Click rate prediction method, system, computer and readable storage medium

Abstract

The present invention provides a click-through rate prediction method, system, computer and readable storage medium, the method comprising: converting a user ID and a product ID into corresponding user embedding vectors and product embedding vectors that can be input into a neural network respectively through an embedding model, and constructing a corresponding user-product interaction graph based on the user ID and the rating matrix in the product ID; performing multiple rounds of iterative information propagation in the user-product interaction graph through a graph neural network, and inputting the user embedding vector and the product embedding vector in its historical behavior into a preset interaction layer and a logic deduction layer to obtain a user interest vector; constructing a click-through rate prediction layer based on a fully connected neural network, and inputting the user interest vector into the click-through rate prediction layer to estimate the click-through rate. The present application can estimate the click-through rate more accurately, greatly improving the accuracy of the prediction results.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a click rate prediction method, system, computer and readable storage medium. Background Art

[0002] With the continuous improvement of computer computing power, online advertising systems have achieved reach efficiency and actual benefits far exceeding traditional advertising models. In the actual e-commerce environment, due to the wide distribution of user groups, diverse needs, and rich types of advertisements, showing similar advertisements to different users through fixed rules greatly limits the revenue capacity of online advertising systems. Click-through rate prediction technology combines the characteristics of different user groups and the advertisements to be distributed, uses neural networks or comprehensive models to perform parameter training on historical data, and then completes the task of predicting the click probability of specific advertisements for different users. It has a wide range of application scenarios in real life.

[0003] In the task of click-through rate prediction, users often have historical records, which not only retain the user's choice of products or diversified features for a long time, but also have rich interest IDs. In practical applications, user interest IDs play an important role in many tasks. For example, in product recommendation, whether a user will buy a product is not only related to his own attributes, but also closely related to the user's recent purchase or browsing history.

[0004] However, most of the existing click-through rate prediction methods focus on the matching degree between the target product and the user's historical behavior, while ignoring the importance of the user's potential needs in the prediction task, which makes the prediction results have certain limitations and reduces the accuracy of the prediction results. Summary of the invention

[0005] Based on this, the purpose of the present invention is to provide a click-through rate prediction method, system, computer and readable storage medium to solve the problem that most of the click-through rate prediction methods in the prior art focus on the matching degree between the target product and the user's historical behavior, but ignore the importance of the user's potential needs in the prediction task, thereby making the prediction results have certain limitations.

[0006] A first aspect of an embodiment of the present invention provides a click rate prediction method, the method comprising:

[0007] The user ID and the product ID are respectively converted into corresponding user embedding vectors and product embedding vectors that can be input into the neural network through the embedding model, and a corresponding user-product interaction graph is constructed based on the rating matrix in the user ID and the product ID. The user-product interaction graph includes a plurality of nodes, each of which represents a user or a product, and two nodes with an interactive relationship are connected by an edge. The neural network includes a graph neural network and a fully connected neural network;

[0008] Perform multiple rounds of iterative information propagation in the user-product interaction graph through the graph neural network, and input the user embedding vector and the product embedding vector in its historical behavior into a preset interaction layer and a logic deduction layer to obtain a user interest vector;

[0009] A click rate prediction layer is constructed based on the fully connected neural network, and the user interest vector is input into the click rate prediction layer to estimate the click rate.

[0010] The beneficial effects of the present invention are as follows: the click-through rate prediction method proposed in the present application proposes for the first time to simultaneously learn the collaborative information between the user ID and the product ID and the logical connection between the user-interacted products in the same model structure, and finally extract the interest information from the user behavior sequence. At the same time, the present application also emphasizes the importance of mining user preferences and needs in the user interest learning process, and proposes an effective user interest learning method to simultaneously model user preferences and needs, and apply it to the click-through rate prediction task. And the present application can effectively utilize the deep collaborative information in the user-product interaction graph and the logical connection in the user behavior sequence to mine user preferences and needs, so as to be able to more accurately estimate the click-through rate, greatly improving the accuracy of the estimation results, and is suitable for large-scale promotion and use.

