conversion rate and the module conversion rate as sample labels to obtain the commodity module ranking model.

[0025] Optionally, the training of the model network by taking the module behavior sequence, the commodity behavior sequence, the module attribute feature, the commodity attribute feature and the context feature as sample data and taking the commodity conversion rate and the module conversion rate as sample labels to obtain the commodity module ranking model comprises:

[0026] converting the module behavior sequence and the commodity behavior sequence through an embedding layer of the model network to obtain an embedding vector feature;

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

[0028] splicing the embedding vector feature and the feature set to obtain a target spliced feature;

[0029] processing the target spliced feature through a plurality of activation layers and a normalization exponential function in sequence to obtain a model estimated output result, wherein the model estimated output result comprises a module conversion rate and a commodity conversion rate.

[0030] Optionally, the method further comprises:

[0031] modifying a loss function adopted by the commodity module ranking model based on a correction factor related to a commodity module size to obtain a commodity module ranking model after modification of the loss function.

[0032] A commodity module ranking device comprises:

[0033] An obtaining unit is configured to obtain user feature data and a commodity module candidate set when a commodity module ranking request sent by a user is received, wherein the user feature data comprises a current module behavior sequence, a current commodity behavior sequence and a current context feature comprising a user basic attribute feature.

[0034] a commodity conversion rate determination unit configured to, for each candidate commodity module in the candidate set of commodity modules, estimate a commodity conversion rate corresponding to each commodity in a 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 target commodity conversion rate, wherein each target commodity conversion rate is higher than a non-target commodity conversion rate that is not selected;

[0036] a module conversion rate determination unit configured to estimate a module conversion rate of each candidate commodity module to which a target commodity belongs based on the commodity module ranking model and the user feature 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 target module conversion rate, wherein each target module conversion rate is higher than a non-target module conversion rate that is not selected;

[0038] a module output unit configured to output each target commodity module and the target commodities included in each target commodity module.

[0039] Optionally, the method further comprises:

[0040] a model training module configured to train the commodity module ranking model;

[0041] The model training module is specifically configured to:

[0042] obtain user historical behavior log data in a preset time period;

[0043] clean the user historical behavior log data to obtain user behavior sequences acting on commodity modules and commodities in the commodity modules, wherein the user behavior sequences include module behavior sequences and commodity behavior sequences;

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

[0045] obtain commodity conversion rates and module conversion rates according to click behaviors and conversion behaviors of users in the user historical behavior log data;

[0046] use the module behavior sequences, the commodity behavior sequences, the module attribute features, the commodity attribute features, and the context features as sample data, use the commodity conversion rates and the module conversion rates as sample labels, train a model network, and obtain the commodity module ranking model.

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

[0048] An electronic device comprises 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] From the above technical solution, the application 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 the user feature data. The first preset number of target commodity conversion rates with high 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 the user feature data, the module conversion rate of each target commodity belonging to the candidate commodity module is estimated. The second preset number of target module conversion rates with high conversion rates are selected from all the module conversion rates, and the 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 application determines the commodity conversion rate and the module conversion rate based on the commodity module sorting model and the user feature data. First, the target commodities with high commodity conversion rates are selected. Then, from each commodity module to which the target commodities belong, the target commodity modules with high module conversion rates are selected. Each target commodity module and the target commodities contained therein are output and displayed. Different commodity modules can be displayed according to different user preferences. That is, the commodity modules and commodities preferred by the user can be displayed in a prominent position. The accuracy of obtaining the preferred commodities of the user and the purchase rate of the commodities are improved. The platform click rate and the final transaction amount are improved, which maximizes the platform revenue. BRIEF DESCRIPTION OF DRAWINGS

[0052] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiments or prior art description will be briefly introduced below. Obviously, the drawings in the following description are only embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor based on the disclosed drawings.

