Advertisement conversion parameter prediction model training method, prediction method, device and equipment
By obtaining ad sample data for feature vector conversion and importance weight calculation, and using parameter prediction model to adjust ad conversion parameters, the problem of advertising placement prediction bias is solved and the advertising conversion rate and display effect is improved.
Patent Information
- Application Number
- CN202310640890.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2023-05-31
- Publication Date
- 2024-12-03
AI Technical Summary
In the prior art, the estimated deviation of the advertising placement affects the advertising conversion parameters, resulting in poor conversion rates and display effects.
By obtaining advertising sample data, performing feature vector conversion and importance weight calculation, using parameter prediction models for classification prediction, calculating model losses and adjusting model parameters to reduce placement prediction bias.
It effectively improves the accuracy and conversion rate of advertising conversion parameter prediction, eliminates the subjective impact of placement on user behavior, and improves advertising display effect.
Smart Images

Figure CN119067725A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of machine learning technology, and more specifically, to an advertising conversion parameter prediction model training method, prediction method, device and equipment. Background Art
[0002] In the related art, when performing data search and display, the relevant data is usually searched based on the user's data search terms, and refreshed and displayed according to the searched relevant data. In addition, when displaying the searched relevant data, the data is usually arranged and displayed according to the relevance between the data and the search terms.
[0003] However, when users view the displayed data, they usually view the relevant data according to their own habits. Therefore, the above display method will affect the conversion efficiency of the data when users view the data. For example, when the above data is advertising data, users usually view the advertising data according to their habits, such as viewing the relevant advertising data in a top-down order according to the display position. This method usually affects the conversion rate of the advertising data and the display effect is poor. Summary of the invention
[0004] In view of this, the embodiments of the present application propose an advertising conversion parameter prediction model training method, prediction method, device and equipment, which can effectively reduce the impact of the estimated deviation of the advertising display position on the advertising conversion parameters.
[0005] In a first aspect, an embodiment of the present application provides a prediction model training method, the method comprising: obtaining advertising sample data, the advertising sample data carrying an advertising conversion label corresponding to a target display position of the advertising sample data; performing vector conversion on user-side feature data and advertising-side feature data in the advertising sample data to obtain a plurality of feature vectors included in the advertising sample data; using the weight prediction model to obtain the importance weight of each feature vector; using the parameter prediction model to perform classification prediction based on the plurality of feature vectors included in the advertising sample data and the importance weight of each feature vector to obtain advertising conversion parameters corresponding to each display position of the advertising sample data; calculating model loss based on the advertising conversion label corresponding to the display position of the advertising sample data and the advertising conversion parameters corresponding to each display position of the advertising sample data; and adjusting the model parameters in the advertising conversion parameter prediction model used to predict that the advertising sample data corresponds to the target display position according to the model loss until the training end condition is met.
[0006] In a second aspect, an embodiment of the present application provides an advertising conversion parameter prediction method, the method comprising: obtaining user-side characteristics of a target user and advertising-side characteristics of a plurality of advertising data to be published; using an advertising conversion parameter prediction model to obtain advertising conversion parameters corresponding to each display position of the advertising data to be published based on the user-side characteristics and the advertising-side characteristics of each of the advertising data to be published; based on the advertising conversion parameters corresponding to each display position of each advertising data to be published, determining target advertising data and the display position corresponding to the target advertising data from the plurality of advertising data to be published.
[0007] In the third aspect, the embodiment of the present application provides an advertising conversion parameter prediction model training device, the device includes: a data acquisition module, a data conversion module, a weight prediction module, a parameter prediction module, a loss acquisition module and a model training module. The data acquisition module is used to acquire advertising sample data, the advertising sample data carries the advertising conversion label corresponding to the target display position of the advertising sample data; the data conversion module is used to perform vector conversion on the advertising sample data to obtain multiple feature vectors included in the advertising sample data; the weight prediction module is used to obtain the importance weight of each feature vector using the weight prediction model; the parameter prediction module is used to perform classification prediction based on the multiple feature vectors included in the advertising sample data and the importance weight of each feature vector using the parameter prediction model to obtain the advertising conversion parameters corresponding to each display position of the advertising sample data; the loss acquisition module is used to calculate the model loss based on the advertising conversion label corresponding to the display position of the advertising sample data and the advertising conversion parameters corresponding to each display position of the advertising sample data; the model training module is used to adjust the model parameters used to predict the advertising sample data corresponding to the target display position in the advertising conversion parameter prediction model according to the model loss until the training end condition is reached.
[0008] In one possible implementation, the model training module includes a masking submodule and a training submodule. The masking submodule is used to mask the target model parameters in the advertising conversion parameter prediction model, wherein the target model parameters are other model parameters except the model parameters used in the classification layer of the parameter prediction model to predict the target display position corresponding to the advertising sample data; and the training submodule is used to adjust the model parameters in the advertising conversion parameter prediction model that are not masked based on the model loss.
[0009] In one possible implementation, the parameter prediction model includes a first sub-model and a second sub-model, and the parameter prediction module includes a feature acquisition sub-module, a first parameter prediction sub-module, and a second parameter prediction sub-module. A feature acquisition submodule is used to multiply each feature vector by the importance weight corresponding to the feature vector to obtain the target feature corresponding to each feature vector; a first parameter prediction submodule is used to use the first submodel to perform classification prediction based on multiple target features corresponding to the advertising sample data to obtain the first advertising conversion parameter corresponding to each display position of the advertising sample data; a second parameter prediction submodule is used to use the second submodel to perform classification prediction based on multiple target features corresponding to the advertising sample data to obtain the second advertising conversion parameter corresponding to each display position of the advertising sample data; the loss acquisition module is also used to calculate the model loss based on the advertising conversion label of the display position corresponding to the advertising sample data, the first advertising conversion parameter corresponding to each display position of the advertising sample data, and the second advertising conversion parameter corresponding to each display position of the advertising sample data; the model training module is also used to adjust the model parameters of the weight prediction model according to the model loss, adjust the model parameters in the first submodel used to predict the advertising sample data corresponding to the target display position, and adjust the model parameters of the second submodel.
[0010] In one possible implementation, the first sub-model is a deep feature cross-network model, and the second sub-model is a linear regression model or a machine learning model based on factor decomposition.
[0011] In one possible implementation, the parameter prediction model also includes a third sub-model, and the parameter prediction module also includes a third parameter prediction sub-module, and the third parameter prediction sub-module is used to use the third sub-model to perform classification prediction based on multiple target features corresponding to the advertising sample data to obtain a third advertising conversion parameter corresponding to each display position of the advertising sample data, wherein the second sub-model is a linear regression model and the third sub-model is a machine learning model based on factor decomposition; the loss acquisition module is also used to calculate the model loss based on the advertising conversion label corresponding to the display position of the advertising sample data and the first advertising conversion parameter corresponding to each display position of the advertising sample data, the second advertising conversion parameter corresponding to each display position of the advertising sample data, and the third advertising conversion parameter corresponding to each display position of the advertising sample data.
[0012] In one possible implementation, the model training module is further used to adjust the model parameters in the advertising conversion parameter prediction model for predicting that the display position of the advertising sample data is the target display position based on the model loss when the advertising type in the advertising sample data is the target advertising type; and is used to adjust at least one model parameter of the advertising conversion parameter prediction model based on the model loss when the advertising type in the advertising sample data is not the target advertising type.
[0013] In a fourth aspect, an embodiment of the present application provides an advertising conversion parameter prediction device, the device comprising: a feature acquisition module, an information prediction module, and an information determination module, the feature acquisition module being used to acquire user-side features of a target user and advertising-side features of a plurality of advertising data to be published; the information prediction module being used to obtain advertising conversion parameters of each display position corresponding to the advertising data to be published based on the user-side features and the advertising-side features of each of the advertising data to be published using an advertising conversion parameter prediction model. The information determination module is used to determine target advertising data and the display position corresponding to the target advertising data from the plurality of advertising data to be published based on the advertising conversion parameters of each of the advertising data to be published corresponding to each display position.
