WO2023029350A1

WO2023029350A1 - Click behavior prediction-based information pushing method and apparatus

Info

Publication number: WO2023029350A1
Application number: PCT/CN2022/071436
Authority: WO
Inventors: 陈浩; 谯轶轩
Original assignee: 平安科技（深圳）有限公司
Priority date: 2021-08-31
Filing date: 2022-01-11
Publication date: 2023-03-09
Also published as: CN113705698A; CN113705698B

Abstract

The present application relates to the fields of artificial intelligence and digital medical treatment, and discloses a click behavior prediction-based information pushing method and apparatus, which aim to solve the existing problem that medical items are directly matched and pushed according to medical results of a patient without taking into consideration behavior information of the patient, which leads to the inability to meet medical information pushing needs of the patient, thereby reducing the effectiveness of receiving push information. The method comprises: acquiring user feature information and product feature information; on the basis of a completed click behavior prediction model, performing prediction processing on the user feature information and the product feature information to obtain a predicted click behavior result; if the result is an expected click behavior, extracting a target keyword in target product information that matches the product feature information, and determining the similarity between each product keyword in a product database and the target keyword; and acquiring associated product information corresponding to product keywords whose similarity is greater than a preset similarity threshold, and outputting the associated product information and the target product information in a linked manner.

Description

Information push method and device based on click behavior prediction

This application claims the priority of the Chinese patent application with the application number 202111015434.5 submitted to the China Patent Office on August 31, 2021, and the application title is "Information Push Method and Device Based on Click Behavior Prediction", the entire content of which is incorporated by reference in the application middle.

technical field

The present application relates to the field of artificial intelligence and digital medical technology, in particular to an information push method and device based on click behavior prediction.

Background technique

With the rapid development of artificial intelligence, artificial intelligence has been fully popularized in the big data field of digital medical business. Among them, after medical consultation based on digital medical treatment, information can be pushed through the intelligent medical system.

The inventor realized that the push of medical information is usually based on the medical results of different patients directly matching related medical items for push. However, the information push based on the direct matching method does not take into account the patient's behavior information, and cannot really satisfy the patient's medical needs. Information push requirements, thereby reducing the effectiveness of receiving information push, making the pushed information invalid.

Contents of the invention

In view of this, the present application provides an information push method and device based on click behavior prediction.

According to one aspect of the present application, a method for pushing information based on click behavior prediction is provided, including:

Obtain user characteristic information and product characteristic information;

Based on the completed click behavior prediction model, the user characteristic information and the product characteristic information are predicted and processed to obtain the predicted click behavior result. The click behavior prediction model is obtained by constructing network weights based on the redirection decomposition machine algorithm and completing model training. of;

If the result of the predicted click behavior is the expected click behavior, then extract the target keyword in the target product information matching the product feature information, and determine the relationship between each product keyword in the product database and the target keyword similarity;

Obtain associated product information corresponding to product keywords whose similarity is greater than a preset similarity threshold, and output the associated product information and the target product information in a linked manner.

According to another aspect of the present application, an information push device based on click behavior prediction is provided, including:

An acquisition module, configured to acquire user feature information and product feature information;

The prediction module is used to perform prediction processing on the user characteristic information and the product characteristic information based on the completed click behavior prediction model, and obtain the predicted click behavior result. The weight is obtained by completing the model training;

A determining module, configured to extract the target keywords in the target product information matching the product feature information if the predicted click behavior result is the expected click behavior, and determine the relationship between each product keyword in the product database and the target keyword similarity between words;

An output module, configured to obtain associated product information corresponding to product keywords whose similarity degree is greater than a preset similarity threshold, and output the associated product information and the target product information in a linked manner.

According to yet another aspect of the present application, a computer-readable storage medium is provided, on which computer-readable instructions are stored, and when the computer-readable instructions are executed by a processor, an information push method based on click behavior prediction is implemented, including:

Obtain user characteristic information and product characteristic information;

According to still another aspect of the present application, a computer device is provided, including a memory, a processor, and computer-readable instructions stored in the memory and operable on the processor, wherein the computer-readable instructions are executed by the processor Realize the information push method based on click behavior prediction, including:

Obtain user characteristic information and product characteristic information;

With the above technical solution, the technical solution provided by the embodiment of the present application has at least the following advantages:

This application uses the click behavior prediction model to predict the behavior of users clicking on products, and pushes the target product and related products based on the prediction results, which meets the user's product push needs and ensures that the pushed information is valid information, thereby improving information push. reception validity.