[0053] Figure 1 A commodity module sorting method flow chart disclosed by an embodiment of the present application;

[0054] Figure 2 A user module and commodity interest portrait generation based on user behavior sequence disclosed by an embodiment of the present application;

[0055] Figure 3 A model network structure schematic diagram disclosed by an embodiment of the present application;

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

[0057] Figure 5 A structure schematic diagram of an electronic device disclosed by an embodiment of the present application. DETAILED DESCRIPTION

[0058] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0059] The embodiments of the present application disclose a commodity module sorting method, device, electronic device and storage medium, determine commodity conversion rate and module conversion rate based on commodity module sorting model and user feature data, first filter out each target commodity with high commodity conversion rate, then filter out each target commodity module with high module conversion rate from each commodity module to which the target commodity belongs, and output and display each target commodity module and the target commodity contained therein, realize displaying different commodity modules according to different user preferences, that is, the user preferred commodity module and commodity can be displayed in a position close to the front, improve the accuracy of obtaining user preferred commodities and the purchase rate of commodities, thereby improving platform click rate and final transaction amount, and maximizing platform revenue.

[0060] Referring to Figure 1 The embodiments of the present application disclose a commodity module sorting method flow chart, and the method comprises:

[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 contains the commodity module candidate set, and the commodity module candidate set contains each commodity module that needs to be sorted and the commodities contained in each commodity module that need to be sorted.

[0063] The commodity module ranking determines the user interest and the use habit. A large amount of user behavior log data is accumulated in the platform database, and the recent behavior of the user can be obtained, including but not limited to exposure, click, purchase, interest, etc. The behavior of the user acts on the commodity module or the commodity on the commodity module, and in combination with rich user behavior log data and corresponding label system, the behavior can be used to represent the interest of the user, including the interest of the user for the commodity module and the interest of the user for the commodity under the module.

[0064] Based on this, when the commodity module ranking request sent by the user is received, the user feature data can be obtained from the platform database according to the unique identification of the user. The user feature data in the application is a multi-dimensional vector, which can specifically include: a current module behavior sequence, a current commodity behavior sequence and a current context feature.

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

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

[0067] The current context feature includes the user basic attribute feature, which mainly refers to the attribute describing the basic situation and background of the user. These features are usually static, that is, they will not change in a short period of time, for example, user interest, habit, etc.

[0068] The current module behavior sequence in combination with the commodity information can effectively depict the commodity module in which the user likes the commodity combination state, and the current commodity behavior sequence in combination with the module information can better depict the commodity module and the commodity under the commodity module which the user is interested in.

[0069] In step S102, for each candidate commodity module in the candidate set of commodity modules, the conversion rate of each commodity in the commodity set of the candidate commodity module is estimated based on the pre-trained commodity module ranking model and the user feature data.

[0070] The commodity module ranking model in the application takes the module behavior sequence, the commodity behavior sequence, the module attribute feature, the commodity attribute feature and the context feature as sample data, takes the commodity conversion rate and the module conversion rate as sample labels, and trains the model network to obtain.

[0071] The commodity conversion rate is a key e-commerce and marketing indicator, which is used to measure the proportion of users who actually complete the purchase behavior among users who visit a certain commodity page in a certain time period. This proportion is usually expressed in percentage. The calculation formula is:

[0072] The commodity conversion rate = (the number of customers who generated a purchase behavior / the number of visitors who arrived at the commodity page) x 100%.

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

[0074] The commodity module candidate set includes multiple candidate commodity modules, and each candidate commodity module includes multiple commodities in a commodity combination. Based on the commodity module ranking model and the user feature data, the application estimates the commodity conversion rate corresponding to each commodity in the commodity set of each candidate commodity module.

[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 target commodity conversion rate.

[0076] Each target commodity conversion rate is higher than a non-target commodity conversion rate that is not selected.

[0077] In actual application, after estimating the commodity conversion rate corresponding to each commodity under each candidate commodity module of the commodity module candidate set, all the commodity conversion rates can be sorted in descending order, a first preset number of target commodity conversion rates with high rankings are selected, and the commodities corresponding to each target commodity conversion rate are used as target commodities. The target commodities are used as commodities displayed in the commodity module. The commodities that are not selected are defined as non-target commodities.