[0014] In one possible implementation, the feature acquisition module includes a request acquisition submodule, a user-side feature acquisition submodule, and an advertisement-side feature acquisition submodule. The request acquisition submodule is used to acquire a content request, wherein the content request carries a user identifier and a search keyword of a target user; the user-side feature acquisition submodule is used to acquire the user-side features of the target user according to the content request; and the advertisement-side feature acquisition submodule is used to acquire multiple advertisement data to be published according to the search keyword and extract advertisement-side features of each advertisement data to be published.
[0015] In one possible implementation, the information determination module is also used to select the to-be-published advertising data corresponding to the advertising conversion parameter with the largest value as the target advertising data based on the advertising conversion parameters corresponding to each display position of each to-be-published advertising data, and use the display position corresponding to the advertising conversion parameter with the largest value in the target advertising data as the display position corresponding to the target advertising data.
[0016] In a fifth aspect, an embodiment of the present application provides an electronic device, comprising a processor and a memory; one or more programs are stored in the memory and configured to be executed by the processor to implement the above method.
[0017] In a sixth aspect, an embodiment of the present application provides a computer-readable storage medium, in which a program code is stored, wherein the above method is executed when the program code is executed by a processor.
[0018] In a seventh aspect, an embodiment of the present application provides a computer program product or a computer program, the computer program product or the computer program including computer instructions, the computer instructions being stored in a computer-readable storage medium. A processor of a computer device obtains the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the above method.
[0019] The embodiments of the present application provide a method for training an advertising conversion parameter prediction model, a prediction method, a device and an apparatus. The method includes: performing vector conversion on the advertising sample data to obtain the feature vectors included in the advertising sample data; using the weight prediction model to obtain the importance weights of each feature vector; using the parameter prediction model to perform classification prediction based on each feature vector and the importance weights of each feature vector, and obtain the advertising conversion parameters corresponding to each display position of the advertising sample data; calculating the model loss based on the advertising conversion labels corresponding to the display positions of the advertising sample data and the advertising conversion parameters corresponding to each display position of the advertising sample data; adjusting the model parameters used to predict the target display position in the advertising conversion parameter prediction model according to the model loss until the training end condition is reached, and obtaining the trained advertising conversion parameter prediction model. By adopting the above method, the feature importance between different features can be learned, thereby strengthening the model's perception of feature importance, and introducing model parameters for display position prediction in the model structure, and adjusting parameters for predictions between different display positions during the model training process, thereby effectively reducing the impact of the prediction deviation of the display position on the advertising display effect, thereby improving the conversion rate and display effect of the display data. BRIEF DESCRIPTION OF THE DRAWINGS
[0020] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.
[0021] Figure 1 An application scenario diagram of an advertisement conversion parameter prediction model training method provided by an embodiment of the present application is shown;
[0022] Figure 2 A schematic diagram of a flow chart of a method for training an advertisement conversion parameter prediction model proposed in an embodiment of the present application is shown;
[0023] Figure 3 Another schematic diagram of the process of training an advertisement conversion parameter prediction model proposed in an embodiment of the present application is shown;
[0024] Figure 4A schematic diagram showing a flow chart of an advertisement conversion parameter prediction method proposed in an embodiment of the present application;
[0025] Figure 5 A flowchart of an online advertisement recommendation provided by an embodiment of the present application is shown;
[0026] Figure 6 A schematic diagram of the model structure of an advertisement conversion parameter prediction model proposed in an embodiment of the present application is shown;
[0027] Figure 7 A schematic diagram of a search page of an information interaction platform provided in an embodiment of the present application is shown;
[0028] Figure 8 A schematic diagram showing the comparison of the AUC of an advertisement conversion parameter prediction method provided by an embodiment of the present application and the AUC of an existing prediction method over time;
[0029] Fig. 9 A connection block diagram of a training device for an advertisement conversion parameter prediction model proposed in an embodiment of the present application is shown;
[0030] Fig.10 A connection block diagram of an advertisement conversion parameter prediction device provided by an embodiment of the present application is shown;
[0031] Fig.11 A structural block diagram of an electronic device for executing the method of an embodiment of the present application is shown. DETAILED DESCRIPTION
[0032] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that this application will be more comprehensive and complete and fully convey the concept of the example embodiments to those skilled in the art.
[0033] In addition, described feature, structure or characteristic can be combined in one or more embodiments in any suitable manner. In the following description, many specific details are provided to provide a full understanding of the embodiments of the present application. However, those skilled in the art will appreciate that the technical scheme of the present application can be put into practice without one or more of the specific details, or other methods, components, devices, steps, etc. can be adopted. In other cases, known methods, devices, realizations or operations are not shown or described in detail to avoid blurring the various aspects of the application.
[0034] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities may be implemented in software form, or in one or more hardware modules or integrated circuits, or in different networks and / or processor devices and / or microcontroller devices.
[0035] The flowcharts shown in the accompanying drawings are only exemplary and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps can be decomposed, and some operations / steps can be combined or partially combined, so the actual execution order may change according to actual conditions.
[0036] It should be noted that the "multiple" mentioned in this article refers to two or more. "And / or" describes the association relationship of the associated objects, indicating that there can be three relationships. For example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone. The character " / " generally indicates that the associated objects before and after are in an "or" relationship.
[0037] It should also be noted that in the embodiments of the present application, the acquisition of information such as job type, education level, age, gender, life stage (for example, whether married), work location, and favorite products in the user data requires the user's permission or consent, and the collection, use, processing and storage of information such as the user's job type, education level, age, life stage (for example, whether married), work location, and favorite products must comply with the regulations of the region where the user is located.
[0038] With the research and advancement of artificial intelligence technology, artificial intelligence technology has been studied and applied in many fields and played an increasingly important role.
[0039] Artificial Intelligence (AI) is the 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. In other words, artificial intelligence is a comprehensive technology in computer science that attempts to understand the essence of intelligence and produce a new intelligent machine that can respond in a similar way to human intelligence. Take the application of artificial intelligence in machine learning as an example:
[0040] Among them, Machine Learning (ML) is a multi-disciplinary interdisciplinary subject involving probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory and other disciplines. It specializes in studying how computers simulate or implement human learning behavior to acquire new knowledge or skills and reorganize existing knowledge structures to continuously improve their performance. Machine learning is the core of artificial intelligence and the fundamental way to make computers intelligent. Its applications are spread across all areas of artificial intelligence. The solution of this application mainly uses machine learning to predict the display location of data and select data.
[0041] The following describes an exemplary application of the device for performing the above-mentioned prediction of placement positions and data selection provided by an embodiment of the present invention. The prediction model training method provided by an embodiment of the present invention can be applied to the following: Figure 1 In the server in the application environment shown.
[0042] Figure 1 is a schematic diagram of an application scenario according to an embodiment of the present application, such as Figure 1 As shown, the application scenario includes a terminal device 10 and a server 20 that is communicatively connected to the terminal device 10 via a network.
[0043] The terminal device 10 may be a mobile phone, a computer, a tablet computer or a vehicle-mounted terminal, etc. The terminal device 10 may be provided with a client for displaying data (eg, displaying advertisement data to be released).
[0044] The network may be a wide area network or a local area network, or a combination of the two. The terminal device 10 may be a smart phone, a smart TV, a tablet computer, a laptop computer, or a desktop computer.
[0045] Server 20 can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers. It can also be a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN, as well as big data and artificial intelligence platforms.
[0046] If using Figure 1When the terminal device 10 and the server 20 in the training of the prediction model and the prediction model are used for prediction, the terminal device 10 can obtain sample data and send the obtained sample data to the server 20, so that after obtaining the sample data, the server 20 performs vector conversion on the user-side feature data and the advertisement-side feature data in the advertisement sample data to obtain multiple feature vectors included in the advertisement sample data; uses the weight prediction model to obtain the importance weight of each feature vector; uses the parameter prediction model to perform classification prediction based on the multiple feature vectors included in the advertisement sample data and the importance weight of each feature vector, and obtains the advertisement conversion parameters corresponding to each display position of the advertisement sample data; calculates the model loss based on the advertisement conversion label corresponding to the display position of the advertisement sample data and the advertisement conversion parameters corresponding to each display position of the advertisement sample data; adjusts the model parameters used to predict the advertisement sample data corresponding to the target display position in the advertisement conversion parameter prediction model according to the model loss, until the training end condition is met to obtain the trained advertisement conversion parameter prediction model.
