Product delivery methods, devices, equipment and computer-readable media

Abstract

This application relates to a product delivery method, apparatus, device, and computer-readable medium. The method includes: acquiring first characteristic information of a target object and second characteristic information of a product to be delivered, wherein the first characteristic information represents the target object's historical behavior on a target platform, and the second characteristic information represents the product characteristics of the product to be delivered, including virtual object resources; extracting cross-features and correlation features between the first and second characteristic information, and using the cross-features and correlation features to determine the target object's preference for the product to be delivered; determining a target product combination from the product to be delivered based on the preference results; and displaying a target display card matching the target product combination to the target object. This application solves the technical problem of low matching between recommended products and users by mining the correlation between user characteristics and the characteristics of the product to be delivered.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computers, and particularly relates to a product putting method and device, equipment and a computer readable medium. BACKGROUND

[0002] With the rapid development of information technology and the Internet, especially in recent years, the emergence of e-commerce websites and social networking sites, information overload has become a serious problem faced by users and suppliers, and the recommendation system has become an effective means to solve this problem. It analyzes the interests of users or the relationship between items according to user information, and provides personalized recommendation services for users.

[0003] At present, in the related art, the recommendation system uses full-amount display and putting of all products, so that all users see the same page after entering, and then the recommendation system can put certain products to the users after guiding the users to perform certain actions. There are also collaborative filtering recommendation algorithms or improvements to collaborative filtering algorithms for product recommendation, but the collaborative filtering algorithm has problems such as feature sparsity and cold start. In the above related art, the recommended products are not highly matched with the users.

[0004] In view of the problem that the recommended products are not highly matched with the users, no effective solution has been proposed so far. SUMMARY

[0005] The present application provides a product putting method, device, equipment and computer readable medium to solve the technical problem that the recommended products are not highly matched with the users.

[0006] According to an aspect of an embodiment of the present application, the present application provides a product putting method, comprising: obtaining first feature information of a target object and second feature information of a to-be-put product, the first feature information being used to represent historical behaviors of the target object on a target platform, and the second feature information being used to represent product features of the to-be-put product, the to-be-put product comprising virtual object resources used for virtual resource exchange; extracting cross features and association features between the first feature information and the second feature information, and determining a preference result of the target object to the to-be-put product by using the cross features and the association features; determining a target product combination from the to-be-put product according to the preference result; and displaying a target display card matched with the target product combination to the target object.

[0007] According to another aspect of the embodiments of the present application, the present application provides a product delivery device, comprising: an information acquisition module, configured to acquire first characteristic information of a target object and second characteristic information of a product to be delivered, the first characteristic information being used to represent historical behaviors of the target object on a target platform, and the second characteristic information being used to represent product characteristics of the product to be delivered, the product to be delivered comprising virtual object resources used for virtual resource exchange; a preference matching module, configured to extract cross characteristics and associated characteristics between the first characteristic information and the second characteristic information, and determine a preference result of the target object for the product to be delivered by using the cross characteristics and the associated characteristics; a combination determination module, configured to determine a target product combination from the product to be delivered according to the preference result; and a delivery display module, configured to display a target display card matched with the target product combination to the target object.

[0008] According to another aspect of the embodiments of the present application, the present application provides an electronic device, comprising a memory, a processor, a communication interface and a communication bus, the memory stores a computer program capable of running on the processor, the memory, the processor and the communication interface communicate through the communication bus, and the processor executes the computer program to implement the above method.

[0009] According to another aspect of the embodiments of the present application, the present application further provides a computer readable medium having a non-volatile program code executable by a processor, the program code causing the processor to execute the above method.

[0010] The above technical solutions provided by the embodiments of the present application have the following advantages compared with related technologies:

[0011] The technical solutions provided by the embodiments of the present application acquire the first characteristic information of the target object and the second characteristic information of the product to be delivered, the first characteristic information being used to represent historical behaviors of the target object on a target platform, and the second characteristic information being used to represent product characteristics of the product to be delivered, the product to be delivered being virtual object resources; extract cross characteristics and associated characteristics between the first characteristic information and the second characteristic information, and determine a preference result of the target object for the product to be delivered by using the cross characteristics and the associated characteristics; determine a target product combination from the product to be delivered according to the preference result; and display a target display card matched with the target product combination to the target object. The present application finds the product to be delivered preferred by the target object by mining the associated relationship between the user characteristics and the product characteristics of the product to be delivered, thereby solving the technical problem that the recommended product is not highly consistent with the user. BRIEF DESCRIPTION OF DRAWINGS

[0012] The accompanying drawings, which are incorporated herein and form part of the specification, illustrate embodiments consistent with the present application and, together with the description, further serve to explain the principles of the present application.

[0013] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the related art, the following will briefly introduce the drawings needed to be used in the embodiments or the related art description. Obviously, for those of ordinary skill in the art, other drawings can also be obtained based on these drawings without any creative effort.