[0011] Preferably, the step of converting the user ID and the product ID into corresponding user embedding vectors and product embedding vectors that can be input into the neural network through the embedding model, and constructing a corresponding user-product interaction graph based on the rating matrix in the user ID and the product ID includes:

[0012] The embedding table E is initialized for the user ID and the product ID respectively through the embedding model u and E i , and calculate the user embedding vector and the product embedding vector respectively according to a first preset formula, wherein the first preset formula is:

[0013]

[0014] Among them, e 0u represents the embedding vector corresponding to the user ID, e 0 i represents the embedding vector corresponding to the product ID, t u represents the one-hot encoding of the user ID, t i A one-hot encoding representing the product ID;

[0015] The user-product interaction graph G=<V, E> is constructed based on the user ID and the rating matrix in the product ID, wherein V represents a node set and E represents an edge set.

[0016] Preferably, the step of performing multiple rounds of iterative information propagation in the user-product interaction graph through the graph neural network includes:

[0017] In the user-product interaction graph, information of the embedding vector of one of the two connected nodes is transmitted to the other node through an information propagation function to update the user embedding vector or the product embedding vector corresponding to each node.

[0018] Preferably, the step of inputting the user embedding vector and the product embedding vector in the user's historical behavior into a preset interaction layer and a logic deduction layer to obtain the user interest vector includes:

[0019] Inputting the product embedding vectors in the user's historical behavior and the product embedding vectors in the target advertisement into the interaction layer in a preset order, and calculating the corresponding interaction vector according to a second preset formula;

[0020] The interaction vector is input into the logic deduction layer, and a corresponding user interest vector is calculated according to a third preset formula.

[0021] Preferably, the second preset formula is:

[0022]

[0023] Among them, [|] represents the splicing operation, W 1 , W 2 , b 1 , b 2 Both represent trainable parameters, e i represents the product embedding vector after the product ID is updated, e u represents the user embedding vector after the user ID is updated, σ 2 Represents the sigmoid activation function;

[0024] The third preset formula is:

[0025]

[0026] Wherein, O represents the user interest vector, represents a trainable parameter, ReLU represents the ReLU activation function, and h n represents the hidden state obtained at time n in the logic derivation layer, It represents the interaction vector obtained after the product embedding vector and the user embedding vector input at the next moment n+1 are input into the interaction layer.

[0027] A second aspect of an embodiment of the present invention provides a click rate prediction system, the system comprising:

[0028] A conversion module, used to convert the user ID and the product ID into corresponding user embedding vectors and product embedding vectors that can be input into the neural network through the embedding model, and construct a corresponding user-product interaction graph based on the rating matrix in the user ID and the product ID, wherein the user-product interaction graph includes a plurality of nodes, each of which represents a user or a product, and two nodes with an interactive relationship are connected by an edge, and the neural network includes a graph neural network and a fully connected neural network;

[0029] An analysis module, configured to perform multiple rounds of iterative information propagation in the user-product interaction graph through the graph neural network, and input the user embedding vector and the product embedding vector in its historical behavior into a preset interaction layer and a logic deduction layer to obtain a user interest vector;

[0030] The estimation module is used to construct a click rate estimation layer based on the fully connected neural network, and input the user interest vector into the click rate estimation layer to estimate the click rate.

[0031] Wherein, in the above-mentioned click rate prediction system, the conversion module is specifically used for:

[0032] The embedding table E is initialized for the user ID and the product ID respectively through the embedding model u and E i , and calculate the user embedding vector and the product embedding vector respectively according to a first preset formula, wherein the first preset formula is:

[0033]

[0034] Among them, e 0 u represents the embedding vector corresponding to the user ID, e 0 i represents the embedding vector corresponding to the product ID, t u represents the one-hot encoding of the user ID, ti A one-hot encoding representing the product ID;

[0035] The user-product interaction graph G is constructed based on the rating matrix in the user ID and the product ID.<V,E> , where V represents the node set and E represents the edge set.

[0036] In the above-mentioned click rate prediction system, the analysis module is specifically used for:

[0037] In the user-product interaction graph, information of the embedding vector of one of the two connected nodes is transmitted to the other node through an information propagation function to update the user embedding vector or the product embedding vector corresponding to each node.