[0047] The server 20 can deploy the trained advertising conversion parameter prediction model online to predict advertising conversion parameters and display advertising for the advertising data to be published. The terminal device can send a content acquisition request to the server, and the server can obtain the user-side characteristics of the target user and the advertising-side characteristics of multiple advertising data to be published based on the content acquisition request; the advertising conversion parameter prediction model obtained by the above training is used to obtain the advertising conversion parameters of each display position corresponding to the advertising data to be published based on the user-side characteristics and the advertising-side characteristics of each advertising data to be published; based on the advertising conversion parameters of each advertising data to be published corresponding to each display position, the target advertising data and the display position corresponding to the target advertising data are determined from the multiple advertising data to be published, and the target advertising data is displayed based on the display position by the terminal device 20. Figure 1 A schematic diagram showing that when a terminal device inputs a content acquisition request including a keyword in a search box of a content interaction platform to acquire data related to the keyword, and the data related to the keyword includes advertisement data, a server displays the display position of the advertisement data including the keyword fed back based on the content acquisition request. Figure 1 The schematic diagram specifically shows that when there are four display positions, namely, display positions 1 to 4, the advertisement data is displayed in display position 3.
[0048] By adopting the above-mentioned advertising conversion parameter prediction model, it is possible to introduce model parameters for display position prediction into the model structure, and adjust the model parameters used to predict the advertising sample data corresponding to the target display position in the advertising conversion parameter prediction model during the model training process, thereby effectively reducing the impact of the display position prediction deviation on the display effect (advertising conversion parameters). Subsequently, when the prediction model is used to obtain the selection prediction results (target advertising data) and display positions of multiple advertising data to be released, and the advertising data to be released is released and displayed based on the obtained selection prediction results and display positions of each advertising data to be released, the accuracy of data selection prediction can be effectively improved, avoiding the problem in related technologies that when users select data to be released, their selection results are mainly related to subjective factors such as user attention, emotions, habits, etc., rather than relying on the user's real interests, thereby affecting the conversion efficiency of the data.
[0049] It should be understood that the above method steps may also be executed only by the terminal device 10 or only by the server 20, that is, the above-described method steps are merely illustrative and are not intended to limit the present solution.
[0050] The embodiments of the present application will be described in detail below with reference to the accompanying drawings.
[0051] Please read Figure 2 , Figure 2 The present application also provides a training method for an advertisement conversion parameter prediction model, which can be applied to an electronic device, which can be the above-mentioned terminal device 10 or server 20. The method includes:
[0052] Step S110: Acquire advertisement sample data, where the advertisement sample data carries an advertisement conversion tag corresponding to the target display position of the advertisement sample data.
[0053] The above-mentioned advertising conversion labels may specifically be user click rate, collection rate, search rate, comment rate or purchase rate, etc.
[0054] The obtained advertisement sample data may include at least user-side data and advertisement-side content data. The obtained advertisement sample data may also include one or more of context feature data, keywords, and the like.
[0055] Among them, user-side data may include one or more of age, gender, job type, education, life stage (such as whether married), work location, and favorite products, and content data on the advertising side may include one or more of the categories of advertising content, the number of times the advertising content is viewed, the number of times the advertising content is collected, the advertiser ID, the advertising industry category, the type of advertising, and the advertising optimization target. Contextual features include one or more of the time or number of times the data is selected. Keywords can specifically be words such as the function or name of an item, or they can be search terms used by users to search for the above content. When keywords are search terms, the search terms are unique features of sample data in the data search scenario, and their main function is to accurately match the strong behavioral information of users when searching.
[0056] It should be understood that the sample data obtained above should be multiple, and the method of obtaining the sample data above can be to obtain the sample data from the server, or to receive the sample data input by the user. It should be understood that the method of obtaining the sample data above is only illustrative, and there can be more ways to obtain it, which is not specifically limited in this embodiment.
[0057] In the advertising conversion label corresponding to the target position of the advertising sample data, the target display position refers to the display position of the advertising sample data when it is actually displayed, and the advertising conversion label corresponding to the target position refers to the advertising conversion data when the advertising sample data is displayed at the target display position, such as advertising conversion rate, advertising collection rate or advertising click-through rate, etc., which are used to characterize the conversion effect of the advertising sample data.
[0058] Step S120: performing vector conversion on the user-side feature data and the advertisement-side feature data in the advertisement sample data to obtain a plurality of feature vectors included in the advertisement sample data.
[0059] The method of performing vector conversion on the user-side feature data in the advertisement sample data may be to encode the user-side feature data and the advertisement-side feature data in the advertisement sample data using a word vector embedding encoding method to obtain a plurality of feature vectors included in the advertisement sample data. Alternatively, the advertisement sample data may be converted into a vector using a vector conversion function.
[0060] By performing vector conversion on the user-side data in the advertisement sample data, the user-side feature vector in the advertisement sample data is obtained, and by performing feature extraction on the advertisement-side content data in the advertisement sample data, the advertisement-side content feature vector is obtained to obtain a plurality of feature vectors including the user-side feature vector and the advertisement-side content feature vector. It should be understood that if the advertisement sample data is a commodity, the above-mentioned search term features may be features corresponding to the commodity name, features corresponding to the function of the commodity, etc. The above-mentioned search term features are only illustrative, and it should be understood that the corresponding search term features in different search scenarios are different.
[0061] Step S130: Obtain the importance weight of each feature vector using a weight prediction model.
[0062] Among them, the above-mentioned weight prediction model can be any model or algorithm that can learn the importance of different feature vectors, such as: using a lightweight attention module (CBAM module, Convolutional Block Attention Module) or SENet (Squeeze-and-Excitation-network) network model. Exemplarily, the weight prediction model can be a SENet (Squeeze-and-Excitation-network) network model, which can first perform a Squeeze operation on the feature map obtained by convolution to obtain a channel-level global feature, and then perform an Excitation operation on the global feature to learn the relationship between each feature, and also obtain the importance weights of different features, and finally multiply the original feature to obtain the final feature. It should be noted that the SENet network model is universal, which means that it can be embedded in the existing network architecture, that is, it can be embedded in the feature extraction stage of this application.
[0063] Step S140: using the parameter prediction model to perform classification prediction based on multiple feature vectors included in the advertisement sample data and the importance weight of each feature vector, and obtaining advertisement conversion parameters of each of the multiple display positions corresponding to the advertisement sample data.
[0064] Among them, the above-mentioned parameter prediction model can be a machine learning model that can be used to predict the advertising conversion parameters corresponding to each display position, which may include one or more of a convolutional neural network model, a deep neural network model, a deep feature cross network model, a logistic regression model, or a machine learning model based on factor decomposition.
[0065] Among them, the parameter prediction model may include a feature extraction layer and a classification layer. The feature extraction layer can be used to multiply each feature vector by the importance weight corresponding to the feature vector to obtain the target feature corresponding to each feature vector, and perform deep feature extraction based on the target feature corresponding to each feature vector. The classification layer can perform classification calculations based on the extracted deep features to obtain advertising sample data corresponding to the advertising conversion parameters of each display position.
[0066] Multiple display locations refer to locations where the sample advertisement data may be displayed when being displayed. The number of multiple display locations may be 4, 5, or 6, etc., which is not specifically limited in the embodiments of the present application. The multiple display locations may be determined based on the number of specific locations corresponding to the target display locations corresponding to each sample advertisement data in the multiple sample advertisement data, or may be pre-set, and may be set according to actual conditions.
[0067] Step S150: Calculate the model loss based on the advertisement conversion label corresponding to the target display position of the advertisement sample data and the advertisement conversion parameters corresponding to each display position of the advertisement sample data.
[0068] Specifically, the loss function can be used to calculate the loss of the advertising conversion parameters corresponding to each display position of the advertising sample data and the advertising conversion labels corresponding to the target display position of the advertising sample data to obtain the model loss, wherein the above-mentioned loss function can be a mean square error loss function, a cross entropy loss function or an exponential loss function, etc.