[0014] Figure 1 An optional product delivery method hardware environment schematic diagram provided according to an embodiment of the present application;

[0015] Figure 2 An optional product delivery method flowchart provided according to an embodiment of the present application;

[0016] Figure 3 An optional user preference mining flowchart provided according to an embodiment of the present application;

[0017] Figure 4 An optional double-tower feature network schematic diagram provided according to an embodiment of the present application;

[0018] Figure 5 An optional product recommendation schematic diagram provided according to an embodiment of the present application;

[0019] Figure 6 An optional product delivery device block diagram provided according to an embodiment of the present application;

[0020] Figure 7 An optional electronic device structure schematic diagram provided according to an embodiment of the present application. DETAILED DESCRIPTION

[0021] In order to make the objects, technical solutions and advantages of the embodiments of the present application clearer, the following will combine the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without any creative effort fall within the scope of protection of the present application.

[0022] In the subsequent description, the suffixes such as "module", "component", or "unit" used to represent elements are only for the convenience of the description of the present application, and have no specific meaning. Therefore, "module" and "component" can be used interchangeably.

[0023] First, some nouns or terms appearing in the description of the embodiments of the present application are applicable to the following explanations:

[0024] Neural network: The neural network can be composed of neural units, and the neural unit can refer to an xs The operation unit takes the intercept b as input, and the output of this operation unit can be:

[0025]

[0026] Where s = 1, 2, ..., n, n is a natural number greater than 1, W s For x s The weights are denoted by b, where b is the bias of the neural unit. f is the activation function of the neural unit, used to introduce nonlinear characteristics into the neural network to convert the input signal in the neural unit into an output signal. The output signal of this activation function can be used as the input to the next convolutional layer. The activation function can be the sigmoid function. A neural network is a network formed by connecting many of the above-mentioned individual neural units together, that is, the output of one neural unit can be the input of another neural unit. The input of each neural unit can be connected to the local receptive field of the previous layer to extract the features of the local receptive field, which can be a region composed of several neural units.

[0027] Deep Neural Networks (DNNs), also known as multi-layered neural networks, can be understood as neural networks with many hidden layers, though there's no specific metric for "many." DNNs can be categorized into three types based on their layer positions: input layers, hidden layers, and output layers. Generally, the first layer is the input layer, the last layer is the output layer, and the layers in between are hidden layers. For example, in a fully connected neural network, all layers are fully connected, meaning that any neuron in layer i is connected to any neuron in layer i+1. Although DNNs appear complex, the operation of each layer is actually quite simple, resembling a linear relationship as shown in the following expression: in, It is the input vector. It is the output vector. α is the offset vector, W is the weight matrix (also called coefficients), and α() is the activation function. Each layer is simply an adjustment of the input vector. The output vector is obtained through such a simple operation. Because DNNs have many layers, the coefficients W and the offset vector... The number of these parameters is therefore quite large. The definitions of these parameters in a DNN are as follows: Taking the coefficient W as an example: Assuming a three-layer DNN, the linear coefficient from the 4th neuron in the second layer to the 2nd neuron in the third layer is defined as... The superscript 3 represents the layer number where coefficient W resides, while the subscript corresponds to the output third layer index 2 and the input second layer index 4. In summary, the coefficients from the k-th neuron in layer L-1 to the j-th neuron in layer L are defined as follows: It's important to note that the input layer does not have a W parameter. In deep neural networks, more hidden layers allow the network to better represent complex real-world situations. Theoretically, the more parameters a model has, the higher its complexity and "capacity," meaning it can perform more complex learning tasks. Training a deep neural network is essentially the process of learning the weight matrix, with the ultimate goal of obtaining the weight matrix of all layers in the trained deep neural network (a weight matrix formed by the vectors W from many layers).

[0028] Cross Networks: Cross networks are used to explicitly and efficiently learn cross features. They consist of multiple cross layers, each represented by the following formula:

[0029]

[0030] in, The output features of the l-th layer and the (l+1)-th layer are respectively. Let w represent the weights and biases of the l-th layer. The special structure of a Cross Network allows the degree of the features to increase with the network depth. For example, in a Cross Network with l layers, the highest polynomial degree is l+1. Since the weights w and biases b of each cross layer are vectors, assuming they all have dimension d, then the number of layers is L. c The total number of parameters in the CrossNetwork is d*L c *2.

[0031] In related technologies, some recommendation systems display and distribute all products at once, ensuring all users see the same page upon entry. Only after guiding users to perform certain actions can the system then distribute specific products. This process, with its complex interactions and information output, leads to "information overload" and a poor user experience. Users often claim products they don't need, resulting in wasted company resources, such as the need for manual verification and calculation after distribution. Furthermore, users' perception of the activities is no different from other platforms, making the products unattractive and failing to motivate users to share or obtain them again.

[0032] Collaborative filtering recommendation algorithms or improvements thereof are also used for product recommendation. However, collaborative filtering algorithms suffer from problems such as feature sparsity and cold start.

[0033] To address the problems mentioned in the background art, according to one aspect of the embodiments of this application, an embodiment of a product delivery method is provided.

[0034] Optionally, in the embodiments of this application, the above-described product delivery method can be applied to, for example... Figure 1 The hardware environment shown consists of terminal 101 and server 103. The terminal can be a user terminal. When a user accesses the software platform through the terminal, the recommendation system displays the target products recommended to the target user on the terminal's display interface. The server is the data server of the software platform. By retrieving user data and product data, it performs correlation mining between users and products, thereby recommending target products that the target user is more likely to like and be interested in.