[0038] In the above-mentioned click rate prediction system, the analysis module is further specifically used for:

[0039] Inputting the product embedding vectors in the user's historical behavior and the product embedding vectors in the target advertisement into the interaction layer in a preset order, and calculating the corresponding interaction vector according to a second preset formula;

[0040] The interaction vector is input into the logic deduction layer, and a corresponding user interest vector is calculated according to a third preset formula.

[0041] Among them, in the above-mentioned click rate prediction system, the second preset formula is:

[0042]

[0043] Among them, [|] represents the splicing operation, W 1 , W 2 、b 1 、b 2 Both represent trainable parameters, e i represents the product embedding vector after the product ID is updated, e u represents the user embedding vector after the user ID is updated, σ 2 Represents the sigmoid activation function;

[0044] The third preset formula is:

[0045]

[0046] Wherein, O represents the user interest vector, represents a trainable parameter, ReLU represents the ReLU activation function, and h n represents the hidden state obtained at time n in the logic derivation layer, It represents the interaction vector obtained after the product embedding vector and the user embedding vector input at the next moment n+1 are input into the interaction layer.

[0047] A third aspect of an embodiment of the present invention provides a computer, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the click rate prediction method described above when executing the computer program.

[0048] A fourth aspect of an embodiment of the present invention provides a readable storage medium on which a computer program is stored. When the program is executed by a processor, the click rate prediction method as described above is implemented.

[0049] Additional aspects and advantages of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 A flowchart of a click rate prediction method provided by the first embodiment of the present invention;

[0051] Figure 2 This is a structural block diagram of a click rate prediction system provided in the third embodiment of the present invention.

[0052] The following specific implementation manner will further illustrate the present invention in conjunction with the above-mentioned drawings. DETAILED DESCRIPTION

[0053] In order to facilitate the understanding of the present invention, the present invention will be described more fully below with reference to the relevant drawings. Several embodiments of the present invention are given in the drawings. However, the present invention can be implemented in many different forms and is not limited to the embodiments described herein. On the contrary, the purpose of providing these embodiments is to make the disclosure of the present invention more thorough and comprehensive.

[0054] It should be noted that when an element is referred to as being "fixed to" another element, it may be directly on the other element or there may be a central element. When an element is considered to be "connected to" another element, it may be directly connected to the other element or there may be a central element at the same time. The terms "vertical", "horizontal", "left", "right" and similar expressions used herein are for illustrative purposes only.

[0055] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art of the present invention. The terms used herein in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The term "and / or" used herein includes any and all combinations of one or more related listed items.

[0056] Most of the existing click-through rate prediction methods focus on the matching degree between the target product and the user's historical behavior, but ignore the importance of the user's potential needs in the prediction task, which makes the prediction results have certain limitations and reduces the accuracy of the prediction results.

[0057] See also Figure 1 , shown is the click-through rate prediction method provided by the first embodiment of the present invention. The click-through rate prediction method provided by this embodiment can effectively utilize the deep collaborative information in the user-product interaction graph and the logical connections in the user behavior sequence to explore the user's preferences and needs, so as to more accurately estimate the click-through rate, greatly improve the accuracy of the prediction results, and is suitable for large-scale promotion and use.

[0058] Specifically, the click rate prediction method provided in this embodiment includes the following steps:

[0059] Step S10, converting the user ID and the product ID into corresponding user embedding vectors and product embedding vectors that can be input into the neural network through the embedding model, and constructing a corresponding user-product interaction graph based on the rating matrix in the user ID and the product ID, wherein the user-product interaction graph includes a plurality of nodes, each of which represents a user or a product, and two nodes with an interactive relationship are connected by an edge, and the neural network includes a graph neural network and a fully connected neural network;

[0060] In this embodiment, it should be noted that the click-through rate prediction method provided in this embodiment is implemented based on an embedding model, a neural network, and a collaborative filtering model. Among them, the embedding model is widely used in the field of deep learning. The embedding model can convert high-dimensional, sparse features into low-dimensional dense vectors, provide initial input for the deep neural network, and enable the deep model to learn and converge better. Among them, the most common practice in the field of click-through rate prediction is to create a trainable nxd-dimensional embedding table for each feature, where n represents all types of the feature, and convert the feature into a d-dimensional vector by looking up the table.