[0069] It should be understood that if the above-mentioned advertising conversion parameter prediction model includes at least one machine learning model, the above-mentioned step S150 may be to obtain the loss value in the training process of each machine learning model, and perform weighted summation on the loss value corresponding to each machine learning model to obtain the model loss of the advertising conversion parameter prediction model.
[0070] Step S160: adjusting the model parameters in the advertisement conversion parameter prediction model used to predict the advertisement sample data corresponding to the target display position according to the model loss, until the training end condition is reached.
[0071] Among them, the model parameters in the advertising conversion parameter prediction model used to predict that the advertising sample data corresponds to the target display position may specifically include the model parameters in the classification layer of the parameter prediction model used to predict that the sample data belongs to the target display position, other model parameters in the parameter prediction model except the classification layer, and model parameters of other models in the advertising parameter prediction model except the parameter prediction model.
[0072] When the number of training times reaches a preset number, or when the model loss is less than a preset loss value, the training end condition is met, and the trained advertising conversion parameter prediction model is obtained.
[0073] In order to achieve the process of adjusting the model parameters used for predicting the target display positions corresponding to the advertising sample data in the advertising conversion parameter prediction model based on the model loss, in the process of obtaining the trained prediction model, avoid adjusting the model parameters other than the model parameters corresponding to the target display positions in the classification layer of the parameter prediction model for predicting the model parameters corresponding to multiple display positions. Before adjusting the parameters, the above-mentioned other model parameters can be masked, and the above-mentioned other model parameters or the model parameters corresponding to the target display positions can be marked.
[0074] When adjusting parameters later, you can adjust the model parameters after masking, or you can adjust the unmarked model parameters, or you can adjust the marked model parameters. You can set them according to actual needs, and there is no specific limitation in this embodiment.
[0075] In one possible implementation of the present application, the execution of the above-mentioned step S160 may specifically include: step S162: masking the target model parameters in the advertising conversion parameter prediction model, the target model parameters being other model parameters except the model parameters in the classification layer of the parameter prediction model used to predict the target display position corresponding to the advertising sample data; step S164: based on the model loss, adjusting the model parameters in the advertising conversion parameter prediction model that are not masked.
[0076] By adopting the above method, since the target parameters have been masked, when adjusting the model parameters, the masked target parameters will not be adjusted.
[0077] In another possible implementation of the present application, before executing the above step S160, the target model parameters in the advertisement conversion parameter prediction model may be marked, and the target model parameters are other model parameters except the model parameters corresponding to the target display position in the model parameters corresponding to the multiple display positions in the classification layer in the parameter prediction model. When executing step S160, the unmarked model parameters in the prediction model are adjusted based on the model loss to obtain a trained prediction model.
[0078] By using mask processing or marking methods, it is possible to achieve physical decoupling between the model parameters corresponding to the target display position in the classification layer in the parameter prediction model and the model parameters corresponding to other display positions, thereby eliminating the influence of the display position on the user's attention, subjective factors, etc. when the user selects the data he needs to view, so that when the user selects the data he needs to view, his selection intention is only related to the user's interest.
[0079] It should be understood that after completing a parameter adjustment, the masked data should be demasked and the marked data should be removed in order to adjust the parameters again later.
[0080] Among them, the advertising sample data usually includes the type of advertising. Among them, the advertising type can be bidding advertising, contract advertising or other advertising categories. Among them, bidding advertising is a form of advertising that pays according to the advertising effect (such as click-through rate, conversion rate, etc.); advertisers can give a bid for the advertisements they place. When an exposure request arrives, the bidding advertisements whose corresponding targeting conditions match the exposure request can compete for the exposure request based on the bids given in advance by the advertisers. Contract advertising is an advertisement generated in the following way: the advertiser signs a contract with the advertising delivery platform, requiring the advertising delivery platform to play a predetermined amount of advertisements to users of the specified type of advertiser within a specified time. If the contract is reached, the advertiser needs to pay the corresponding advertising delivery fee to the advertising delivery platform. If the contract is not reached, that is, the actual amount of advertisements played does not reach its corresponding predetermined amount of advertisements, the advertising delivery platform needs to compensate the advertiser for a certain amount of fees. When playing such contract advertisements, if the actual amount of advertisements played exceeds its corresponding predetermined amount of advertisements, the advertising delivery platform will not charge additional fees. Other advertisements refer to advertisements other than the above two types of advertisements.
[0081] In order to achieve ad position prediction and click-through rate prediction for advertisements, in one possible implementation mode of the present application, if the ad type in the ad sample data is the target ad type, based on the model loss, the model parameters used to predict the display position of the ad sample data as the target display position in the ad conversion parameter prediction model are adjusted; if the ad type in the ad sample data is not the target ad type, based on the model loss, at least one model parameter of the ad conversion parameter prediction model is adjusted.
[0082] The target advertisement type may be a bidding advertisement, or a bidding advertisement and a contract advertisement. It can be set according to actual needs. In one possible implementation of the present application, the target advertisement type is a bidding advertisement.
[0083] By adopting the above-mentioned advertising conversion parameter prediction model training method, in the model training stage, the weight prediction model is used to obtain the importance weight of each feature vector, and the parameter prediction model is used to perform classification prediction based on the multiple feature vectors included in the advertising sample data and the importance weight of each feature vector, so as to obtain the advertising conversion parameters of each display position among the multiple display positions corresponding to the advertising sample data, calculate the model loss based on the advertising conversion label corresponding to the target display position of the advertising sample data and the advertising conversion parameters corresponding to each display position of the advertising sample data, and when adjusting the model parameters based on the model loss, the model's perception of the importance of features can be enhanced by learning the feature importance between different features, and subsequently specifically adjust the model parameters used to predict the advertising sample data corresponding to the target display position in the advertising conversion parameter prediction model to obtain the trained advertising conversion parameters. When predicting the model, it is possible to introduce model parameters for display position prediction into the model structure, and adjust the parameters of the model parameters corresponding to the target display position during the model training stage. This can achieve physical decoupling between the model parameters corresponding to the target display position and the model parameters corresponding to other display positions in the classification layer of the parameter prediction model, thereby eliminating the influence of the display position on the user's attention, subjective factors, etc. when the user selects the data to be viewed, so that when the user selects the data to be viewed, the user's selection intention is only related to the user's interest, thereby effectively reducing the influence of the display position prediction deviation on the display effect (such as reducing the influence of the display position of goods or advertisements on the purchase rate, and reducing the influence of the display position of text, goods, advertisements or videos on the click-through rate), thereby improving the accuracy of the advertising conversion parameter prediction model.
[0084] See also Figure 3 The present application also provides an advertisement conversion parameter prediction model training method, the method comprising:
[0085] Step S210: Acquire advertisement sample data, where the advertisement sample data carries an advertisement conversion tag corresponding to the target display position of the advertisement sample data.
[0086] Step S220: performing vector conversion on the user-side feature data and the advertisement-side feature data in the advertisement sample data to obtain a plurality of feature vectors included in the advertisement sample data.
[0087] Step S230: Obtain the importance weight of each feature vector using the weight prediction model.
[0088] Step S240: for each feature vector, multiply the feature vector by the importance weight corresponding to the feature vector to obtain the target feature corresponding to each feature vector.
[0089] Step S250: Utilize the first sub-model to perform classification prediction based on multiple target features corresponding to the advertisement sample data, and obtain the first advertisement conversion parameter corresponding to each display position of the advertisement sample data.
[0090] The first sub-model can be any one of a convolutional neural network model, a deep neural network model, a logistic regression model, a deep feature cross network model, or a machine learning model based on factor decomposition.
[0091] In one possible implementation, the first sub-model may be a factorization-based machine learning model.
[0092] Among them, the machine learning model based on factor decomposition can specifically be any one of the FM model (FM, Factorization Machines), FwFM model (FwFM, Field-weighted Factorization Machines, weighted Field factorization machine), FmFM model (FmFM, Field-matrixed Factorization Machines, matrix Field factorization machine), FwFFM model (FwFFM, Field weight factorization machine) or FINT model (FINT, Field-aware INTeraction Neural Network, Field-based cross neural network), etc.