[0078] The value of the first preset number is determined according to actual needs, which is not limited in the application.

[0079] Step S104, based on the commodity module ranking model and the user feature data, estimating the module conversion rate of each target commodity belonging to a candidate commodity module.

[0080] The module conversion rate is a term specific to the e-commerce or online sales field, which refers to the proportion of visitors who are converted into actual purchasers in a certain commodity module (such as a commodity detail page, a commodity list page, etc.) of an e-commerce platform. This proportion is usually expressed in percentage, reflecting the ability of the commodity module to attract visitors and promote them to generate a purchase behavior.

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

[0082] Module conversion rate = (number of customers who complete a purchase through a product module in a certain time period / number of visitors who visit the product module in the same time period) x 100%.

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

[0084] Each target module conversion rate is higher than a non-target module conversion rate that is not selected.

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

[0086] Step S106, outputting each target product module and the target product included in each target product module.

[0087] Each target product module and the target product included in each target product module output by the present application are the product modules and the products included in the product modules recommended according to the user's preferences.

[0088] In conclusion, the application 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 acquired, for each candidate commodity module in the commodity module candidate set, a commodity conversion rate corresponding to each commodity in a commodity set of the candidate commodity module is estimated based on a pre-trained commodity module sorting model and the user feature data, a target commodity conversion rate with a high conversion rate is 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 the user feature data, a module conversion rate of each target commodity belonging to the candidate commodity module is estimated, a target module conversion rate with a high conversion rate is selected from all the module conversion rates, and a 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 outputted.

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

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

[0091] The module attribute features are key information for describing commodity features, which are crucial for user selection of commodities and management of e-commerce platforms. Commodity attributes usually include commodity name, commodity price, commodity brand, etc.

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

[0093] The commodity attribute features are features for describing and distinguishing commodity characteristics, performance, function, appearance, specifications, etc., which provide a basis for consumers to select and compare.

[0094] (3) inputting each commodity attribute feature and its corresponding module attribute feature combined with the user feature data into the commodity module sorting model to obtain the commodity conversion rate corresponding to each commodity in the commodity set of the candidate commodity module.

[0095] The module attribute features of the candidate commodity module and the commodity attribute features corresponding to each commodity in the commodity set of the candidate commodity module are combined one by one, and user feature data is combined and input into a commodity module sorting model for processing, so that the commodity conversion rate corresponding to each commodity can be obtained.

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

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

[0098] In the determination of the module conversion rate, the commodity attribute features input into the commodity module sorting model are the commodity attribute features of each target commodity included in the candidate commodity module.

[0099] In one embodiment, the application further discloses a training process of a commodity module sorting model, specifically as follows:

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

[0101] The user historical behavior log data refers to the operation behaviors of the user in the commodity module and the commodities of the platform in the preset time period, which can include exposure, clicking, adding to shopping cart, completing an order, collecting, etc.

[0102] The value of the preset time period is determined according to actual needs, for example, one month, which is not limited in the application.

[0103] (2) Clean the user historical behavior log data to obtain a user behavior sequence acting on the commodity module and the commodities in the commodity module.

[0104] The user behavior sequence includes a module behavior sequence and a commodity behavior sequence, and the user module and the commodity interest portrait can be generated according to the user behavior sequence.

[0105] The user historical behavior log data is cleaned, so that the cleaned data can more accurately reflect the interests, preferences and behavior patterns of the user, which is helpful to build a more detailed user portrait.

[0106] Referring to Figure 2 The embodiment of the application discloses a schematic diagram of a user module and a commodity interest portrait generated based on a user behavior sequence, and the module behavior sequence includes the module itself and the commodities on the module. Figure 2In the embodiment, the module behavior sequence involves three operation behaviors of exposure, click and conversion, the commodity module 1 includes commodity A1, commodity B1, …, commodity T1, the commodity module 2 includes commodity A2, commodity B2, …, commodity T2, …, and the commodity module N includes commodity AN, commodity BN, …, commodity TN.