Description of drawings

Various other advantages and benefits will become apparent to those of ordinary skill in the art upon reading the following detailed description of the preferred embodiment. The drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the application. Also throughout the drawings, the same reference numerals are used to designate the same components. In the attached picture:

FIG. 1 shows a flowchart of an information push method based on click behavior prediction provided by an embodiment of the present application;

FIG. 2 shows a flow chart of another information push method based on click behavior prediction provided by the embodiment of the present application;

FIG. 3 shows a block diagram of an information push device based on click behavior prediction provided by an embodiment of the present application;

Fig. 4 shows a schematic structural diagram of a computer device provided by an embodiment of the present application.

Detailed ways

Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be embodied in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided for more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

The embodiments of the present application may acquire and process relevant data based on artificial intelligence technology. Among them, artificial intelligence (AI) is a theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results. .

Artificial intelligence basic technologies generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation/interaction systems, and mechatronics. Artificial intelligence software technology mainly includes computer vision technology, robotics technology, biometrics technology, speech processing technology, natural language processing technology, and machine learning/deep learning.

Based on this, in one embodiment, as shown in Figure 1, a method for pushing information based on click behavior prediction is provided, and the application of this method to computer equipment such as servers is used as an example for illustration, wherein the server may be an independent server , can also provide cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery network (Content Delivery Network, CDN), and big data Cloud servers for basic cloud computing services such as artificial intelligence platforms, such as intelligent medical systems, digital medical platforms, etc. The above method comprises the following steps:

101. Acquire user feature information and product feature information.

The embodiments of the present application can be applied to any electronic platform with information push function, such as an insurance electronic trading platform, a smart medical platform, and the like. Taking the insurance electronic trading platform as an example, when the user chooses to log in and register, the user's characteristic information is recorded and saved in the current platform. The user's characteristic information includes but is not limited to the user's age, gender, occupation, salary level, user click behavior, etc. So that the current system can recommend information. In order to recommend applicable push information and insurance products to users, as the insurance electronic trading platform at the current execution end, obtain product feature information, product feature information includes but not limited to insurance amount, claim settlement time limit, payment method, claim settlement amount, etc., so that based on user characteristics information to identify whether a click will be made on this insurance product. It can be understood that the characteristic information of the insurance product is a discrete characteristic. User feature information is a group of feature vectors corresponding to one user, and insurance product feature information may be feature vectors corresponding to multiple products, which is not specifically limited in this embodiment of the present application. For example, user A may click on multiple products such as product B, product C, product D, and so on.

It should be noted that in the embodiment of the present application, user feature information can be entered in a way that is obtained when the user logs in and registers, and product feature information can not only be entered based on development technicians, in order to achieve artificial intelligence and make up for product updates To solve the problem of untimely feature update, feature extraction can be performed on specific marks in the product to ensure the accuracy of product feature information.

102. Based on the completed click behavior prediction model, perform prediction processing on user characteristic information and product characteristic information to obtain a predicted click behavior result.

Among them, the click behavior prediction model is obtained by constructing network weights based on the redirection decomposition machine algorithm and completing model training.

In the embodiment of the present application, based on user feature information and product feature information, a click behavior prediction model is used to predict whether the user will click on the product. Since the prediction results based on the characteristic information of a user and the characteristic information of multiple products include whether the user will expect to click on each product, the result of the predicted click behavior is a vector containing 0 and 1 to determine Whether the user is expected to click on each insurance product.

It should be noted that, in the process of predicting the click behavior prediction model, the discrete features of user U and product I can first be encoded by one-hot to obtain the corresponding vector representation X, and input X to the redirection decomposition machine algorithm In the end, if

It means that user U will click item I, and vice versa.

103. If the result of the predicted click behavior is the expected click behavior, extract the target keyword in the target product information that matches the product feature information, and determine the similarity between each product keyword and the target keyword in the product database.

In the embodiment of the present application, if a certain product in the predicted click behavior result is 1, it means that the predicted user will click on the product. Therefore, when it is determined to be the expected click behavior, the keywords in the target product are extracted, and the key words in the product database are determined. The keyword of each product, the similarity with the keyword of the target product, in order to find the associated product information. Among them, the keywords include but are not limited to product types such as commercial insurance, medical insurance, and vehicle insurance, or include but are not limited to specific product content, such as serious illness, accident, personal life, and property. Therefore, lookups can be based on the product database. In the embodiment of the present application, different product information and corresponding keywords are pre-stored in the product database, so as to calculate the similarity.