[0035] like Figure 1 As shown, server 103 is connected to terminal 101 via a network and can be used to provide services to the terminal or clients installed on the terminal. Database 105 can be set up on the server or independently of the server to provide data storage services for server 103. The network mentioned above includes, but is not limited to, wide area network, metropolitan area network or local area network. Terminal 101 includes, but is not limited to, PC, mobile phone, tablet computer, etc.

[0036] One product delivery method in this application embodiment can be executed by server 103, or it can be jointly executed by server 103 and terminal 101, such as... Figure 2 As shown, the method may include the following steps:

[0037] Step S202: Obtain the first feature information of the target object and the second feature information of the product to be deployed. The first feature information is used to represent the historical behavior of the target object on the target platform, and the second feature information is used to represent the product features of the product to be deployed. The product to be deployed includes virtual object resources used for virtual resource exchange.

[0038] In this application embodiment, the target object is the user who accesses the target software platform through a terminal device. Preferably, the target software platform in this application is an application, mini-program, website, etc. for insurance business.

[0039] The primary characteristic information of the target user includes basic user information such as name, age, frequently used address, gender, and current mobile phone number location. This basic information is filled in by the user when they first register to use the insurance business software platform. Primary characteristic information may also include user behavioral habits and preferences, such as whether they smoke, drink alcohol, or stay up late. This information can be collected through user behavior surveys. Primary characteristic information may also include records of the user's specific browsing, clicking, and following activities on the insurance business software platform, such as browsing promotional activities, platform points redemption activities, product introduction pages, and personal centers.

[0040] The products to be advertised can be insurance products, or virtual objects used for virtual resource exchange, such as prizes from platform activities, like cash red envelopes, vouchers, gifts, or VIP cards. The second set of information for the products to be advertised includes the product type, price, tags, reward amount, and the number of clicks and click-through rate in the past week.

[0041] Step S204: Extract the cross features and correlation features between the first feature information and the second feature information, and use the cross features and correlation features to determine the target object's preference results for the product to be delivered.

[0042] In this embodiment, we can start from both user characteristics and product characteristics, capture the correlation between user characteristics and product characteristics, obtain cross features, and obtain deeper nonlinear correlations to obtain associated features, thereby finding products that are more likely to match user preferences for deployment.

[0043] Optionally, such as Figure 3 As shown, step S204, which extracts the cross-features and correlation features between the first feature information and the second feature information, and uses these cross-features and correlation features to determine the target object's preference for the product to be delivered, may specifically include the following steps:

[0044] Step 302: Convert the first feature information into a first feature vector and convert the second feature information into a second feature vector. The first feature information includes at least one of the numerical feature information and the attribute feature information of the target object. The second feature information includes at least one of the numerical feature information and the attribute feature information of the product to be delivered.

[0045] In this embodiment, embedding can be used to convert the first feature information into a first feature vector and the second feature information into a second feature vector. The first feature information includes basic user features such as gender, age, province of residence, and current phone number location, as well as behavioral preference features such as the user's product browsing history and product purchase history on the target platform over the past week. Both user and product features include numerical and attribute-based feature information. Since a single feature may contain multiple values ​​(e.g., product prices vary at different times, and a person's age increases over time), embedding_lookup cannot be used directly for vector conversion. Embedding_lookup only accepts one-hot encoding with a single value. Therefore, the first and second feature information need to be discretized to convert them into multi-hot encoding before embedding into the first and second feature vectors.

[0046] Optionally, converting the first feature information into a first feature vector and converting the second feature information into a second feature vector includes:

[0047] Determine the data types of the first and second feature information;

[0048] When the data type is an attribute type, the first feature information is discretized to obtain a first multi-hot code; the first multi-hot code is input into the embedding layer to obtain a first feature vector obtained by vectorizing the first multi-hot code by the embedding layer; the second feature information is discretized to obtain a second multi-hot code; the second multi-hot code is input into the embedding layer to obtain a second feature vector obtained by vectorizing the second multi-hot code by the embedding layer.

[0049] When the data type is numeric, the initial feedback dataset is obtained, which is the behavioral data generated when a user logs into the target platform for the first time. The initial feedback dataset is divided into multiple subsets, and the first feature information is assigned to each subset. Each subset is discretized to obtain a third multi-hot encoding. The third multi-hot encoding is input into the embedding layer to obtain the first feature vector obtained by the embedding layer vectorizing the third multi-hot encoding. The length of the first feature vector matches the number of subsets. The initial feedback dataset is divided into multiple subsets, and the second feature information is assigned to each subset. Each subset is discretized to obtain a fourth multi-hot encoding. The fourth multi-hot encoding is input into the embedding layer to obtain the second feature vector obtained by the embedding layer vectorizing the fourth multi-hot encoding. The length of the second feature vector matches the number of subsets. For example, if the vector length matches the number of subsets, and the dataset is divided into 10 buckets, then a vector of length 10 is used to record this feature.

[0050] In this embodiment of the application, the feature information can be discretized. Both the first feature information and the second feature information contain attribute-type feature information and numerical continuous features. The discretization methods for different feature information are different.