[0061] In addition, the core idea of ​​the above collaborative filtering model is "Birds of a feather flock together". Analyze user interests, find similar users to the specified user in the user group, and integrate the evaluations of these similar users on a certain product as collaborative information to form a model to predict the degree of preference of the specified user for this information. Similarly, you can also find similar products to the specified product, and integrate the user's evaluation of these similar products as collaborative information for prediction. The graph-based collaborative filtering model uses graphs to mine the relationship between products and products, users and users, and users and products. Compared with traditional collaborative filtering that calculates similarity through a rating matrix, graph-based collaborative filtering can mine collaborative information between each other at a deeper level.

[0062] Therefore, in this step, it should be noted that this step will convert the collected user ID and product ID into corresponding user embedding vectors and product embedding vectors that can be input into the neural network based on the created embedding model, where both the user ID and the product ID are unique.

[0063] Furthermore, in this step, it should be pointed out that the user ID and product ID collected in this embodiment both carry a rating matrix. Therefore, in this step, a corresponding user-product interaction graph will be further constructed based on the rating matrix in the user ID and product ID. Specifically, the user-product interaction graph includes a number of nodes, each of which represents a user or a product, and two nodes with an interactive relationship are connected by an edge, and the neural network includes a graph neural network and a fully connected neural network.

[0064] Step S20, performing multiple rounds of iterative information propagation in the user-product interaction graph through the graph neural network, and inputting the user embedding vector and the product embedding vector in its historical behavior into a preset interaction layer and a logic deduction layer to obtain a user interest vector;

[0065] Furthermore, in this step, it should be noted that after obtaining the user-product interaction graph, this step will perform secondary processing on the user-product interaction graph. Specifically, this step will perform multiple rounds of iterative information propagation in the user-product interaction graph through the above-mentioned graph neural network. Furthermore, this step will also input the obtained user embedding vector and the product embedding vector in its historical behavior into the preset interaction layer and logic deduction layer to obtain the user interest vector.

[0066] Step S30: construct a click rate prediction layer based on the fully connected neural network, and input the user interest vector into the click rate prediction layer to estimate the click rate.

[0067] Finally, in this step, it should be noted that this step will further construct a corresponding click-through rate prediction layer based on the created fully connected neural network, and input the above-obtained user interest vector into the click-through rate prediction layer to estimate the click-through rate and obtain the final prediction result.

[0068] When in use, the collaborative information between the user ID and the product ID and the logical connection between the user-interacted products are simultaneously learned in the same model structure, and interest information is finally extracted from the user behavior sequence. At the same time, this application also emphasizes the importance of mining user preferences and needs in the user interest learning process, and proposes an effective user interest learning method to model user preferences and needs at the same time, and apply it to the click-through rate prediction task. And this application can effectively use the deep collaborative information in the user-product interaction graph and the logical connection in the user behavior sequence to mine user preferences and needs, so that the click-through rate can be estimated more accurately, which greatly improves the accuracy of the prediction results and is suitable for large-scale promotion and use.

[0069] It should be noted that the above implementation process is only to illustrate the feasibility of the present application, but this does not mean that the click-through rate prediction method of the present application has only the above-mentioned single implementation process. On the contrary, as long as the click-through rate prediction method of the present application can be implemented, it can be included in the feasible implementation plan of the present application.

[0070] In summary, the click-through rate prediction method provided by the above embodiment of the present invention can effectively utilize the deep collaborative information in the user-product interaction graph and the logical connections in the user behavior sequence to explore the user's preferences and needs, so as to more accurately estimate the click-through rate, greatly improve the accuracy of the prediction results, and is suitable for large-scale promotion and use.

[0071] The second embodiment of the present invention also provides a click rate prediction method. The click rate prediction method provided by this embodiment specifically includes the following steps:

[0072] Similarly, in this embodiment, it should be first explained that the click rate prediction method provided in this embodiment is also implemented based on the embedding model, neural network and collaborative filtering model.