It should be noted that, in the embodiment of the present application, when calculating the similarity of keywords, in addition to calculating the similarity of text meaning, the similarity between the number of keywords can also be calculated, that is, the number of keywords carried by the target product The similarity value between the number of keywords with other product information, select the product information whose similarity value is greater than the preset similarity threshold as the associated product information.

104. Obtain associated product information corresponding to product keywords whose similarity is greater than a preset similarity threshold, and output associated product information and target product information in a linkage manner.

In the embodiment of this application, since the result of the predicted click behavior is the expected click behavior, the corresponding product information is the target product that the user expects to click, and the associated product information is determined based on similarity, and the number is much larger than the target product that the user expects to click Quantity, therefore, when outputting, in order to avoid users from having a redundant experience due to viewing a large number of recommended products, before outputting, sort according to similarity, and then sort the sorted products and related products according to the scrolling rendering method output in the form of information.

It should be noted that the scrolling rendering method can be to configure a display box, which is divided into two parts, one part is used to push the target product information corresponding to the predicted click behavior result as the expected click behavior, and the other part is scrolled according to the similar Display the relevant product information in order of degree to improve recommendation efficiency.

For further explanation and limitation, the embodiment of the present application provides another information push method based on click behavior prediction, as shown in Figure 2, the method includes:

201. Obtain a product click behavior training sample set, and construct a basic neural network.

202. Determine the neural network multilayer perceptron of the basic neural network, and establish a connection relationship between the neural network multilayer perceptron and the fully connected layer network structure, deactivation network, and activation network.

203. Multiply the vector obtained after constructing the connection relationship with the embedded vector layer parameters of the preset decomposition machine algorithm to obtain the redirection decomposition machine algorithm of the weight vector.

204. Perform the weight vector inner product operation based on the redirected factorization machine algorithm to generate a limited and reset prediction model.

205. Define a loss function according to the sample data in the product click behavior training sample set, wherein the loss function is defined based on the redirection decomposition machine algorithm and obtained by combining the cross-entropy loss function and the activation function.

206. Perform model training on the prediction model based on the product click behavior training sample set to obtain a click behavior prediction model.

Specifically, the product click training sample set is pre-built, and any training sample in the data set can be expressed as (U, I, Y), where the discrete features of the user can be expressed as

Among them, u _j represents the jth feature of the user, such as the user's age, gender and other discrete features, and p represents the number of features of each user. The discrete features of any product in the data set can be expressed as

Among them, i _j represents the qth feature of the product, such as discrete features such as shape and category, and q represents the number of features of each product. Y indicates whether the user U clicked on the product I, where Y∈{0,1}, 0 indicates that the user did not click on the product, and 1 indicates that the user clicked on the product.

In the constructed neural network, the input vector represents: perform one-hot encoding on the user characteristic information U and product characteristic information I in the training samples (U, I, y), and obtain the input vector X=[x ₁ ,x ₂ of the model ,..., _xi ,...,x _n ], where _xi represents the i-th dimension of the vector X, and _xi ∈ {0,1}, n represents the input vector after one-hot encoding dimension. Then the training samples of the data set G can be expressed as (X, Y) after one-hot encoding.

It should be noted that in the limited reset of the weight vector in the neural network combined with the fully connected layer network, inactivation network, and activation layer network, it is realized based on the redirection decomposition machine algorithm. Among them, the traditional factorization machine The algorithm (FM) can be expressed as:

Among them, w ₀ ∈ R is the bias parameter of the factorization machine algorithm (FM), w _i ∈ R is the characteristic weight parameter of the factorization machine algorithm (FM), v _i , v _j are the factorization machine algorithm (FM) Embedding vector layer parameters, for any dimension x _i of vector X, there is a corresponding embedding vector v _i corresponding to it, and v _i , v _j ∈ R ^k , k is the dimension of embedding vector, <v _i , v _j ＞ Indicates the inner product operation of vectors v _i and v _j , and represents the feature cross operation of the factorization machine algorithm (FM). In the embodiment of the present application, a redirection decomposition machine algorithm is provided, and the specific formula is as follows:

M _X =mlps(X)

Among them, mlps represents the neural network multi-layer perceptron, M _X represents the vector obtained by inputting the input vector X into the fully connected layer network, and passing through the corresponding deactivation network and activation layer network, and M _X ∈ R ^k , that is, the multi-layer The number of neurons in the last layer of the perceptron network is k, and v _i and v _j are the same as the above steps, which are the parameters of the embedding vector layer of the redirection decomposition machine algorithm.