[0051] For attribute-type features, such as user gender, province of residence, product category tags, and product type, this application can directly discretize the attributes. Specifically, the first feature information is discretized to obtain a first multi-hot code, and the second feature information is discretized to obtain a second multi-hot code. The first multi-hot code is then input into the embedding layer, and embedding is used to obtain the first feature vector. Similarly, the second multi-hot code is input into the embedding layer, and embedding is used to obtain the second feature vector.

[0052] For numerical continuous features, such as user age, number of views for each product, and number of clicks on a product in the past week, this application can retrieve cold start samples (i.e., the initial feedback dataset) and divide them into multiple subsets, i.e., bucketing. During bucketing, the data in each bucket can be of equal or unequal quantity. Then, the first feature information is allocated to each subset, either evenly or randomly. Each subset is discretized to obtain the third multi-hot encoding. The initial feedback dataset is further divided into multiple subsets, preferably using equal-frequency bucketing. The second feature information is also allocated to each subset, preferably evenly. Each subset is then discretized to obtain the fourth multi-hot encoding. This enhances the statistical significance of the features and avoids the sensitivity to maximum / minimum values ​​inherent in equal-distance bucketing.

[0053] The aforementioned equal-frequency binning means that the number of samples in each bin (sub-dataset) is the same or similar. After binning the feature information, segmentation points can be extracted, and then this type of feature can be discretized. For example, if a feature A belongs to the 7th segment of a 10-segment bin, this application can use a vector of length 10 to record this feature, and mark the 7th value in the vector as 1 to represent the current feature A. Then, the third multi-hot encoding is input into the embedding layer, and the first feature vector is obtained using Embedding. The fourth multi-hot encoding is input into the embedding layer, and the second feature vector is obtained using Embedding.

[0054] In this embodiment of the application, after discretizing the feature information, a 0 / 1 vector is obtained, namely the first multi-hot encoding, the second multi-hot encoding, the third multi-hot encoding and the fourth multi-hot encoding mentioned above. The multi-hot encoding is then input into the embedding layer to obtain the feature vector.

[0055] Optionally, before obtaining the initial feedback dataset, the method further includes constructing the initial feedback dataset as follows:

[0056] Determine the initial user base;

[0057] Divide the initial user base into multiple targeting groups;

[0058] Serve at least one product to each delivery group;

[0059] Obtain user feedback data for the advertised products in each ad group to obtain the initial feedback dataset.

[0060] In this embodiment, for software platforms using insurance services for the first time, the system automatically identifies these first-time users as initial users, assigns them to an initial user group, and further divides this initial user group into multiple delivery groups. At least one product is delivered to each delivery group, and feedback data on the delivered products from users in each delivery group is obtained, resulting in an initial feedback dataset, also known as a cold start sample. If the user feedback data indicates that the user dislikes the product initially delivered, it means that the user's preference does not match the initially delivered product, i.e., a conversion has occurred, and the user's preference is marked as 1. Conversely, if no conversion occurs and the initially delivered product matches the user's preference, it is marked as 0.

[0061] Step S304: Using the first feature vector and the second feature vector, determine the cross features and related features between the first feature information and the second feature information. The cross features and related features are used to represent the relationship between the target object and the product to be delivered.

[0062] In this embodiment of the application, the first feature vector and the second feature vector can be input into the dual-tower feature network to utilize the dual-tower feature network to mine the relationship between the product to be launched and the target object.

[0063] In the embodiments of this application, such as Figure 4 As shown, the dual-tower feature network includes a cross-network and a deep network. The cross-network, referencing the FM (Factor Machine) model, automatically captures the correlation between user-side and product-side features through feature cross-analysis. For example, it learns that users aged 50 prefer WeSure vouchers. The cross-network transforms each feature into a vector and automatically learns the correlation between any two vector pairs. For instance, the features of being 50 years old and liking WeSure vouchers are vectorized into a pair, and the correlation between these pairs is learned. Of course, any two vectors can be paired; those with low correlation are automatically ignored. The deep network contains multiple fully connected layers to capture deeper, non-linear relationships between users and products. For example, it uses a form like wx+b to non-linearly combine input features. The more fully connected layers in a deep network, the stronger its ability to capture non-linearity.

[0064] Optionally, step S304 may further include the following steps:

[0065] The first feature vector and the second feature vector are input into the cross network, and the cross features output by the cross network are obtained. The cross layer of the cross network is used to determine the cross features between the first feature vector and the second feature vector. The relationship includes the relationship represented by the cross features.

[0066] The first feature vector and the second feature vector are input into the deep network, and the associated features output by the deep network are obtained. At least one fully connected layer in the deep network is used to extract the associated features between the first feature vector and the second feature vector. The association relationship includes the relationship represented by the associated features.

[0067] Optionally, inputting the first feature vector and the second feature vector into the cross-network and obtaining the cross features output by the cross-network includes:

[0068] The first and second feature vectors are passed to the cross layer;

[0069] In the cross layer, extract co-occurrence features that appear in the first feature vector and appear in pairs in the second feature vector;

[0070] The weights of co-occurrence features are determined based on their total proportion in the first and second feature vectors.