[0107] In the embodiment, the module behavior sequence involves three operation behaviors of exposure, click and conversion, the commodity module 1 includes commodity A1, commodity B1, …, commodity T1, the commodity module 2 includes commodity A2, commodity B2, …, commodity T2, …, and the commodity module N includes commodity AN, commodity BN, …, commodity TN.

[0108] In the embodiment, the module behavior sequence involves three operation behaviors of exposure, click and conversion, the commodity module 1 includes commodity A1, commodity B1, …, commodity T1, the commodity module 2 includes commodity A2, commodity B2, …, commodity T2, …, and the commodity module N includes commodity AN, commodity BN, …, commodity TN.

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

[0110] The module attribute features are key information for describing commodity features, which are crucial for user selection of commodities and management of e-commerce platforms. Commodity attributes usually include commodity name, commodity price, commodity brand, etc.

[0111] The commodity attribute features are features for describing and distinguishing the characteristics, performance, function, appearance, specification, etc. of commodities, which provide the basis for consumers to select and compare.

[0112] The context features include user basic attribute features, which mainly refer to attributes for describing the basic situation and background of users. These features are usually static, i.e., they do not change in a short period of time, such as user interests, habits, etc.

[0113] (4) Obtaining commodity conversion rate and module conversion rate according to the click behavior and conversion behavior of the user in the user historical behavior log data.

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

[0115] Module conversion rate = (number of customers who complete purchase through the commodity module in a certain time period / number of visitors who visit the commodity module in the same time period) x 100%.

[0116] The calculation formula of the commodity conversion rate is as follows:

[0117] Commodity conversion rate = (number of customers who generate purchase behavior / number of visitors who arrive at the commodity page) x 100%.

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

[0119] Specifically, referring to the model network structure diagram shown in FIG. 1, the training process of the model network is as follows: Figure 3

[0120] (1) The module behavior sequence and the commodity behavior sequence are converted into embedding vector features through an embedding layer of the model network.

[0121] The embedding layer is an important component in a deep learning model, and has multiple functions such as dimension reduction, expression of word relationship, introduction of additional information, and optimization effect. In the word vector space, the embedding layer can represent each word as a fixed-dimensional vector. 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, so that words with similar semantics are closer in the vector space.

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

[0123] The hidden layer in the model network refers to the intermediate layer between the input layer and the output layer. The hidden layer provides powerful expression and learning capabilities for the neural network through functions such as feature extraction and transformation, non-linear 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 key factors that affect the performance of the model.

[0124] In this application, the commodity attribute feature, the module attribute feature, and the context feature are processed through the hidden layer to obtain the feature set.

[0125] (3) The embedding vector feature and the feature set are spliced to obtain a target splicing feature.

[0126] (4) The target splicing feature is processed through multiple activation layers and a normalization exponential function in sequence to obtain a model prediction output result.

[0127] The model prediction output result includes the module conversion rate and the commodity conversion rate.

[0128] ​The multi-layer activation layer solves the problem of insufficient expression ability of the linear model by introducing a nonlinear activation function. Each activation layer of the multi-layer activation layer can be regarded as a nonlinear transformation of a feature space. Such transformation can convert a problem that is originally linearly inseparable into a problem that is linearly separable, thereby greatly increasing the ability of the model 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 log normalization of a finite term discrete probability distribution.

[0130] The present application obtains a model prediction output result including a module conversion rate and a commodity conversion rate by sequentially processing the target splicing feature through the multi-layer activation layer and the normalized exponential function.