[0093] In one possible implementation of the present application, the above-mentioned factor decomposition-based machine learning model can specifically be a FwFFM model.
[0094] When training the FwFFM model, the above step S230 can specifically be: using the first sub-model to perform feature transformation processing on the feature vector to obtain the feature vector after feature transformation; using the first sub-model to normalize the feature vector after feature transformation to obtain the normalized feature vector; using the first sub-model to perform feature abstraction processing on the normalized feature vector to obtain the abstracted feature vector; using the first sub-model to obtain the first advertising conversion parameter of each of the multiple display positions corresponding to the advertising sample data based on the abstracted feature vector.
[0095] When performing feature transformation processing, the sample features may be subjected to linear transformation processing, or the feature vector may be subjected to encoding processing. When normalizing the feature vector after feature transformation processing, the value of the feature vector after feature transformation processing may be normalized to [0, 1], and the specific normalization method may be to perform normalization processing using a normalization function. When performing feature abstraction processing on the normalized feature vector, the hidden layer in the model may be used for abstraction processing.
[0096] In another possible implementation of the present application, the first sub-model is a deep feature cross network model, that is, the first sub-model is a model constructed based on a deep cross network (Deep & Cross Network, DCN). In the model constructed based on the deep cross network (Deep & Cross Network), the feature vector is first input into an embedding and stacking layer, and then the output of the embedding and stacking layer is input into a cross network and a deep network (deep network) parallel to the cross network, and then the outputs of the cross network and the deep network are combined through a combination output layer (Combination output Layer) for prediction, and the output training advertising sample data corresponds to the advertising conversion parameters of each of the multiple placements.
[0097] The first sub-model built on the Deep & Cross Network (DCN) explicitly applies feature crossover on each layer, which can effectively learn bounded predictive crossover features without manual feature engineering or exhaustive search. Moreover, the network is simple and effective, the polynomial series of each layer is the highest and is determined by the layer depth. The network consists of all cross terms, and the cross term coefficients are different. And the cross-network memory is efficient, memory-saving, and easy to implement. At the same time, the number of parameters is reduced by nearly an order of magnitude compared with DNN in cross entropy (LogLoss). The ReLu function can be used as an activation function in this model; and Dropout can be added during the training process, that is, for the neural network unit, the process of temporarily discarding it from the network according to a certain probability is added. Then, the first sub-model built on the Deep & Cross Network (DCN) is trained to obtain efficient and reliable high-performance prediction of advertising conversion parameters for each of multiple placements to improve the accuracy of the prediction results.
[0098] Step S260: Utilize the second sub-model to perform classification prediction based on multiple target features corresponding to the advertisement sample data, and obtain second advertisement conversion parameters corresponding to each display position of the advertisement sample data.
[0099] The model types of the first sub-model and the second sub-model may be the same or different. The second sub-model may be any one of a convolutional neural network model, a deep neural network model, a logistic regression model, a deep feature cross network model, or a machine learning model based on factor decomposition.
[0100] It should be understood that the above-mentioned method of obtaining the second advertisement conversion parameter corresponding to the advertisement sample data for each display position should be similar to the method of obtaining the first advertisement conversion parameter corresponding to the advertisement sample data for each display position, and will not be described one by one in the embodiments of the present application.
[0101] In one possible implementation of the present application, the first sub-model is a deep feature cross-network model, and the second sub-model is a linear regression model or a machine learning model based on factor decomposition.
[0102] Step S270: Calculate the model loss based on the advertisement conversion label of the display position corresponding to the advertisement sample data, the first advertisement conversion parameter of each display position corresponding to the advertisement sample data, and the second advertisement conversion parameter of each display position corresponding to the advertisement sample data.
[0103] In one possible implementation, the above-mentioned model loss can be obtained by calculating the first loss using the advertising conversion label of the display position corresponding to the advertising sample data and the first advertising conversion parameter of each display position corresponding to the advertising sample data, and calculating the second loss using the advertising conversion label of the display position corresponding to the advertising sample data and the second advertising conversion parameter of each display position corresponding to the advertising sample data, and performing a weighted sum of the first loss and the second loss to obtain the model sum.
[0104] In another possible implementation, the above-mentioned model loss may also be obtained by using a first loss calculation formula corresponding to the first sub-model to calculate the loss based on the advertisement conversion label of the display position corresponding to the advertisement sample data, and a first advertisement conversion parameter of each display position corresponding to the advertisement sample data, and using a second loss calculation formula corresponding to the second sub-model to calculate the loss based on the advertisement conversion label of the display position corresponding to the advertisement sample data, and a first advertisement conversion parameter of each display position corresponding to the advertisement sample data, to obtain a second loss, and selecting the loss with a larger loss value between the first loss and the second loss as the model loss.
[0105] It should be understood that the above-mentioned method of obtaining the target loss is only illustrative, and there may be other methods of obtaining the target loss, which are not specifically limited in this embodiment.
[0106] Step S280: adjusting the model parameters of the weight prediction model according to the model loss, adjusting the model parameters of the first sub-model used to predict the advertising sample data corresponding to the target display position, and adjusting the model parameters of the second sub-model.
[0107] Among them, regarding the specific description of adjusting the model parameters of the weight prediction model according to the model loss in the above steps, and adjusting the model parameters in the first sub-model used to predict the advertising sample data corresponding to the target display position, please refer to the specific description of step S160 above, which will not be repeated one by one in this embodiment.
[0108] The method of adjusting the model parameters in the second sub-model based on the model loss may be to adjust at least one model parameter in the second sub-model.
[0109] It should be understood that when the number of training times reaches a preset number, or when the model loss is less than a preset loss value, a trained advertising conversion parameter prediction model is obtained, and the parameter prediction model in the advertising conversion parameter prediction model includes a first sub-model and a second sub-model.
[0110] In one possible implementation, the parameter prediction model may include other sub-models in addition to the first sub-model and the second sub-model, such as a third sub-model.
[0111] The method also includes:
[0112] Step S290: Use the third sub-model to perform classification prediction based on multiple target features corresponding to the advertising sample data to obtain a third advertising conversion parameter corresponding to each display position of the advertising sample data, wherein the second sub-model is a linear regression model and the third sub-model is a machine learning model based on factor decomposition.
[0113] The above-mentioned step S270 can specifically be: calculating the model loss based on the advertising conversion label corresponding to the display position of the advertising sample data and the first advertising conversion parameter corresponding to each display position of the advertising sample data, the second advertising conversion parameter corresponding to each display position of the advertising sample data, and the third advertising conversion parameter corresponding to each display position of the advertising sample data.
[0114] Accordingly, based on the model loss, the model parameters that are not masked in the ad conversion parameter prediction model are adjusted, which also includes:
[0115] The model parameters of the third sub-model are adjusted based on the model loss to obtain the trained third sub-model.
[0116] The manner of adjusting the model parameters of the third sub-model based on the model loss may be similar to the manner of adjusting the model parameters of the second sub-model based on the model loss, and will not be elaborated again.
[0117] By adopting the above-mentioned method of the present application, the model parameters for display position prediction are introduced into the model structure of the first sub-model, and the model parameters corresponding to the target display position in the classification layer of the first sub-model are adjusted during the model training stage, so that the model parameters corresponding to the target display position in the classification layer of the first sub-model and the model parameters corresponding to other display positions can be physically decoupled, thereby eliminating the influence of the attention and subjective factors brought to the user by the display position when the user selects the advertisement to be viewed, so that when the user selects the advertisement to be viewed, the selection intention of the user is only related to the user's interest, thereby effectively reducing the influence of the prediction deviation of the display position on the display effect. Furthermore, in the training stage of the advertising conversion parameter prediction model, multiple sub-models in the advertising conversion parameter prediction model are used to participate in the training together, so that the trained advertising conversion parameter prediction model is more accurate, and the stability of the conversion parameter prediction model can also be improved, thereby achieving more accurate prediction of the advertising conversion parameters of each display position in the subsequent period.