Multiplication operation for the corresponding elements of the vector.

It can be seen from the above formula that the embodiment of the present application uses a fully connected layer network. Even if a similar input vector X passes through a multi-layer perceptron network, the corresponding M _X is different. In the subsequent feature crossover process, although the same feature Shared embedding layer vectors v _i , v _j , but v _iX , v _jX obtained after M _X weighting are also different. Therefore, the difference caused by similar vectors can be effectively avoided.

In addition, during the model training process, for any sample (X, Y) in the data set G, the model loss can be defined as:

Among them, sigmoid represents the sigmoid activation function, and CE represents the cross-entropy loss function. Input the data in the data set G into the formula of the redirection decomposition machine algorithm, and use the gradient descent algorithm to continuously optimize the parameters of the model until the model loss converges to complete the model training.

In a specific application scenario, for example, a data set G is constructed, and a total of 100,000 historical click records of users are collected. Among them, 12 discrete features of users are selected, and 7 of them are discrete features, which are encoded by one-hot , the dimension of the final input vector X is 72, the number of layers of the multilayer perceptron network is 4, and the number of neurons in each layer is 64, 128, 64, and 32 respectively. The embedding layer vector dimension k is 32 in the redirection decomposition machine algorithm.

In order to ensure that the pushed information is valid information, in the embodiment of this application, based on the completed click behavior prediction model, the user characteristic information and product characteristic information are predicted and processed, and after the predicted click behavior result is obtained, the method of this embodiment also includes: if If the result of the predicted click behavior is the expected non-click behavior, the replacement product information is searched from the product matching relationship database based on the user characteristic information and pushed.

Wherein, the product matching relationship database stores correspondences between different user characteristic information and different product information. When the result of the predicted click behavior is the expected non-click behavior, it means that the user will not click on the target product with a high probability. In order to ensure the validity of the push information, it is necessary to search and push the replacement product information from the product matching relationship database according to the user characteristic information. For example, if the user is a 55-year-old woman, she can search for products related to serious gynecological diseases and push them as replacement products.

In the embodiment of the present application, it is further preferred that outputting related product information and target product information in a linked manner includes: rendering the target product information into the first display frame, and sorting at least one related product information in order of similarity , render the sorted associated product information into the second display frame, the first display frame and the second display frame are combined into a floating display window; output the associated product information in the second display frame according to the scrolling rendering method, and Output target product information in a display box.

In order to display the output content clearly to the user, the output content may be divided into two parts, which are respectively displayed in another display box, and displayed to the user in the form of a floating display window. Wherein, the first display frame is used to display the target product expected to be clicked by the user, and the second display frame is used to display related product information associated with the target product information. In order to prevent users from being bored by viewing a large number of recommended products, thereby reducing the sense of experience, when displaying related product information in the second display box, it can be sorted according to the degree of similarity, and the sorted related product information can be displayed.

It should be noted that, for the convenience of viewing, the associated product information may be output in the second display frame in a scrolling rendering manner.

In order to ensure that the user can receive information that meets the user's product push needs, in the embodiment of the present application, after determining the similarity between each product keyword in the product database and the target keyword, the method in this embodiment further includes: if the similarity If it is less than or equal to the preset similarity threshold, search for replacement product information from the product matching relationship database based on business requirements, and push it.

Specifically, when the similarity is less than or equal to the preset similarity threshold, it means that no associated product similar to the target product has been found, and at this time, the associated product can be replaced based on business requirements. For example, push products that have a relatively strong discount in the near future, or products that are the main push, as associated products to users.

This application provides an information push method based on click behavior prediction. Firstly, user characteristic information and product characteristic information are obtained; based on the completed click behavior prediction model, the user characteristic information and product characteristic information are predicted and processed to obtain Predict the click behavior result, the click behavior prediction model is based on the redirection decomposition machine algorithm to construct the network weight to complete the model training; if the predicted click behavior result is the expected click behavior, then extract the target that matches the product feature information Target keywords in the product information, and determine the similarity between each product keyword in the product database and the target keyword; obtain the associated product information corresponding to the product keyword whose similarity is greater than the preset similarity threshold , and output the associated product information and the target product information in a linked manner. Compared with the prior art, the embodiment of the present application uses the click behavior prediction model to predict the behavior of users clicking on products, and pushes the target product and related products based on the prediction results, which meets the user's product push needs and ensures that the pushed information For effective information, thereby improving the effectiveness of receiving information push.