[0131] It can be understood that different commodity modules differ in commodity form and display size. A module with a larger size has certain advantages in conversion efficiency, but consumes more display resources. Therefore, to ensure fairness in module conversion calculation and avoid model bias towards modules with larger sizes, the commodity module sorting model also needs to be corrected. In the model correction process, the module size is introduced for conversion rate correction. For modules with larger sizes and larger commodity image sizes, a certain degree of weight reduction is given to ensure fair and reasonable exposure of each module.

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

[0133] The loss function adopted by the commodity module sorting model is corrected based on the correction factor related to the size of the commodity module to obtain the commodity module sorting model after correction of the loss function.

[0134] The present application corrects the loss function adopted by the commodity module sorting model based on the modification factor related to the size of the commodity module, which can eliminate the deviation caused by the size of the first commodity module and obtain the commodity module sorting model after correction of the loss function.

[0135] In specific practice, this purpose is achieved by improving the loss function of the neural network. On the basis of the original loss function, the product of the difference between the estimated conversion rate and the actual conversion rate and the module size term is added. For example, in the present application, the focal loss loss function is used for the neural network, and a bias term is added. 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(p-y)] / (Mxa My b );

[0137] In the formula, alpha, gamma, a, b are all common parameters, the best value is obtained by adjusting parameters, y is a label value in a sample, a positive sample is 1, and a negative sample is 0, p is a model estimated click rate, Mx is a module long size, My is a module width size, the larger the area is, the larger the loss function is, the second term of the loss function is reduced when the positive sample estimated value is small, and the effect of adjusting the module area bias is achieved.

[0138] Corresponding to the method embodiments, the application further discloses a commodity module sorting device.

[0139] Referring to Figure 4 The commodity module sorting device disclosed by the embodiment of the application can include:

[0140] The acquisition unit 201 is configured 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 the commodity module candidate set, the commodity module candidate set includes various commodity modules that need to be sorted, and each commodity module includes commodities that need to be sorted.

[0142] The commodity module sorting determines user interests and use habits. A platform database accumulates a large amount of user behavior log data, and can obtain recent user behaviors, including but not limited to exposure, clicks, purchases, interests, and the like. The behaviors of the user act on commodity modules or commodities on the commodity modules, and in combination with rich user behavior log data and a corresponding label system, the behaviors can be used to represent user interests, including user interests in commodity modules and commodities under the modules.

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

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

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

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

[0147] The current module behavior sequence combines the product information, which can effectively depict the product module in which the user likes the product combination. The current product behavior sequence includes the module information, which can better depict the product module and the product under the product module that the user is interested in.

[0148] The product conversion rate determination unit 202 is configured to, for each candidate product module in the candidate product module set, estimate the product conversion rate corresponding to each product in the product set of the candidate product module based on the pre-trained product module ranking model and the user feature data.

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

[0150] The product conversion rate is a key e-commerce and marketing indicator, which is used to measure the proportion of users who actually complete the purchase behavior among users who visit a certain product page in a certain time period. This proportion is usually expressed in percentage. The calculation formula is:

[0151] Product conversion rate = (number of customers who generate purchase behavior / number of visitors who arrive at the product page) x 100%.

[0152] The product conversion rate can directly reflect the attractiveness of the product, the effectiveness of the sales strategy and the smoothness of the user 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, thereby attracting more users to complete the purchase.

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

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

[0155] In practical applications, after estimating the conversion rates of each product corresponding to each candidate product module in the candidate set of product modules, all the product conversion rates can be sorted in descending order, the top first preset number of target product conversion rates are selected, and the products corresponding to each target product conversion rate are defined as target products, which are also the products displayed in the product module. The products that are not selected are defined as non-target products.

[0156] The first preset number is determined according to actual needs, which is not limited in the present application.

[0157] The module conversion rate determination unit 204 is configured to estimate the module conversion rate of each candidate product module to which the target product belongs based on the product module ranking model and the user feature data.

[0158] Module conversion rate is a term specific to the e-commerce or online sales field, which refers to the proportion of visitors who actually purchase in a certain product module (such as product detail page, product list page, etc.) of an e-commerce platform. This proportion is usually expressed in percentage, reflecting the ability of the product module to attract visitors and promote them to make a purchase.