[0118] See also Figure 4 The present application also provides an advertisement conversion parameter prediction method, the method comprising:
[0119] Step S310: Acquire user-side features of a target user and advertisement-side features of a plurality of advertisement data to be published.
[0120] The method of obtaining the user-side features of the target user may be to obtain the user-side features of the target user input by the user; or to obtain the user ID input by the user, and search the database for the user-side features corresponding to the user ID based on the user ID.
[0121] The method for obtaining the advertising side features of multiple advertising data to be published may be to obtain data identifiers corresponding to the multiple advertising data to be published, obtain the advertising data to be published according to each data identifier, and perform feature extraction on the advertising data to be published to obtain the advertising side features of the advertising data to be published; or it may be to select the advertising data to be published related to the preset keywords from the data according to the preset keywords; perform feature extraction on the advertising data to be published to obtain the advertising side features of the advertising data to be published.
[0122] In one possible implementation of the present application, the above step S310 includes:
[0123] Step S312: Obtain a content request, which carries the user ID of the target user and the search keyword.
[0124] The above-mentioned content request may be obtained based on a user's voice search operation, or may be obtained based on a user's content search operation.
[0125] Step S314: Obtain user-side features of the target user according to the content request.
[0126] The above method of obtaining the user-side feature of the target user according to the content request may specifically be to search for the user-side feature corresponding to the user identifier according to the user identifier of the target user included in the content request. The server or a memory associated with the server stores multiple user identifiers and the user-side features corresponding to each user identifier.
[0127] Step S316: Acquire multiple pieces of advertisement data to be published according to the search keyword, and extract advertisement-side features of each piece of advertisement data to be published.
[0128] The method of obtaining multiple advertisement data to be published according to the search term may be to obtain multiple search data according to the search term, and determine multiple advertisement data to be published from the multiple search data according to the relevance of each search data to the search term. Alternatively, after obtaining multiple search data according to the search term, determine multiple advertisement data to be published from the multiple search data according to at least one of the corresponding display times, relevance and popularity of each search data.
[0129] When the search data is an advertisement, the popularity corresponding to the search data may be the popularity of the advertiser, brand or product.
[0130] When multiple advertisement data to be published are obtained, feature extraction is performed on each advertisement data to obtain advertisement side features of each advertisement data to be published, wherein the advertisement side features may include one or more of advertisement category, advertisement browsing times, advertisement collection times, and the like.
[0131] Step S320: using the advertisement conversion parameter prediction model to obtain advertisement conversion parameters of each display position corresponding to the advertisement data to be published based on user-side characteristics and advertisement-side characteristics of each advertisement data to be published.
[0132] It should be understood that the display position corresponding to the advertisement data to be published obtained based on the user-side characteristics and the advertisement-side characteristics of each advertisement data to be published is used to indicate at which display position the advertisement data to be published is displayed.
[0133] In order to send the advertisement data to be published that needs to be displayed to the terminal device so that the terminal device displays the advertisement data to be published that needs to be displayed, in one possible implementation of the present application, after executing step S320, the method further includes:
[0134] Step S330: based on the advertisement conversion parameters of each advertisement data to be published corresponding to each display position, target advertisement data and the display position corresponding to the target advertisement data are determined from the plurality of advertisement data to be published.
[0135] Specifically, the above steps may be, based on the advertisement conversion parameters corresponding to each display position of each to-be-published advertisement data, selecting the to-be-published advertisement data corresponding to the advertisement conversion parameter with the largest value as the target advertisement data, and using the display position corresponding to the advertisement conversion parameter with the largest value in the target advertisement data as the display position corresponding to the target advertisement data. Alternatively, based on the advertisement conversion parameters corresponding to each display position of each to-be-published advertisement data, selecting the to-be-published advertisement data corresponding to the advertisement conversion parameter with a value greater than a preset conversion parameter threshold as the target advertisement data, and using the display position corresponding to the advertisement conversion parameter with the largest value in the target advertisement data as the display position corresponding to the target advertisement data.
[0136] See also Figure 5 and Figure 6 , for example, training an advertisement conversion parameter prediction model for predicting click-through rates of advertisement data obtained through search at different display positions is used as an example for explanation, wherein the advertisement conversion parameter prediction model includes a weight prediction model, a first sub-model, a second sub-model, and a third sub-model, and the first sub-model is a deep feature cross network model, the second sub-model is a linear regression, and the third sub-model is an FM model (or an FMFFM model). In the training phase, the types of input advertisement sample data include bidding sample advertisements, contract sample advertisements, and sample advertisements of other advertisement types as examples, and the specific data processing process of the training phase and the prediction phase of the prediction model is as follows:
[0137] Training phase:
[0138] Ad samples may be obtained by obtaining from a database advertisements searched by different users in different search scenarios, and operation data obtained based on click-view operations of users on the searched advertisements, or by receiving advertisement data input by users.
[0139] Among them, the sample data in the database can specifically be an advertisement search acquisition request sent by a user through a terminal device, and in response to the advertisement search acquisition request, multiple advertisements corresponding to the search keywords are searched from a database or content server. The server uses a model to predict the click-through rate of multiple advertisements and publishes multiple advertisements based on the click-through rate prediction results, and obtains the user's selection operation results for the multiple advertisements to obtain multiple sample advertisements and store them in the database.
[0140] The sample features of the advertising sample data may specifically include user-side features, advertising-side features, search term features, and context features, among which user-side features include user gender, age, interests, etc.; advertising-side features include advertiser ID, advertising industry category, advertising type, optimization goals, etc.; context features include advertising pull time and number of advertising pulls, etc. The search term feature is a unique feature of the search-search advertising scenario, and its main function is to accurately match the strong behavioral information of users during search.
[0141] After obtaining the above-mentioned sample features, the information of these features can be subjected to a series of encryption, hashing, and discretization operations, and then mapped to generate an embedding (feature vector) with a fixed dimension. The weight prediction model is then used to obtain the importance weight of each feature vector, and each feature vector is multiplied by the corresponding importance weight to obtain the target feature corresponding to each feature vector. After that, the target feature corresponding to each feature vector is detected and input into the first sub-model, the second sub-model, and the third sub-model respectively.
[0142] The first sub-model is used to perform classification prediction based on multiple target features corresponding to the advertising sample data, and the first advertising conversion parameter corresponding to each display position of the advertising sample data is obtained. The second sub-model is used to perform classification prediction based on multiple target features corresponding to the advertising sample data, and the second advertising conversion parameter corresponding to each display position of the advertising sample data is obtained; the third sub-model is used to perform classification prediction based on multiple target features corresponding to the advertising sample data, and the third advertising conversion parameter corresponding to each display position of the advertising sample data is obtained.
[0143] After obtaining the prediction results of each sub-model for the advertising sample data, the model loss is calculated based on the advertising conversion label of the advertising sample data corresponding to the display position and the first advertising conversion parameter of the advertising sample data corresponding to each display position, the second advertising conversion parameter of the advertising sample data corresponding to each display position, and the third advertising conversion parameter of the advertising sample data corresponding to each display position.
[0144] Specifically, the click prediction result of the final first sub-model for the advertisement sample data at the target display position is marked as f θ (.), where θ is the hyperparameter of the model. In the process of constructing the loss function of the first sub-model, it is assumed that during a model training process, the advertising sample data input to the prediction model is {x 1 ,x 2 ,......,x n}, where the corresponding positions in the above n advertisement samples include positions 1, 2, ..., m. The sample label of the advertisement sample data is {y 1 ,y 2 ,......,y n}, then the loss of the ad sample data corresponding to the jth placement is: BCE j (y i ,f)=-ylogσ(f)-(1-y)log(1-σ(f)) and σ(.) represents the sigmoid function, the model loss is L, and Among them, a j is the weight coefficient of the jth placement. The physical meaning of BCE is to make the estimated score f θ (x) is close to the sample label y. The loss of the first sub-model can be calculated by using the above method. It should be understood that the calculation process of the loss of each sub-model is similar and will not be repeated here. After obtaining the loss of each sub-model, the model loss of the advertising conversion parameter prediction model can be obtained.