[0073] Step S11, initializing embedding table E for the user ID and the product ID respectively through the embedding model u and E i , and calculate the user embedding vector and the product embedding vector respectively according to a first preset formula, wherein the first preset formula is:

[0074]

[0075] Among them, e 0 u represents the embedding vector corresponding to the user ID, e 0 i represents the embedding vector corresponding to the product ID, t u represents the one-hot encoding of the user ID, t i A one-hot encoding representing the product ID;

[0076] The user-product interaction graph G=<V, E> is constructed based on the user ID and the rating matrix in the product ID, wherein V represents a node set and E represents an edge set.

[0077] Specifically, in this embodiment, it should be noted that this step will first initialize the embedding table E for the acquired user ID by using the created embedding model. u , and initialize the embedded table E for the obtained product ID i , further, the user embedding vector and the product embedding vector are calculated respectively according to the above first preset formula.

[0078] Furthermore, the present application will also construct the above-mentioned user-product interaction graph G=<V, E> according to the user ID and the rating matrix in the product ID, where V represents the node set and E represents the edge set.

[0079] Step S21, in the user-product interaction graph, the information of the embedding vector of one of the two connected nodes is transmitted to the other node through the information propagation function, so as to update the user embedding vector or the product embedding vector corresponding to each node.

[0080] Furthermore, in this step, this embodiment transmits the information of the embedding vector of one of the two connected nodes to the other node through the information propagation function in the acquired user-product interaction graph, so as to update the user embedding vector or product embedding vector corresponding to each node.

[0081] Specifically, in this step, the amount of information propagation from one node to another node in the above user-product interaction graph is calculated by the following formula:

[0082]

[0083] Among them, e h and e t The embedding vector corresponding to the outgoing node and the receiving node of the information, D h and D t They represent the degree of a node, that is, the number of nodes connected to it. s , Wi and b both represent trainable parameters.

[0084] Furthermore, this step will combine the information propagated by the adjacent nodes with its own information according to the following formula to obtain a new node vector.

[0085]

[0086] Among them, σ represents the sigmoid activation function iterates L times of information propagation, and each node will get L vectors And L vectors are finally represented as vector e by this node. t .

[0087] Step S31, inputting the product embedding vector in the user's historical behavior and the product embedding vector in the target advertisement into the interaction layer in a preset order, and calculating the corresponding interaction vector according to a second preset formula;

[0088] The interaction vector is input into the logic deduction layer, and a corresponding user interest vector is calculated according to a third preset formula.

[0089] Specifically, in this step, it should be noted that this step will input the product embedding vector in the current user's historical behavior record and the product embedding vector in its target advertisement into the interaction layer in a preset order, and calculate the corresponding interaction vector according to the following second preset formula.

[0090] Specifically, the second preset formula is:

[0091]

[0092] Among them, [|] represents the splicing operation, W 1 , W 2 、b 1 、b 2 Both represent trainable parameters, e i represents the product embedding vector after the product ID is updated, e u represents the user embedding vector after the user ID is updated, σ 2 Represents the sigmoid activation function;

[0093] Furthermore, after obtaining the interaction vectors, this step will sequentially input the inference network in the logic inference layer, and perform a non-operation on the interaction vectors corresponding to each behavior before inputting the loop structure according to the following formula:

[0094]

[0095] in, These are all trainable parameters.

[0096] Furthermore, an iterative calculation method of the derivation network is designed according to the following formula:

[0097]

[0098] in, The vector representing the product embedding vector at time t converted by the above formula.

[0099] Furthermore, in this step, the interaction vector is input into the logic deduction layer, and the corresponding user interest vector is calculated by the following third preset formula:

[0100] Specifically, the third preset formula is:

[0101]

[0102] Wherein, O represents the user interest vector, represents a trainable parameter, ReLU represents the ReLU activation function, and h n represents the hidden state obtained at time n in the logic derivation layer, It represents the interaction vector obtained after the product embedding vector and the user embedding vector input at the next moment n+1 are input into the interaction layer.

[0103] Step S41, constructing a click rate prediction layer based on the fully connected neural network, and inputting the user interest vector into the click rate prediction layer to estimate the click rate.

[0104] In this step, it should be noted that this step will further construct a corresponding click-through rate prediction layer based on the created fully connected neural network, and input the above-obtained user interest vector into the click-through rate prediction layer to estimate the click-through rate and obtain the final prediction result.