[0071] First-order and second-order features are determined using co-occurrence features and weights, resulting in cross features, which include both first-order and second-order features.

[0072] In this embodiment, the length of the multi-hot encoded vector can be denoted as M. By default, the dimension of the feature vector can be set; preferably, the feature vector can be set to 128 dimensions. Figure 4 As shown, the first feature vector is partially input into the cross-network and partially into the deep network, and the second feature vector is similarly input. Therefore, after extracting the corresponding vectors, this application processes the vectors to be input into the cross-network and the deep network separately: For the vector input into the cross-network, no processing is performed, and it is denoted as Hi, then the vector dimension is M*128; for the vector input into the deep network, all vectors are summed to represent the feature vector of the current input sample, denoted as X. i The specific calculation method is as follows:

[0073]

[0074] Among them, emb j This represents the obtained eigenvectors, namely the first eigenvector and the second eigenvector.

[0075] In this embodiment, the method of capturing cross features in the cross network by referring to the FM model is used to optimize feature cross calculation. After the feature vector is input into the cross layer of the cross network, the cross network can automatically learn: (1) which features have co-occurrence responses, i.e., features that match or appear in pairs; (2) how important these co-occurrence response features are, i.e., the magnitude of their weights. At the same time, this application optimizes the original FM algorithm, no longer using element-level feature cross, but using vector-level feature cross. Refer to the FM formula:

[0076]

[0077] In this application, the embedding layer input is used as the cross term, i.e. and and Let w be an element in Hi, and w be a first-order weight. As an implicit vector, The first-order features mentioned above are the cross features output by the cross network:

[0078] <w,H i >

[0079] The above second-order features are:

[0080]

[0081] The advantage of this approach is that it can capture the cross features between feature vectors, making the statistical significance of feature crosses more significant and reducing noise. At the same time, it reduces the number of latent vectors that need to be calculated from M*128 to M, which greatly reduces the model complexity, reduces the time cost of online prediction, and improves the efficiency of online prediction.

[0082] Optionally, inputting the first feature vector and the second feature vector into the deep network and obtaining the associated features output by the deep network includes:

[0083] The first feature vector and the second feature vector are fitted by target regularization to determine the associated features while avoiding overfitting.

[0084] In this embodiment, sparsity in deep networks can be controlled, specifically through L1 regularization to avoid feature overfitting and enhance the model's generalization ability. The specific calculation method for L1 regularization is as follows:

[0085] l1-norm=λ·|Ω|

[0086] Where λ represents the penalty coefficient, used to control the degree of penalty; |Ω| represents the sum of the absolute values ​​of all parameters in the model, this term is used to characterize the complexity of the model. Let vector X... i Inputting data into a deep network allows the network to capture deeper, non-linear relationships between users and products, thereby outputting correlation features.

[0087] In deep networks, if a DNN consists of M fully connected layers, the output of the j-th fully connected layer is: Then the output of the (j+1)th layer is:

[0088]

[0089] Among them, W j b is the weight term of the j-th layer. j Let σ be the bias term of the j-th layer, and σ be the activation function, which can be ReLU, as shown below:

[0090]

[0091] The correlation features output by the deep network are the output of the last fully connected layer:

[0092]

[0093] Step S306: Concatenate the cross features and related features into the target feature vector.

[0094] Optionally, step S306, concatenating the cross features and associated features into the target feature vector, may further include:

[0095] Pass the cross-features and associated features to the connection layer;

[0096] Obtain the target feature vector obtained by concatenating the cross features and related features by the connection layer.

[0097] In this embodiment of the application, the connection layer can collect the output results of the dual-tower network, that is, the cross features output by the cross network. Correlation features with deep network output Among them, the cross features include first-order features:

[0098] <w,H i >

[0099] Second-order features:

[0100]

[0101] Therefore, the first-order features, second-order features, and associated features of the cross features can be concatenated into the target feature vector K. i The details are as follows:

[0102] q 1,i = <w,H i >

[0103]

[0104]

[0105] Where concat represents the join operation, q 1,i Let q represent the first-order feature of sample i in the cross features. 2,i This represents the second-order feature of sample i in the cross features.

[0106] Step S308: Determine the preference result based on the target feature vector.

[0107] In this embodiment of the application, the target feature vector K can be... i The input is fed into the loss layer, where the final output is calculated using the target loss function. This application modifies the loss function, directly employing the softmax function to calculate the probability output p. i (X i ∈G j ), where G j If p represents the j-th category, then i It can be represented as:

[0108]

[0109] Where exp represents the natural index. The weights of the last softmax layer. These are the bias parameters for the last softmax layer.

[0110] When calculating the loss function, you can directly refer to the calculation method of the softmax function. Let the final output label be y. i Then the objective loss function can be expressed as:

[0111]

[0112] Where N represents the number of samples, y i This represents the true classification result of the current sample, and log represents the natural logarithm. The labels output above indicate the products to be launched that are consistent with or similar to the user's preferences and habits.

[0113] In this embodiment, user characteristics can be analyzed to assign feature tags to users, thereby reducing the target range and facilitating the formulation of operational activity strategies and precise targeting. This further stimulates users to achieve their desired business goals and improves business achievement efficiency.

[0114] Step S206: Determine the target product mix from the products to be launched based on the preference results.