[0145] When adjusting model parameters based on model loss, the second sub-model and the third sub-model can adjust at least one model parameter of the model to which they belong based on the model loss, and the first sub-model can adjust the model parameters and general model parameters in its classification layer used to predict the target display position of the predicted advertising sample data based on the model loss of the bidding advertisement in the sample advertising data, wherein the general model parameters specifically refer to the model parameters that do not participate in position prediction.
[0146] Specifically, in order to enable the first sub-model to adjust the model parameters and general model parameters in its classification layer for predicting the target display position corresponding to the predicted advertising sample data based on the model loss, the target model parameters in the first sub-model can be masked, and the target model parameters are other model parameters except the model parameters in the classification layer of the parameter prediction model for predicting the target display position corresponding to the advertising sample data.
[0147] Exemplarily, if there are four display positions, namely b1, b2, b3 and b4, and the model parameter corresponding to b1 is μ1, the model parameter corresponding to b2 is μ2, the model parameter corresponding to b3 is μ3, the model parameter corresponding to b4 is μ4, and the general model parameter is β, then after masking, μ1 and β are adjusted using the loss of the sample data corresponding to b1, μ2 and β are adjusted using the loss of the sample data corresponding to b2, μ3 and β are adjusted using the loss of the sample data corresponding to b3, and μ4 and β are adjusted using the loss of the sample data corresponding to b4.
[0148] When the number of model adjustments reaches a preset number, or the model loss is less than a preset loss threshold, a trained advertising conversion parameter prediction model is obtained.
[0149] Prediction Phase
[0150] The user can input keywords through the terminal device to send an advertising content request to the server (such as an advertising platform), and the content request carries the user ID and search keywords of the target user. When the advertising platform obtains the content request, it obtains the user-side features of the target user according to the user ID, and obtains multiple to-be-published advertising data from the database (such as the content server) according to the search keywords based on the content acquisition request, and feeds back to the advertising platform, so that the advertising platform extracts the content features of each to-be-published advertising data. The trained prediction model obtained in the above-mentioned training stage is used to obtain the advertising conversion parameters of each display position corresponding to the to-be-published advertising data based on the user-side features and the advertising-side features of each to-be-published advertising data; and based on the advertising conversion parameters of each to-be-published advertising data corresponding to each display position, the target advertising data and the display position corresponding to the target advertising data are determined from multiple to-be-published advertising data.
[0151] like Figure 7 As shown, an advertisement search is performed on a content interaction platform, wherein: Figure 7 It shows that there are 4 display positions (i.e., display position 1 to display position 4) set on the content interaction platform, and the display results of the target advertisement to be published are obtained by performing an advertisement search, wherein: Figure 8 Only the display result diagram for display at display position 3 is shown. When displaying at different positions, the advertising conversion parameters of each display position corresponding to the advertising data to be published predicted by the advertising conversion parameter prediction model and the advertising conversion parameters of each display position corresponding to the advertising data to be published in the related technology are shown in the following table:
[0152] Placement number Related technology advertisement selection prediction (ctr) Ad selection prediction for this scheme (ctr) overall 3.04% 3.04% 1 14.3% 4.03% 2 3.26% 3.51% 3 2.27% 2.97% 4 1.88% 2.6%
[0153] It can be seen that the advertisement selection prediction in the related art will significantly affect the click-through rate of advertisements due to different display positions, and it is worth noting that this influence may not be related to the user's real interest, but to subjective factors such as the user's attention, emotions and habits. However, this solution makes a fine distinction in the display position information to eliminate the deviation of the model's prediction between different display positions, thereby eliminating the difference in CTR levels between different display positions.
[0154] like Figure 8As shown, in order to verify the effect of the present solution, the prediction model obtained above is deployed on an online advertising platform, and a series of offline tests are carried out. The experimental time is from 16:00 on the first day to 16:00 on the second day, and the experimental data is the search-to-search click data of the content interaction platform. The AUC (Area under the curve, the most important indicator for measuring advertisement ranking, the value range of AUC is 0-1, the closer to 1, the better the model ranking ability) is tested in the present solution and the advertisement click-through rate prediction solution used in the prior art. It can be seen that the AUC of the prediction method of the present application is improved by 0.14% (the mean AUC is improved from 0.8234 to 0.8246). In addition to offline testing, the GMV (Gross merchandise volume) of online testing also increased by more than 4%. GMV is a concept in online shopping and e-commerce, referring to the transaction amount of the website. It is used to indicate the sum of the order amount that users ultimately did not pay after taking the order (including unpaid orders placed in the shopping cart after taking the order, canceled orders, rejected orders and returned orders) and the paid order amount after taking the order.
[0155] It should be understood that, although the steps in the flowcharts involved in the above embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps is not strictly limited in order, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above embodiments may include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.
[0156] See also Fig. 9Another embodiment of the present application provides an advertising conversion parameter prediction model training device 400, the device 400 includes: a data acquisition module 410, a data conversion module 420, a weight prediction module 430, a parameter prediction module 440, a loss acquisition module 450 and a model training module 460. A data acquisition module 410 is used to acquire advertising sample data, where the advertising sample data carries an advertising conversion label corresponding to a target display position of the advertising sample data; a data conversion module 420 is used to perform vector conversion on the advertising sample data to obtain a plurality of feature vectors included in the advertising sample data; a weight prediction module 430 is used to obtain the importance weight of each feature vector using a weight prediction model; a parameter prediction module 440 is used to perform classification prediction based on a plurality of feature vectors included in the advertising sample data and the importance weight of each feature vector using a parameter prediction model to obtain advertising conversion parameters corresponding to each display position of the advertising sample data; a loss acquisition module 450 is used to calculate a model loss based on the advertising conversion label corresponding to the display position of the advertising sample data and the advertising conversion parameters corresponding to each display position of the advertising sample data; and a model training module 460 is used to adjust the model parameters used to predict the advertising sample data corresponding to the target display position in the advertising conversion parameter prediction model according to the model loss until the training end condition is met.
[0157] In one possible implementation, the model training module 460 includes a masking submodule and a training submodule. The masking submodule is used to mask the target model parameters in the advertisement conversion parameter prediction model, where the target model parameters are other model parameters except the model parameters used in the classification layer of the parameter prediction model to predict the target display position corresponding to the advertisement sample data; the training submodule is used to adjust the model parameters in the advertisement conversion parameter prediction model that are not masked based on the model loss.
[0158] In one possible implementation, the parameter prediction model includes a first sub-model and a second sub-model, and the parameter prediction module 440 includes a feature acquisition sub-module, a first parameter prediction sub-module and a second parameter prediction sub-module. The feature acquisition sub-module is used to multiply the feature vector by the importance weight corresponding to the feature vector for each feature vector to obtain the target feature corresponding to each feature vector; the first parameter prediction sub-module is used to perform classification prediction based on the first sub-model based on the multiple target features corresponding to the advertising sample data to obtain the first advertising conversion parameter corresponding to each display position of the advertising sample data; the second parameter prediction sub-module is used to perform classification prediction based on the multiple target features corresponding to the advertising sample data using the second sub-model to obtain the second advertising conversion parameter corresponding to each display position of the advertising sample data; the loss acquisition module 450 is also used to calculate the model loss based on the advertising conversion label of the display position corresponding to the advertising sample data, the first advertising conversion parameter corresponding to each display position of the advertising sample data, and the second advertising conversion parameter corresponding to each display position of the advertising sample data; the model training module 460 is also used to adjust the model parameters of the weight prediction model according to the model loss, adjust the model parameters used in the first sub-model to predict the advertising sample data corresponding to the target display position, and adjust the model parameters of the second sub-model.
[0159] In one possible implementation, the first sub-model is a deep feature cross-network model, and the second sub-model is a linear regression model or a factor decomposition-based machine learning model.