[0105] It should be pointed out that the implementation principle and some technical effects of the method provided in the second embodiment of the present invention are the same as those of the first embodiment. For the sake of brief description, for matters not mentioned in this embodiment, reference may be made to the corresponding contents provided in the first embodiment.

[0106] In summary, the click-through rate prediction method provided by the above embodiment of the present invention can effectively utilize the deep collaborative information in the user-product interaction graph and the logical connections in the user behavior sequence to explore the user's preferences and needs, so as to more accurately estimate the click-through rate, greatly improve the accuracy of the prediction results, and is suitable for large-scale promotion and use.

[0107] See also Figure 2 , which is a click rate prediction system provided by a third embodiment of the present invention, and the system comprises:

[0108] A conversion module 12 is used to convert the user ID and the product ID into corresponding user embedding vectors and product embedding vectors that can be input into the neural network through an embedding model, and construct a corresponding user-product interaction graph based on the rating matrix in the user ID and the product ID, wherein the user-product interaction graph includes a plurality of nodes, each of which represents a user or a product, and two nodes with an interactive relationship are connected by an edge, and the neural network includes a graph neural network and a fully connected neural network;

[0109] An analysis module 22, configured to perform multiple rounds of iterative information propagation in the user-product interaction graph through the graph neural network, and input the user embedding vector and the product embedding vector in its historical behavior into a preset interaction layer and a logic deduction layer to obtain a user interest vector;

[0110] The estimation module 32 is used to construct a click rate estimation layer based on the fully connected neural network, and input the user interest vector into the click rate estimation layer to estimate the click rate.

[0111] In the above-mentioned click rate prediction system, the conversion module 12 is specifically used for:

[0112] The embedding table E is initialized for the user ID and the product ID respectively through the embedding model u and E i , and calculate the user embedding vector and the product embedding vector respectively according to a first preset formula, wherein the first preset formula is:

[0113]

[0114] Among them, e 0 u represents the embedding vector corresponding to the user ID, e 0 i represents the embedding vector corresponding to the product ID, t u represents the one-hot encoding of the user ID, t i A one-hot encoding representing the product ID;

[0115] The user-product interaction graph G=<V, E> is constructed based on the user ID and the rating matrix in the product ID, wherein V represents a node set and E represents an edge set.

[0116] In the above-mentioned click rate prediction system, the analysis module 22 is specifically used for:

[0117] In the user-product interaction graph, information of the embedding vector of one of the two connected nodes is transmitted to the other node through an information propagation function to update the user embedding vector or the product embedding vector corresponding to each node.

[0118] In the above-mentioned click rate prediction system, the analysis module 22 is further specifically used for:

[0119] Inputting the product embedding vectors in the user's historical behavior and the product embedding vectors in the target advertisement into the interaction layer in a preset order, and calculating the corresponding interaction vector according to a second preset formula;

[0120] The interaction vector is input into the logic deduction layer, and a corresponding user interest vector is calculated according to a third preset formula.

[0121] Among them, in the above-mentioned click rate prediction system, the second preset formula is:

[0122]

[0123] Among them, [|] represents the splicing operation, W 1 , W 2 , b 1 , b 2 Both represent trainable parameters, e i represents the product embedding vector after the product ID is updated, e u represents the user embedding vector after the user ID is updated, σ 2 Represents the sigmoid activation function;

[0124] The third preset formula is:

[0125]

[0126] Wherein, O represents the user interest vector, represents a trainable parameter, ReLU represents a ReLU activation function, hn represents a hidden state obtained at time n in the logical derivation layer, It represents the interaction vector obtained after the product embedding vector and the user embedding vector input at the next moment n+1 are input into the interaction layer.

[0127] A fourth embodiment of the present invention provides a computer, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the click rate prediction method provided in the first embodiment or the second embodiment described above is implemented.

[0128] A fifth embodiment of the present invention provides a readable storage medium having a computer program stored thereon. When the program is executed by a processor, the click rate prediction method provided in the first embodiment or the second embodiment described above is implemented.

[0129] In summary, the click-through rate prediction method, system, computer and readable storage medium provided by the above embodiments of the present invention can effectively utilize the deep collaborative information in the user-product interaction diagram and the logical connections in the user behavior sequence to explore the user's preferences and needs, thereby being able to more accurately estimate the click-through rate, greatly improving the accuracy of the estimation results, and being suitable for large-scale promotion and use.