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

[0160] Module conversion rate = (number of customers who complete purchase through the product module in a certain time period / number of visitors who visit the product module in the same time period) x 100%.

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

[0162] In practical applications, after estimating the module conversion rate corresponding to each product module in the candidate set of products, all the module conversion rates can be sorted in descending order, the top second preset number of target module conversion rates are selected, and the product module corresponding to each target module conversion rate is determined as the target product module. The product module that is not selected is defined as a non-target product module.

[0163] The module output unit 206 is configured to output each target product module and the target products included in each target product module.

[0164] The target commodity module output by the application and the target commodity contained in each target commodity module are the commodity module recommended according to the user preference and the commodity contained in the commodity module.

[0165] In conclusion, the application discloses a commodity module sorting device, when receiving a commodity module sorting request sent by a user, 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 the 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 the user feature data, a module conversion rate of each target commodity belonging to the candidate commodity module 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, and each target commodity module and the target commodity contained in each target commodity module are output. The commodity conversion rate and the module conversion rate are determined based on the commodity module sorting model and the user feature data, the target commodities with high commodity conversion rates are first screened out, then the target commodity modules with high module conversion rates are screened out from the commodity modules to which the target commodities belong, and each target commodity module and the target commodity contained therein are output and displayed, different commodity modules are displayed according to different user preferences, that is, the commodity modules and commodities preferred by the user can be displayed in a position close to the front, the accuracy of obtaining the commodities preferred by the user and the purchase rate of the commodities are improved, the platform click rate and the final transaction amount are improved, and the platform revenue maximization is beneficial.

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

[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] combining each commodity attribute feature and its corresponding module attribute feature with the user feature data, and inputting the combination into the commodity module sorting 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 can be specifically used for:

[0171] The product attribute features corresponding to each target product, 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 each candidate product module to which the target product belongs.

[0172] In one embodiment, the product module ranking device can further include:

[0173] a model training module configured to train the product module ranking model.

[0174] The model training module can be specifically configured to:

[0175] obtain user historical behavior log data in a preset time period;

[0176] clean the user historical behavior log data to obtain user behavior sequences acting on product modules and products in the product modules, wherein the user behavior sequences include module behavior sequences and product behavior sequences;

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

[0178] obtain product conversion rates and module conversion rates according to click behaviors and conversion behaviors of users in the user historical behavior log data;

[0179] use the module behavior sequences, the product behavior sequences, the module attribute features, the product attribute features, and the context features as sample data, and use the product conversion rates and the module conversion rates as sample labels to train a model network to obtain the product module ranking model.

[0180] In one embodiment, the model training module can be specifically configured to:

[0181] convert the module behavior sequences and the product behavior sequences through an embedding layer of the model network to obtain embedding vector features;

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

[0183] splice the embedding vector features and the feature set to obtain target spliced features;

[0184] process the target spliced features through multiple activation layers and a normalization exponential function in sequence to obtain a model estimation output result, wherein the model estimation output result includes the module conversion rate and the product conversion rate.

[0185] In one embodiment, the model training module can be further specifically used for:

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

[0187] It should be noted that the specific working principles of the components in the device embodiment can be found in the corresponding parts of the method embodiment, which will not be repeated here.

[0188] Corresponding to the above embodiments, the present application also discloses a computer storage medium, which stores at least one instruction, and the at least one instruction is executed by a processor to implement the steps shown in the commodity module ranking method embodiment.

[0189] The computer storage medium can be a tangible medium, which can contain or store programs for use by or in connection with an instruction execution system, apparatus or device. The computer storage medium can be a machine-readable signal medium or a machine-readable storage medium. The computer storage medium can include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatuses or devices, or any suitable combination of the above. More specific examples of machine-readable storage media can include one or more wires, portable computer disks, hard drives, random access memories (RAM), read-only memories (ROM), erasable programmable read-only memories (EPROM or flash memory), optical fibers, compact disk read-only memories (CD-ROMs), optical storage devices, magnetic storage devices, or any suitable combination of the above.