[0115] Optionally, the preference results include the target object's preference for each product to be offered (i.e., the probability output p calculated by the softmax function above). i The process of determining the target product mix from the products to be launched based on preference results includes:

[0116] The multiple preferences are sorted according to the target order to obtain a preference sequence;

[0117] Select the target number of preferences from the preference sequence in sorted order;

[0118] The products to be deployed corresponding to each preference level are taken as target products, and the target number of target products are sorted and combined according to the order of their corresponding preference levels to obtain the target product combination.

[0119] In this embodiment of the application, the preference results of each product to be deployed and the user are calculated through the above steps, that is, the preference degree of the target object for the product to be deployed and the corresponding tags. Then, this application can sort the preference degree in a certain order, such as from large to small, or from small to large, and select the top 3 products with the largest preference degree as the target products, and combine them according to the corresponding preference degree sorting order to obtain the target product combination.

[0120] Step S208: Display a target display card that matches the target product combination to the target audience.

[0121] In this embodiment, the target products corresponding to the target product combination are determined, and the data packets corresponding to each target product are obtained. The data packets include display styles and corresponding virtual resources. Virtual resources refer to resources that can be used to offset consumption after the user triggers (click, touch, swipe, etc.) through the front-end page. The data packets corresponding to each target product are combined, and the interaction displayed on the front end is simplified into a card form. Furthermore, the card content information (target product combination) may include prize images, prize descriptions, and corresponding jump links or virtual function buttons. The user can directly claim the prize after clicking or jump to a landing page to claim the prize.

[0122] like Figure 5 As shown, the main content W1 on the display page is the same, but the target product combination in the specific recommendation area W2 changes according to the user's corresponding targeting strategy. If the calculated preference results for the first user indicate that gifts A, B, and C are the top three in terms of preference from highest to lowest, then targeting strategy 1 is adopted for the first user, determining display combination 1, and the corresponding interface is ultimately displayed to the first user. Similarly, if the second user matches targeting strategy 2, then display combination 2 is displayed to the second user, and if the third user matches targeting strategy 3, then display combination 3 is displayed to the third user. The target product combination matches user preferences, increasing the accuracy of product targeting and thus improving the conversion rate.

[0123] According to another aspect of the embodiments of this application, such as Figure 6 As shown, a product dispensing device is provided, comprising:

[0124] The information acquisition module 601 is used to acquire the first feature information of the target object and the second feature information of the product to be deployed. The first feature information is used to represent the historical behavior of the target object on the target platform, and the second feature information represents the product features of the product to be deployed. The product to be deployed is a virtual object resource.

[0125] The preference matching module 603 is used to extract the cross features and correlation features between the first feature information and the second feature information, and to use the cross features and correlation features to determine the target object's preference results for the product to be delivered;

[0126] The combination determination module 605 is used to determine the target product combination from the products to be deployed based on the preference results.

[0127] The delivery and display module 607 is used to display target display cards that match the target product combination to the target audience.

[0128] It should be noted that the information acquisition module 601 in this embodiment can be used to execute step S202 in this application embodiment, the preference matching module 603 in this embodiment can be used to execute step S204 in this application embodiment, the combination determination module 605 in this embodiment can be used to execute step S206 in this application embodiment, and the delivery display module 607 in this embodiment can be used to execute step S208 in this application embodiment.

[0129] It should be noted that the examples and application scenarios implemented by the above modules and corresponding steps are the same, but are not limited to the content disclosed in the above embodiments. It should also be noted that the above modules, as part of a device, can operate in situations such as... Figure 1 The hardware environment shown can be implemented either through software or through hardware.

[0130] Optionally, the preference matching module includes:

[0131] An embedding unit is used to convert first feature information into a first feature vector and second feature information into a second feature vector. The first feature information includes at least one of numerical feature information and attribute feature information of the target object, and the second feature information includes at least one of numerical feature information and attribute feature information of the product to be delivered.

[0132] The feature extraction unit is used to determine the cross features and correlation features between the first feature information and the second feature information using the first feature vector and the second feature vector. The cross features and correlation features are used to represent the correlation between the target object and the product to be delivered.

[0133] The vector concatenation unit is used to concatenate cross features and related features into a target feature vector;

[0134] The preference result determination unit is used to determine the preference result based on the target feature vector.

[0135] Optionally, the embedded unit includes:

[0136] The data type determination subunit is used to determine the data type of the first feature information and the second feature information;

[0137] The first embedding subunit is used to discretize the first feature information to obtain a first multi-hot code when the data type is an attribute class; input the first multi-hot code into the embedding layer to obtain a first feature vector obtained by vectorizing the first multi-hot code by the embedding layer; discretize the second feature information to obtain a second multi-hot code; input the second multi-hot code into the embedding layer to obtain a second feature vector obtained by vectorizing the second multi-hot code by the embedding layer.