[0130] It should be noted that the above modules can be functional modules or program modules, and can be implemented by software or hardware. For modules implemented by hardware, the above modules can be located in the same processor; or the above modules can be located in different processors in any combination.

[0131] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0132] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering or, if necessary, processing in another suitable manner, and then stored in a computer memory.

[0133] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0134] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "examples", "specific examples", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0135] The above-mentioned embodiments only express several implementation methods of the present invention, and the description thereof is relatively specific and detailed, but it cannot be understood as limiting the scope of the patent of the present invention. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

Claims

1. A click rate prediction method, characterized in that: The method comprises: The user ID and the product ID are respectively converted into corresponding user embedding vectors and product embedding vectors that can be input into the neural network through the embedding model, and a corresponding user-product interaction graph is constructed based on the rating matrix in the user ID and the product ID. The user-product interaction graph includes a plurality of nodes, each of which represents a user or a product, and two nodes with an interactive relationship are connected by an edge. The neural network includes a graph neural network and a fully connected neural network; Perform multiple rounds of iterative information propagation in the user-product interaction graph through the graph neural network, and input the user embedding vector and the product embedding vector in its historical behavior into a preset interaction layer and a logic deduction layer to obtain a user interest vector; Constructing a click rate prediction layer based on the fully connected neural network, and inputting the user interest vector into the click rate prediction layer to estimate the click rate; The step of performing multiple rounds of iterative information propagation in the user-product interaction graph through the graph neural network includes: In the user-product interaction graph, information of an embedding vector of one of two connected nodes is transmitted to another node through an information propagation function to update a user embedding vector or a product embedding vector corresponding to each node; The expression of the amount of information propagated from one node to another node in two connected nodes is: Among them, e h and e t The embedding vector corresponding to the outgoing node and the receiving node of the information, D h and D t They represent the degree of a node, that is, the number of nodes connected to it, and W s , W i and b both represent trainable parameters; The expression for updating the user embedding vector or product embedding vector corresponding to each node is: Among them, σ represents the sigmoid activation function iterates L times of information propagation, and each node will get L vectors And L vectors are finally represented as vector e by this node. t ; The step of inputting the user embedding vector and the product embedding vector in the historical behavior thereof into a preset interaction layer and a logic deduction layer to obtain the user interest vector comprises: Inputting the product embedding vector in the user's historical behavior and the product embedding vector in the target advertisement into the interaction layer in a preset order, and calculating the corresponding interaction vector according to a second preset formula; The interaction vectors are sequentially input into the inference network in the logic inference layer for iterative calculation, and a non-operation is performed on the interaction vector corresponding to each behavior before inputting into the loop structure; Inputting the interaction vector after non-operation into the logic deduction layer, and calculating the corresponding user interest vector according to a third preset formula; The second preset formula is: Among them, [|] represents the concatenation operation, W1, W2, b1, b2 are all trainable parameters, e i represents the product embedding vector after the product ID is updated, e u represents the user embedding vector after the user ID is updated, σ2 represents the sigmoid activation function; The expression for iterative calculation of the derivation network is: The expression of the non-operation is: in, All are trainable parameters; The third preset formula is: Wherein, O represents the user interest vector, represents a trainable parameter, ReLU represents the ReLU activation function, and h n represents the hidden state obtained at time n in the logic derivation layer, It represents the interaction vector obtained after the product embedding vector and the user embedding vector input at the next moment n+1 are input into the interaction layer.

2. The click rate prediction method according to claim 1, characterized in that: The steps of converting the user ID and the product ID into corresponding user embedding vectors and product embedding vectors that can be input into the neural network through the embedding model, and constructing a corresponding user-product interaction graph based on the rating matrix in the user ID and the product ID include: The embedding table E is initialized for the user ID and the product ID respectively through the embedding model u and E i , and calculate the user embedding vector and the product embedding vector respectively according to a first preset formula, wherein the first preset formula is: Among them, e 0 u represents the embedding vector corresponding to the user ID, e 0 i represents the embedding vector corresponding to the product ID, t u represents the one-hot encoding of the user ID, t i A one-hot encoding representing the product ID; The user-product interaction graph G is constructed based on the rating matrix in the user ID and the product ID.<V,E> , where V represents the node set and E represents the edge set.