Further, as an implementation of the method shown in Figure 1 above, an embodiment of the present application provides an information push device based on click behavior prediction, as shown in Figure 3, the device includes:

An acquisition module 31 , a prediction module 32 , a determination module 33 , and an output module 34 .

Obtaining module 31, for acquiring user feature information and product feature information;

Prediction module 32, for predicting the user feature information and the product feature information based on the completed click behavior prediction model to obtain the predicted click behavior result. The click behavior prediction model is constructed based on the redirection decomposition machine algorithm The network weight is obtained by completing the model training;

A determining module 33, configured to extract the target keywords in the target product information matching the product feature information if the predicted click behavior result is the expected click behavior, and determine the relationship between each product keyword in the product database and the target keyword. The similarity between keywords;

The output module 34 is configured to obtain associated product information corresponding to product keywords whose similarity is greater than a preset similarity threshold, and output the associated product information and the target product information in a linked manner.

In a specific application scenario, before the prediction module 32, the device also includes:

The building block is used to obtain the product click behavior training sample set and build the basic neural network;

The limiting module is used to limit and reset the weight vectors corresponding to the fully connected layer network, the deactivation network and the activation network in the basic neural network based on the redirection decomposition machine algorithm, so as to obtain a prediction model that completes the limited reset;

A training module, configured to perform model training on the prediction model based on the product click behavior training sample set to obtain a click behavior prediction model.

In a specific application scenario, the limiting module includes:

A determining unit, configured to determine the neural network multilayer perceptron of the basic neural network, and establish a connection relationship between the neural network multilayer perceptron and the fully connected layer network structure, deactivation network, and activation network;

The multiplication unit is used to multiply the vector obtained after building the connection relationship with the embedded vector layer parameters of the preset decomposition machine algorithm to obtain the redirection decomposition machine algorithm of the weight vector;

A generating unit, configured to perform the weight vector inner product operation based on the redirection factorization machine algorithm, and generate a defined and reset prediction model.

In a specific application scenario, before the training module, the device also includes:

The definition module is used to define a loss function according to the sample data in the product click behavior training sample set, wherein the loss function is defined based on the redirection decomposition machine algorithm and obtained by combining a cross-entropy loss function and an activation function.

In a specific application scenario, after the prediction module 32, the device further includes:

The replacement module is used to search for replacement product information from the product matching relationship database based on the user characteristic information if the result of the predicted click behavior is an expected non-click behavior, and push it. The product matching relationship database stores different Correspondence between user feature information and different product information.

In a specific application scenario, the output module 34 includes:

A rendering unit, configured to render the target product information into the first display frame, sort at least one related product information in order of similarity, and render the sorted related product information into the second display frame wherein, the first display frame and the second display frame are combined into a floating display window;

An output unit, configured to output the associated product information in the second display frame in a scrolling rendering manner, and output the target product information in the first display frame.

In a specific application scenario, after the determining module 33, the device further includes:

The push module is configured to search for replacement product information from the product matching relationship database based on business requirements and push the information if the similarity is less than or equal to the preset similarity threshold.

The present application provides an information push device based on click behavior prediction. First, user characteristic information and product characteristic information are obtained; based on the completed click behavior prediction model, the user characteristic information and product characteristic information are predicted and processed to obtain Predict the click behavior result, the click behavior prediction model is based on the redirection decomposition machine algorithm to construct the network weight to complete the model training; if the predicted click behavior result is the expected click behavior, then extract the target that matches the product feature information Target keywords in the product information, and determine the similarity between each product keyword in the product database and the target keyword; obtain the associated product information corresponding to the product keyword whose similarity is greater than the preset similarity threshold , and output the associated product information and the target product information in a linked manner. Compared with the prior art, the embodiment of the present application uses the click behavior prediction model to predict the behavior of users clicking on products, and pushes the target product and related products based on the prediction results, which meets the user's product push needs and ensures that the pushed information For effective information, thereby improving the effectiveness of receiving information push.

According to one embodiment of the present application, a computer-readable storage medium is provided, the storage medium stores at least one executable instruction, and the computer-executable instruction can execute the information push method based on click behavior prediction in any of the above method embodiments , the computer-readable storage medium may be non-volatile or volatile.