[0190] Corresponding to the above embodiments, as shown in Figure 5 The present application also provides an electronic device, which can include a processor 1 and a memory 2.

[0191] The processor 1 and the memory 2 can complete mutual communication through a communication bus 3.

[0192] The processor 1 is used to execute at least one instruction.

[0193] The memory 2 is used to store at least one instruction.

[0194] The processor 1 can be a central processing unit CPU, or an application specific integrated circuit ASIC, or one or more integrated circuits configured to carry out embodiments of the application.

[0195] The memory 2 can comprise a high speed RAM memory, and possibly also a 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 method of ordering the modules of the product.

[0197] Finally, it is to be understood that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting; it is also possible that the use of the same term in different places in the same or different embodiments can have different meanings. It is also possible that various terms used herein can be interchanged under appropriate circumstances. Furthermore, the use of the terms "a", "an", "the", etc. and "comprising", "having", "containing", etc. are used generically herein to describe individual embodiments and can be dependent on the embodiment being described. As used herein, the article "a" is intended to include one or more items. For example, the articles "a" and "an" are used herein to refer to one or more items unless otherwise indicated. For example, the term "an item" can mean one or more items. Similarly, the term "comprising" is used herein to mean, and is used interchangeably with, the term "including". The term "including" is, therefore, synonymous with and means the inclusion of one or more items, elements, components, etc. listed after this term in the claims or the specification. Further, it is to be understood that the use of "one or more of" preceded by a comma should be understood to impart "one", "two", "three", "four", "five", "six", "seven", "eight", "nine", "ten", "eleven", "twelve", "thirteen", "fourteen", "fifteen", "sixteen", "seventeen", "eighteen", "nineteen", or "twenty" or a range or list combing any of these, such as "one or more of the two of the items", "one or more of the three of the items", "one or more of the four of the items" and the like.

[0198] The various embodiments described herein are presented by way of example only and are not intended, nor should they be construed, to limit the scope of the application. Embodiments described herein can be combined in any way, and the description of the various embodiments should not be construed to limit the scope of the application.

[0199] The above description of disclosed embodiments is summarised to enable a person skilled in the art to carry out or use the application. Numerous modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined can be applied to other embodiments without departing from the spirit or scope of the application. Thus, the present application is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method of ordering items in a module, characterized by, The method comprises the following steps: When receiving a commodity module sorting request sent by a user, user feature data and a commodity module candidate set are obtained, wherein the user feature data comprises a current module behavior sequence, a current commodity behavior sequence and a current context feature comprising a user basic attribute feature; 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 the user feature data; From all the commodity conversion rates, a first preset number of target commodity conversion rates are selected, and a target commodity corresponding to each target commodity conversion rate is determined, wherein each target commodity conversion rate is higher than a non-target commodity conversion rate that is not selected; Based on the commodity module sorting model and the user feature data, a module conversion rate of a candidate commodity module to which each target commodity belongs is estimated; From all the module conversion rates, a second preset number of target module conversion rates are selected, and a target commodity module corresponding to each target module conversion rate is determined, wherein each target module conversion rate is higher than a non-target module conversion rate that is not selected; Each target commodity module and the target commodities contained in each target commodity module are outputted.