[0138] The second embedding subunit, when the data type is numerical, is used to obtain the initial feedback dataset, which is the behavioral data generated when the user first logs into the target platform; divide the initial feedback dataset into multiple subsets and assign the first feature information to each subset; discretize each subset to obtain a third multi-hot encoding; input the third multi-hot encoding into the embedding layer to obtain the first feature vector obtained by the embedding layer vectorizing the third multi-hot encoding, the length of the first feature vector matching the number of subsets; divide the initial feedback dataset into multiple subsets and assign the second feature information to each subset; discretize each subset to obtain a fourth multi-hot encoding; input the fourth multi-hot encoding into the embedding layer to obtain the second feature vector obtained by the embedding layer vectorizing the fourth multi-hot encoding, the length of the second feature vector matching the number of subsets.

[0139] Optionally, the feature extraction unit includes:

[0140] The cross network subunit is used to input the first feature vector and the second feature vector into the cross network and obtain the cross features output by the cross network. The cross layer of the cross network is used to determine the cross features between the first feature vector and the second feature vector. The relationship includes the relationship represented by the cross features.

[0141] A deep network subunit is used to input a first feature vector and a second feature vector into a deep network and obtain the associated features output by the deep network. At least one fully connected layer in the deep network is used to extract the associated features between the first feature vector and the second feature vector. The associated relationship includes the relationship represented by the associated features.

[0142] Optionally, the vector concatenation unit also includes:

[0143] The output transfer subunit is used to transfer cross features and associated features to the connection layer;

[0144] The concatenation vector acquisition sub-unit is used to obtain the target feature vector obtained by concatenating the cross features and related features by the connection layer.

[0145] Optionally, the combination determination module includes:

[0146] The preference sorting unit is used to sort multiple preferences according to the target order to obtain a preference sequence;

[0147] The preference selection unit is used to select a target number of preferences from the preference sequence in sorted order.

[0148] The combination determination unit is used to take the products to be deployed corresponding to each preference degree as target products, and sort and combine the target number of target products according to the order of their corresponding preference degrees to obtain the target product combination.

[0149] According to another aspect of the embodiments of this application, this application provides an electronic device, such as... Figure 7 As shown, the device includes a memory 701, a processor 703, a communication interface 705, and a communication bus 707. The memory 701 stores a computer program that can run on the processor 703. The memory 701 and the processor 703 communicate through the communication interface 705 and the communication bus 707. When the processor 703 executes the computer program, it implements the steps of the above method.

[0150] The memory and processor in the aforementioned electronic devices communicate with each other via a communication bus and a communication interface. The communication bus can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus can be divided into an address bus, a data bus, a control bus, etc.

[0151] The memory may include random access memory (RAM) or non-volatile memory, such as at least one disk storage device. Optionally, the memory may also be at least one storage device located remotely from the aforementioned processor.

[0152] The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0153] According to another aspect of the embodiments of this application, a computer program product or computer program is also provided, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the steps of any of the above embodiments.

[0154] Optionally, in embodiments of this application, the computer program product or computer program is program code used by the processor to perform the following steps:

[0155] Obtain the first characteristic information of the target object and the second characteristic information of the product to be deployed. The first characteristic information is used to represent the historical behavior of the target object on the target platform, and the second characteristic information is used to represent the product characteristics of the product to be deployed.

[0156] The first and second feature information are used to determine the target audience's preference for the product to be delivered.

[0157] Based on the preference results, the target products to be delivered to the target audience are determined from the products to be delivered.

[0158] Optionally, specific examples in this embodiment can refer to the examples described in the above embodiments, and will not be repeated here.

[0159] In specific implementation, the embodiments of this application can be referred to the above embodiments and have corresponding technical effects.

[0160] It is understood that the embodiments described herein can be implemented in hardware, software, firmware, middleware, microcode, or a combination thereof. For hardware implementation, the processing unit can be implemented in one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), general-purpose processors, controllers, microcontrollers, microprocessors, other electronic units for performing the functions described herein, or combinations thereof.

[0161] For software implementation, the techniques described herein can be implemented by units that perform the functions described herein. The software code can be stored in memory and executed by a processor. The memory can be implemented in the processor or external to the processor.

[0162] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0163] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0164] In the embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection through some interfaces, devices, or units, and may be electrical, mechanical, or other forms.

[0165] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0166] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0167] If the aforementioned function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiments of this application, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, ROM, RAM, magnetic disks, or optical disks. It should be noted that in this document, relational terms such as "first" and "second" are used only 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. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. In the absence of further restrictions, an element defined by the phrase "comprising a..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0168] The above description is merely a specific embodiment of this application, enabling those skilled in the art to understand or implement this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features claimed herein.