Based on this understanding, the technical solution of the present application can be embodied in the form of software products, which can be stored in a non-volatile storage medium (which can be CD-ROM, U disk, mobile hard disk, etc.), including several The instructions are used to make a computer device (which may be a personal computer, a server, or a network device, etc.) execute the methods described in various implementation scenarios of the present application.

FIG. 4 shows a schematic structural diagram of a computer device provided according to an embodiment of the present application. The specific embodiment of the present application does not limit the specific implementation of the computer device.

As shown in FIG. 4 , the computer device may include: a processor (processor) 402, a communication interface (Communications Interface) 404, a memory (memory) 406, and a communication bus 408.

Wherein: the processor 402 , the communication interface 404 , and the memory 406 communicate with each other through the communication bus 408 .

The communication interface 404 is used to communicate with network elements of other devices such as clients or other servers.

The processor 402 is configured to execute the program 410, specifically, may execute relevant steps in the above embodiment of the method for pushing information based on click behavior prediction.

Specifically, the program 410 may include program codes including computer operation instructions.

The processor 402 may be a central processing unit CPU, or an ASIC (Application Specific Integrated Circuit), or one or more integrated circuits configured to implement the embodiments of the present application. The one or more processors included in the computer device may be of the same type, such as one or more CPUs, or may be of different types, such as one or more CPUs and one or more ASICs.

The memory 406 is used to store the program 410 . The memory 406 may include a high-speed RAM memory, and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory.

The program 410 can specifically be used to make the processor 402 perform the following operations:

Obtain user characteristic information and product characteristic information;

The storage medium may also include an operating system and a network communication module. The operating system is a program that manages the hardware and software resources of the physical equipment for business data processing based on the multi-modal hybrid model, and supports the operation of information processing programs and other software and/or programs. The network communication module is used to realize the communication between various components inside the storage medium, and communicate with other hardware and software in the information processing entity device.

Obviously, those skilled in the art should understand that each module or each step of the above-mentioned application can be realized by a general-purpose computing device, and they can be concentrated on a single computing device, or distributed in a network composed of multiple computing devices Alternatively, they may be implemented in program code executable by a computing device so that they may be stored in a storage device to be executed by a computing device, and in some cases in an order different from that shown here The steps shown or described are carried out, or they are separately fabricated into individual integrated circuit modules, or multiple modules or steps among them are fabricated into a single integrated circuit module for implementation. As such, the present application is not limited to any specific combination of hardware and software.

The above descriptions are only preferred embodiments of the present application, and are not intended to limit the present application. For those skilled in the art, there may be various modifications and changes in the present application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of this application shall be included within the protection scope of this application.

Claims

An information push method based on click behavior prediction, including:

Obtain user characteristic information and product characteristic information;

Based on the completed click behavior prediction model, the user characteristic information and the product characteristic information are predicted and processed to obtain the predicted click behavior result. The click behavior prediction model is obtained by constructing network weights based on the redirection decomposition machine algorithm and completing model training. of;

If the result of the predicted click behavior is the expected click behavior, then extract the target keyword in the target product information matching the product feature information, and determine the relationship between each product keyword in the product database and the target keyword similarity;

Obtain associated product information corresponding to product keywords whose similarity is greater than a preset similarity threshold, and output the associated product information and the target product information in a linked manner.
The method according to claim 1, wherein, performing prediction processing on the user feature information and the product feature information based on the completed click behavior prediction model, and before obtaining the predicted click behavior result, the method further includes:

Obtain a product click behavior training sample set and build a basic neural network;

Based on the redirection decomposition machine algorithm, the weight vectors corresponding to the fully connected layer network, the deactivation network, and the activation network in the basic neural network are limited and reset, and a prediction model that completes the limited reset is obtained;

Model training is performed on the prediction model based on the product click behavior training sample set to obtain a click behavior prediction model.
The method according to claim 2, wherein, the weight vectors corresponding to the fully connected layer network, the inactivation network, and the activation network in the basic neural network are limited and reset based on the redirection decomposition machine algorithm, to obtain completion Predictive models that qualify reset include:

Determining the neural network multilayer perceptron of the basic neural network, and establishing a connection relationship between the neural network multilayer perceptron and the fully connected layer network structure, deactivation network, and activation network;

Multiply the vector obtained after constructing the connection relationship with the embedded vector layer parameters of the preset decomposition machine algorithm to obtain the redirection decomposition machine algorithm of the weight vector;