3. A click rate prediction system, characterized in that: The system comprises: A conversion module, used to convert the user ID and the product ID into corresponding user embedding vectors and product embedding vectors that can be input into the neural network through the embedding model, and construct a corresponding user-product interaction graph based on the rating matrix in the user ID and the product ID, wherein the user-product interaction graph includes a plurality of nodes, each of which represents a user or a product, and two nodes with an interactive relationship are connected by an edge, and the neural network includes a graph neural network and a fully connected neural network; An analysis module, configured to perform multiple rounds of iterative information propagation in the user-product interaction graph through the graph neural network, and input the user embedding vector and the product embedding vector in its historical behavior into a preset interaction layer and a logic deduction layer to obtain a user interest vector; An estimation module, configured to construct a click rate estimation layer based on the fully connected neural network, and input the user interest vector into the click rate estimation layer to estimate the click rate; The analysis module is specifically used for: In the user-product interaction graph, information of an embedding vector of one of two connected nodes is transmitted to another node through an information propagation function to update a user embedding vector or a product embedding vector corresponding to each node; The step of performing multiple rounds of iterative information propagation in the user-product interaction graph through the graph neural network includes: In the user-product interaction graph, information of an embedding vector of one of two connected nodes is transmitted to another node through an information propagation function to update a user embedding vector or a product embedding vector corresponding to each node; The expression of the amount of information propagated from one node to another node in two connected nodes is: Among them, e h and e t The embedding vector corresponding to the outgoing node and the receiving node of the information, D h and D t They represent the degree of a node, that is, the number of nodes connected to it, and W s , W i and b both represent trainable parameters; The expression for updating the user embedding vector or product embedding vector corresponding to each node is: Among them, σ represents the sigmoid activation function iterates L times of information propagation, and each node will get L vectors And L vectors are finally represented as vector e by this node. t ; The step of inputting the user embedding vector and the product embedding vector in the historical behavior thereof into a preset interaction layer and a logic deduction layer to obtain the user interest vector comprises: Inputting the product embedding vector in the user's historical behavior and the product embedding vector in the target advertisement into the interaction layer in a preset order, and calculating the corresponding interaction vector according to a second preset formula; The interaction vectors are sequentially input into the inference network in the logic inference layer for iterative calculation, and a non-operation is performed on the interaction vector corresponding to each behavior before inputting into the loop structure; Inputting the interaction vector after non-operation into the logic deduction layer, and calculating the corresponding user interest vector according to a third preset formula; The second preset formula is: Among them, [|] represents the concatenation operation, W1, W2, b1, b2 are all trainable parameters, e i represents the product embedding vector after the product ID is updated, e u represents the user embedding vector after the user ID is updated, σ2 represents the sigmoid activation function; The expression for iterative calculation of the derivation network is: The expression of the non-operation is: in, All are trainable parameters; The third preset formula is: Wherein, O represents the user interest vector, represents a trainable parameter, ReLU represents the ReLU activation function, and h n represents the hidden state obtained at time n in the logic derivation layer, It represents the interaction vector obtained after the product embedding vector and the user embedding vector input at the next moment n+1 are input into the interaction layer.

4. The click rate prediction system according to claim 3, characterized in that: The conversion module is specifically used for: The embedding table E is initialized for the user ID and the product ID respectively through the embedding model u and E i , and calculate the user embedding vector and the product embedding vector respectively according to a first preset formula, wherein the first preset formula is: Among them, e 0 u represents the embedding vector corresponding to the user ID, e 0 i represents the embedding vector corresponding to the product ID, t u represents the one-hot encoding of the user ID, t i A one-hot encoding representing the product ID; The user-product interaction graph G is constructed based on the rating matrix in the user ID and the product ID.<V,E> , where V represents the node set and E represents the edge set.

5. A computer comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the click rate prediction method according to any one of claims 1 to 2 is implemented.

6. A readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the click rate prediction method as described in any one of claims 1 to 2 is implemented.

Images