Claims

1. A product placement method, characterized by, The method comprises the following steps: obtaining first characteristic information of a target object and second characteristic information of a product to be launched, wherein the first characteristic information is used to represent historical behaviors of the target object on a target platform, and the second characteristic information is used to represent product characteristics of the product to be launched, and the product to be launched comprises virtual object resources for virtual resource exchange; extracting cross characteristics and associated characteristics between the first characteristic information and the second characteristic information, and determining a preference result of the target object for the product to be launched by using the cross characteristics and the associated characteristics; determining a target product combination from the product to be launched according to the preference result; displaying a target display card matched with the target product combination to the target object; the step of extracting the cross characteristics and the associated characteristics between the first characteristic information and the second characteristic information, and determining the preference result of the target object for the product to be launched by using the cross characteristics and the associated characteristics comprises the following steps: converting the first characteristic information into a first characteristic vector, and converting the second characteristic information into a second characteristic vector, wherein the first characteristic information comprises at least one of numerical characteristic information and attribute characteristic information of the target object, and the second characteristic information comprises at least one of numerical characteristic information and attribute characteristic information of the product to be launched; determining the cross characteristics and the associated characteristics between the first characteristic information and the second characteristic information by using the first characteristic vector and the second characteristic vector, wherein the cross characteristics and the associated characteristics are used to represent an association relationship between the target object and the product to be launched; splicing the cross characteristics and the associated characteristics into a target characteristic vector; and determining the preference result according to the target characteristic vector; The converting the first feature information into a first feature vector and the converting the second feature information into a second feature vector comprises: determining a data type of the first feature information and the second feature information; in a case where the data type is the attribute type, discretizing the first feature information to obtain a first one-hot code; inputting the first one-hot code into an embedding layer to obtain the first feature vector obtained by vectorizing the first one-hot code by the embedding layer; discretizing the second feature information to obtain a second one-hot code; inputting the second one-hot code into the embedding layer to obtain the second feature vector obtained by vectorizing the second one-hot code by the embedding layer; in a case where the data type is the numerical type, obtaining an initial feedback data set, wherein the initial feedback data set is behavior data generated when a user logs in the target platform for the first time; dividing the initial feedback data set into a plurality of sub-data sets, and classifying the first feature information into each of the sub-data sets; discretizing each of the sub-data sets to obtain a third one-hot code; inputting the third one-hot code into the embedding layer to obtain the first feature vector obtained by vectorizing the third one-hot code by the embedding layer, wherein a vector length of the first feature vector matches a number of the sub-data sets; dividing the initial feedback data set into a plurality of sub-data sets, and classifying the second feature information into each of the sub-data sets; discretizing each of the sub-data sets to obtain a fourth one-hot code; inputting the fourth one-hot code into the embedding layer to obtain the second feature vector obtained by vectorizing the fourth one-hot code by the embedding layer, wherein a vector length of the second feature vector matches the number of the sub-data sets.

2. The method of claim 1, wherein, The determining the cross feature and the association feature between the first feature information and the second feature information by using the first feature vector and the second feature vector comprises: inputting the first feature vector and the second feature vector into a cross network, and obtaining the cross feature output by the cross network, wherein a cross layer of the cross network is configured to determine the cross feature between the first feature vector and the second feature vector, and the association relationship comprises a relationship represented by the cross feature; inputting the first feature vector and the second feature vector into a deep network, and obtaining the association feature output by the deep network, wherein at least one fully connected layer in the deep network is configured to extract the association feature between the first feature vector and the second feature vector, and the association relationship comprises a relationship represented by the association feature.

3. The method of claim 2, wherein, The inputting the first feature vector and the second feature vector into the cross network and obtaining the cross feature output by the cross network comprises: passing the first feature vector and the second feature vector to the cross layer; extracting, in the cross layer, a co-occurrence feature that appears in the first feature vector and is pairedly present in the second feature vector; determine a weight of the co-occurrence feature according to a total proportion of the co-occurrence feature in the first feature vector and the second feature vector; determine a first-order feature and a second-order feature by using the co-occurrence feature and the weight, to obtain a cross feature, wherein the cross feature includes the first-order feature and the second-order feature.

4. The method of claim 1, wherein, concatenating the cross feature and the association feature into a target feature vector includes: delivering the cross feature and the association feature to a connection layer; obtaining the target feature vector obtained by concatenating the cross feature and the association feature by the connection layer.

5. The method according to any one of claims 1 to 4, characterized in that, The preference result includes a preference degree of each of the to-be-launched products by the target object, and determining a target product combination from the to-be-launched products according to the preference result includes: sorting a plurality of the preference degrees according to a target order to obtain a preference degree sequence; selecting a target number of the preference degrees from the preference degree sequence according to the sorting order; arranging the target number of the target products corresponding to the preference degrees according to the arrangement order of the preference degrees to obtain the target product combination.

6. A product dispensing apparatus for implementing the product dispensing method according to any one of claims 1 to 5, characterized by comprise: an information acquisition module, configured to acquire first feature information of a target object and second feature information of a to-be-launched product, wherein the first feature information is used to represent historical behaviors of the target object on a target platform, and the second feature information is used to represent product features of the to-be-launched product, and the to-be-launched product includes virtual object resources for virtual resource exchange; a preference matching module, configured to extract cross features and association features between the first feature information and the second feature information, and determine a preference result of the target object for the to-be-launched product by using the cross features and the association features; a combination determination module, configured to determine a target product combination from the to-be-launched products according to the preference result; a launch display module, configured to display a target display card matched with the target product combination to the target object.

7. An electronic device comprising a memory, a processor, a communication interface and a communication bus, the memory storing a computer program executable on the processor, the memory, the processor communicating through the communication bus and the communication interface, characterized in that, The processor executes the computer program to implement the method of any one of claims 1 to 5.

8. A computer readable medium having a non-transitory program code executable by a processor, the program code comprising instructions for: The program code causes the processor to execute the method of any one of claims 1 to 5.

Images