The weight vector inner product operation is performed based on the reorientation factorization machine algorithm to generate a limited and reset prediction model.
The method according to claim 3, wherein, performing model training on the prediction model based on the product click behavior training sample set, and before obtaining the click behavior prediction model, the method further comprises:

A loss function is defined according to the sample data in the product click behavior training sample set, wherein the loss function is defined based on the redirection decomposition machine algorithm and obtained by combining a cross-entropy loss function and an activation function.
The method according to claim 1, wherein the user feature information and the product feature information are predicted based on the completed click behavior prediction model, and after obtaining the predicted click behavior result, the method further includes:

If the result of the predicted click behavior is an expected non-click behavior, search for replacement product information from the product matching relationship database based on the user characteristic information, and push it. The product matching relationship database stores different user characteristic information and different Correspondence between product information.
The method according to any one of claims 1-5, wherein said outputting said associated product information and said target product information in a linked manner comprises:

Rendering the target product information into the first display frame, sorting at least one related product information in order of similarity, rendering the sorted related product information into the second display frame, the first A display frame and the second display frame are combined into a floating display window;

Outputting the associated product information in the second display frame in a scrolling rendering manner, and outputting the target product information in the first display frame.
The method according to claim 6, wherein, after determining the similarity between each product keyword in the product database and the target keyword, the method further comprises:

If the similarity is less than or equal to the preset similarity threshold, search for replacement product information from the product matching relationship database based on business requirements, and push it.
An information push device based on click behavior prediction, including:

An acquisition module, configured to acquire user feature information and product feature information;

The prediction module is used to perform prediction processing on the user characteristic information and the product characteristic information based on the completed click behavior prediction model, and obtain the predicted click behavior result. The weight is obtained by completing the model training;

A determining module, configured to extract the target keywords in the target product information matching the product feature information if the predicted click behavior result is the expected click behavior, and determine the relationship between each product keyword in the product database and the target keyword similarity between words;

An output module, configured to obtain associated product information corresponding to product keywords whose similarity degree is greater than a preset similarity threshold, and output the associated product information and the target product information in a linked manner.
A computer-readable storage medium, on which computer-readable instructions are stored, wherein, when the computer-readable instructions are executed by a processor, an information push method based on click behavior prediction is implemented, including:

Obtain user characteristic information and product characteristic information;

Based on the completed click behavior prediction model, the user characteristic information and the product characteristic information are predicted and processed to obtain the predicted click behavior result. The click behavior prediction model is obtained by constructing network weights based on the redirection decomposition machine algorithm and completing model training. of;

If the result of the predicted click behavior is the expected click behavior, then extract the target keyword in the target product information matching the product feature information, and determine the relationship between each product keyword in the product database and the target keyword similarity;

Obtain the associated product information corresponding to the product keywords whose similarity is greater than the preset similarity threshold, and output the associated product information and the target product information in a linkage manner.
The computer-readable storage medium according to claim 9, wherein, when the computer-readable instructions are executed by the processor, the completed click behavior prediction model is used to perform the user characteristic information and the product characteristic information. Prediction processing, before obtaining the predicted click behavior result, the method also includes:

Obtain a product click behavior training sample set and build a basic neural network;

Based on the redirection decomposition machine algorithm, the weight vectors corresponding to the fully connected layer network, the deactivation network, and the activation network in the basic neural network are limited and reset, and a prediction model that completes the limited reset is obtained;

Model training is performed on the prediction model based on the product click behavior training sample set to obtain a click behavior prediction model.
The computer-readable storage medium according to claim 10, wherein, when the computer-readable instructions are executed by the processor, the algorithm based on the redirection decomposition machine is implemented for the fully connected layer network and the deactivation network in the basic neural network. , The weight vector corresponding to the activation network is limited to reset, and the prediction model that completes the limited reset is obtained, including:

Determining the neural network multilayer perceptron of the basic neural network, and establishing a connection relationship between the neural network multilayer perceptron and the fully connected layer network structure, deactivation network, and activation network;

Multiply the vector obtained after constructing the connection relationship with the embedded vector layer parameters of the preset decomposition machine algorithm to obtain the redirection decomposition machine algorithm of the weight vector;

The weight vector inner product operation is performed based on the reorientation factorization machine algorithm to generate a limited and reset prediction model.
The computer-readable storage medium according to claim 11, wherein when the computer-readable instructions are executed by the processor, the model training of the prediction model based on the product click behavior training sample set is implemented to obtain the click behavior Before predicting the model, the method also includes:

A loss function is defined according to the sample data in the product click behavior training sample set, wherein the loss function is defined based on the redirection decomposition machine algorithm and obtained by combining a cross-entropy loss function and an activation function.
The computer-readable storage medium according to claim 9, wherein, when the computer-readable instructions are executed by the processor, the completed click behavior prediction model is used to perform the user characteristic information and the product characteristic information. Prediction processing, after obtaining the predicted click behavior result, the method also includes:

If the result of the predicted click behavior is an expected non-click behavior, search for replacement product information from the product matching relationship database based on the user characteristic information, and push it. The product matching relationship database stores different user characteristic information and different Correspondence between product information.
The computer-readable storage medium according to any one of claims 9-13, wherein, when the computer-readable instructions are executed by the processor, the output of the associated product information and the target product information includes :

Rendering the target product information into the first display frame, sorting at least one related product information in order of similarity, rendering the sorted related product information into the second display frame, the first A display frame and the second display frame are combined into a floating display window;

Outputting the associated product information in the second display frame in a scrolling rendering manner, and outputting the target product information in the first display frame.
A computer device, comprising a memory, a processor, and computer-readable instructions stored on the memory and operable on the processor, wherein, when the computer-readable instructions are executed by the processor, an information push method based on click behavior prediction is realized ,include:

Obtain user characteristic information and product characteristic information;

Based on the completed click behavior prediction model, the user characteristic information and the product characteristic information are predicted and processed to obtain the predicted click behavior result. The click behavior prediction model is obtained by constructing network weights based on the redirection decomposition machine algorithm and completing model training. of;

If the result of the predicted click behavior is the expected click behavior, then extract the target keyword in the target product information matching the product feature information, and determine the relationship between each product keyword in the product database and the target keyword similarity;

Obtain associated product information corresponding to product keywords whose similarity is greater than a preset similarity threshold, and output the associated product information and the target product information in a linked manner.
The computer device according to claim 15, wherein when the computer-readable instructions are executed by the processor, the completed click behavior prediction model is used to predict the user characteristic information and the product characteristic information, Before obtaining the predicted click behavior result, the method further includes:

Obtain a product click behavior training sample set and build a basic neural network;

Based on the redirection decomposition machine algorithm, the weight vectors corresponding to the fully connected layer network, the deactivation network, and the activation network in the basic neural network are limited and reset, and a prediction model that completes the limited reset is obtained;

Model training is performed on the prediction model based on the product click behavior training sample set to obtain a click behavior prediction model.
The computer device according to claim 16, wherein, when the computer-readable instructions are executed by the processor, the algorithm based on the redirection decomposition machine is implemented for the fully connected layer network, the deactivation network, and the activation network in the basic neural network. The corresponding weight vector is limited to reset, and the prediction model that completes the limited reset is obtained including:

Determining the neural network multilayer perceptron of the basic neural network, and establishing a connection relationship between the neural network multilayer perceptron and the fully connected layer network structure, deactivation network, and activation network;

Multiply the vector obtained after constructing the connection relationship with the embedded vector layer parameters of the preset decomposition machine algorithm to obtain the redirection decomposition machine algorithm of the weight vector;

The weight vector inner product operation is performed based on the reorientation factorization machine algorithm to generate a limited and reset prediction model.
The computer device according to claim 17, wherein when the computer-readable instructions are executed by the processor, the model training of the prediction model based on the product click behavior training sample set is implemented, and the click behavior prediction model is obtained before , the method also includes:

A loss function is defined according to the sample data in the product click behavior training sample set, wherein the loss function is defined based on the redirection decomposition machine algorithm and obtained by combining a cross-entropy loss function and an activation function.
The computer device according to claim 15, wherein when the computer-readable instructions are executed by the processor, the completed click behavior prediction model is used to predict the user characteristic information and the product characteristic information, After obtaining the predicted click behavior result, the method also includes:

If the result of the predicted click behavior is an expected non-click behavior, search for replacement product information from the product matching relationship database based on the user characteristic information, and push it. The product matching relationship database stores different user characteristic information and different Correspondence between product information.
The computer device according to any one of claims 15-19, wherein when the computer-readable instructions are executed by the processor, the output of the associated product information and the target product information in a linkage manner includes:

Rendering the target product information into the first display frame, sorting at least one related product information in order of similarity, rendering the sorted related product information into the second display frame, the first A display frame and the second display frame are combined into a floating display window;

Outputting the associated product information in the second display frame in a scrolling rendering manner, and outputting the target product information in the first display frame.