2. The merchandise module sequencing method of claim 1, wherein, The method comprises the following steps: 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 the user feature data; For each candidate commodity module in the commodity module candidate set, a corresponding module attribute feature is extracted; For each commodity in the commodity set of each candidate commodity module, a corresponding commodity attribute feature is extracted; 3. The merchandise module sequencing method of claim 1, wherein, Each commodity attribute feature and its corresponding module attribute feature are combined with the user feature data and input into the commodity module sorting model to estimate the commodity conversion rate corresponding to each commodity in the commodity set of the candidate commodity module. The method comprises the following steps:

4. The method of claim 1, 2 or 3, wherein, The commodity attribute feature corresponding to each target commodity, the module attribute feature of the candidate commodity module to which the target commodity belongs, and the user feature data are combined and then input into the commodity module sorting model again to estimate the module conversion rate of the candidate commodity module to which each target commodity belongs. The training process of the commodity module sorting model comprises the following steps: User historical behavior log data in a preset time period is obtained; The user historical behavior log data is cleaned to obtain user behavior sequences acting on commodity modules and commodities in the commodity modules, wherein the user behavior sequences comprise module behavior sequences and commodity behavior sequences; Module attribute features, commodity attribute features and context features are extracted from the user historical behavior log data; Based on the click behavior and conversion behavior of the user in the user historical behavior log data, commodity conversion rates and module conversion rates are obtained; The module behavior sequence, the commodity behavior sequence, the module attribute feature, the commodity attribute feature, and the context feature are taken as sample data, the commodity conversion rate and the module conversion rate are taken as sample labels, a model network is trained, and the commodity module ranking model is obtained.

5. The method of claim 4, wherein, The module behavior sequence, the commodity behavior sequence, the module attribute feature, the commodity attribute feature, and the context feature are taken as sample data, the commodity conversion rate and the module conversion rate are taken as sample labels, a model network is trained, and the commodity module ranking model is obtained. The module behavior sequence and the commodity behavior sequence are converted into embedding vector features through an embedding layer of a model network; The commodity attribute feature, the module attribute feature, and the context feature are processed through a hidden layer of the model network to obtain a feature set; The embedding vector features and the feature set are spliced to obtain target spliced features; The target spliced features are sequentially processed through multiple activation layers and a normalization exponential function to obtain a model estimation output result, wherein the model estimation output result includes a module conversion rate and a commodity conversion rate.

6. The merchandise module sequencing method of claim 4, wherein, Further comprising: A loss function adopted by the commodity module ranking model is modified based on a correction factor related to a commodity module size to obtain a commodity module ranking model after loss function modification.

7. A merchandise module sequencing apparatus characterized by, Further comprising: An acquisition unit is configured to acquire user feature data and a commodity module candidate set when a commodity module ranking request sent by a user is received, wherein the user feature data includes a current module behavior sequence, a current commodity behavior sequence, and a current context feature including user basic attribute features; A commodity conversion rate determination unit is configured to, for each candidate commodity module in the commodity module candidate set, estimate a commodity conversion rate corresponding to each commodity in a 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 is 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 target commodity conversion rate, wherein each target commodity conversion rate is higher than a non-target commodity conversion rate that is not selected; A module conversion rate determination unit is configured to estimate a module conversion rate of a candidate commodity module to which each target commodity belongs based on the commodity module ranking model and the user feature data; A target commodity module determination unit is 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 target module conversion rate, wherein each target module conversion rate is higher than a non-target module conversion rate that is not selected; A module output unit is configured to output each target commodity module and the target commodities included in each target commodity module.

8. The merchandise module sequencing arrangement of claim 7, wherein, Further comprising: A model training module is configured to train the commodity module ranking model; The model training module is specifically configured to: Obtaining user historical behavior log data in 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 comprises a module behavior sequence and a commodity behavior sequence; Extracting module attribute features, commodity attribute features and context features from the user historical behavior log data; Obtaining a commodity conversion rate and a module conversion rate according to a click behavior and a conversion behavior of a user in the user historical behavior log data; Taking the module behavior sequence, the commodity behavior sequence, the module attribute features, the commodity attribute features and the context features as sample data, taking the commodity conversion rate and the module conversion rate as sample labels, training a model network to obtain the commodity module ranking model.

9. A computer storage medium, characterized in that The computer storage medium stores at least one instruction, and the at least one instruction is executed by a processor to implement the commodity module ranking method in any one of claims 1-6.

10. An electronic device, comprising: 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 ranking method in any one of claims 1-6.

Images