[0160] In one possible implementation, the parameter prediction model also includes a third sub-model, and the parameter prediction module also includes a third parameter prediction sub-module, the third parameter prediction sub-module is used to use the third sub-model to perform classification prediction based on multiple target features corresponding to the advertising sample data, to obtain the third advertising conversion parameter corresponding to each display position of the advertising sample data, wherein the second sub-model is a linear regression model, and the third sub-model is a machine learning model based on factor decomposition; the loss acquisition module 450 is also used to calculate the model loss based on the advertising conversion label corresponding to the display position of the advertising sample data and the first advertising conversion parameter corresponding to each display position of the advertising sample data, the second advertising conversion parameter corresponding to each display position of the advertising sample data, and the third advertising conversion parameter corresponding to each display position of the advertising sample data.
[0161] In one possible implementation, the model training module 460 is further used to adjust, based on the model loss, the model parameters in the advertising conversion parameter prediction model used to predict the display position of the advertising sample data as the target display position when the advertising type in the advertising sample data is the target advertising type; and is used to adjust, based on the model loss, at least one model parameter of the advertising conversion parameter prediction model when the advertising type in the advertising sample data is not the target advertising type.
[0162] See also Fig.10 Another embodiment of the present application provides an advertising conversion parameter prediction device 500, the device 500 includes: a feature acquisition module 510, an information prediction module 520 and an information determination module 530, the feature acquisition module 510 is used to obtain the user-side features of the target user and the advertising-side features of multiple advertising data to be published; the information prediction module 520 is used to obtain the advertising conversion parameters of the advertising data to be published corresponding to each display position based on the user-side features and the advertising-side features of each advertising data to be published using the advertising conversion parameter prediction model. The information determination module 530 is used to determine the target advertising data and the display position corresponding to the target advertising data from the multiple advertising data to be published based on the advertising conversion parameters of each advertising data to be published corresponding to each display position.
[0163] In one possible implementation, the feature acquisition module 510 includes a request acquisition submodule, a user-side feature acquisition submodule, and an advertisement-side feature acquisition submodule. The request acquisition submodule is used to acquire a content request, which carries a user identifier and a search keyword of a target user; the user-side feature acquisition submodule is used to acquire the user-side features of the target user according to the content request; and the advertisement-side feature acquisition submodule is used to acquire multiple advertisement data to be published according to the search keyword and extract the advertisement-side features of each advertisement data to be published.
[0164] In one possible implementation, the information determination module 530 is also used to select the to-be-published advertising data corresponding to the advertising conversion parameter with the largest value as the target advertising data based on the advertising conversion parameters corresponding to each display position of each to-be-published advertising data, and use the display position corresponding to the advertising conversion parameter with the largest value in the target advertising data as the display position corresponding to the target advertising data.
[0165] Each module in the above-mentioned device can be implemented in whole or in part by software, hardware and a combination thereof. Each of the above-mentioned modules can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute operations corresponding to each of the above modules. It should be noted that the device embodiment in this application corresponds to the aforementioned method embodiment. The specific principles in the device embodiment can be found in the contents of the aforementioned method embodiment, which will not be repeated here.
[0166] The following will be combined Fig.11 An electronic device provided by the present application is described.
[0167] See also Fig.11Based on the molecular property prediction model training method provided in the above embodiment, the embodiment of the present application also provides another electronic device 100 including a processor 102 that can execute the above method. The electronic device 100 can be a server or a terminal device, and the terminal device can be a smart phone, tablet computer, computer or portable computer and other devices.
[0168] The electronic device 100 further includes a memory 104 . The memory 104 stores a program that can execute the contents of the aforementioned embodiments, and the processor 102 can execute the program stored in the memory 104 .
[0169] Among them, the processor 102 may include one or more cores for processing data and a message matrix unit. The processor 102 uses various interfaces and lines to connect various parts of the entire electronic device 100, and executes various functions and processes data of the electronic device 100 by running or executing instructions, programs, code sets or instruction sets stored in the memory 104, and calling data stored in the memory 104. Optionally, the processor 102 can be implemented in at least one hardware form of digital signal processing (Digital Signal Processing, DSP), field programmable gate array (Field-Programmable Gate Array, FPGA), and programmable logic array (Programmable Logic Array, PLA). The processor 102 can integrate one or a combination of a central processing unit (Central Processing Unit, CPU), a graphics processing unit (Graphics Processing Unit, GPU) and a modem. Among them, the CPU mainly processes the operating system, user interface and application programs; the GPU is responsible for rendering and drawing display content; and the modem is used to process wireless communications. It can be understood that the above-mentioned modem may not be integrated into the processor 102, but may be implemented separately through a communication chip.
[0170] The memory 104 may include a random access memory (RAM) or a read-only memory (ROM). The memory 104 may be used to store instructions, programs, codes, code sets or instruction sets. The memory 104 may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for implementing at least one function, instructions for implementing the following various method embodiments, etc. The data storage area may also store data (e.g., advertisement sample data) acquired by the electronic device 100 during use.
[0171] The electronic device 100 may also include a network module and a screen. The network module is used to receive and send electromagnetic waves, realize the mutual conversion between electromagnetic waves and electrical signals, and thus communicate with a communication network or other devices, such as communicating with an audio playback device. The network module may include various existing circuit components for performing these functions, such as antennas, radio frequency transceivers, digital signal processors, encryption / decryption chips, user identity modules (SIM) cards, memories, etc. The network module may communicate with various networks such as the Internet, corporate intranets, wireless networks, or communicate with other devices via wireless networks. The above-mentioned wireless networks may include cellular telephone networks, wireless local area networks, or metropolitan area networks. The screen may display interface content and perform data interaction.
[0172] In some embodiments, the electronic device 100 may further include: a peripheral interface 106 and at least one peripheral device. The processor 102, the memory 104 and the peripheral interface 106 may be connected via a bus or a signal line. Each peripheral device may be connected to the peripheral interface via a bus, a signal line or a circuit board. Specifically, the peripheral device includes: at least one of a radio frequency component 108, a positioning component 112, a camera 114, an audio component 116, a display screen 118 and a power supply 122.
[0173] The peripheral interface 106 may be used to connect at least one peripheral device related to I / O (Input / Output) to the processor 102 and the memory 104. In some embodiments, the processor 102, the memory 104, and the peripheral interface 106 are integrated on the same chip or circuit board; in some other embodiments, any one or two of the processor 102, the memory 104, and the peripheral interface 106 may be implemented on a separate chip or circuit board, which is not limited in the embodiments of the present application.
[0174] The radio frequency component 108 is used to receive and transmit RF (Radio Frequency) signals, also known as electromagnetic signals. The radio frequency component 108 communicates with the communication network and other communication devices through electromagnetic signals. The radio frequency component 108 converts electrical signals into electromagnetic signals for transmission, or converts received electromagnetic signals into electrical signals. Optionally, the radio frequency component 108 includes: an antenna system, an RF transceiver, one or more amplifiers, a tuner, an oscillator, a digital signal processor, a codec chipset, a user identity module card, and the like. The radio frequency component 108 can communicate with other terminals through at least one wireless communication protocol. The wireless communication protocol includes, but is not limited to: the World Wide Web, a metropolitan area network, an intranet, various generations of mobile communication networks (2G, 3G, 4G and 5G), a wireless local area network and / or a WiFi (Wireless Fidelity) network. In some embodiments, the radio frequency component 108 may also include circuits related to NFC (Near Field Communication), which is not limited in this application.
[0175] The positioning component 112 is used to locate the current geographic location of the electronic device to implement navigation or LBS (Location Based Service). The positioning component 112 can be a positioning component based on the US GPS (Global Positioning System), Beidou system or Galileo system.
[0176] The camera 114 is used to capture images or videos. Optionally, the camera 114 includes a front camera and a rear camera. Usually, the front camera is arranged on the front panel of the electronic device 100, and the rear camera is arranged on the back of the electronic device 100. In some embodiments, there are at least two rear cameras, which are any one of a main camera, a depth of field camera, a wide-angle camera, and a telephoto camera, so as to realize the fusion of the main camera and the depth of field camera to realize the background blur function, the fusion of the main camera and the wide-angle camera to realize the panoramic shooting and VR (Virtual Reality) shooting function or other fusion shooting functions. In some embodiments, the camera 114 may also include a flash. The flash can be a monochrome temperature flash or a dual-color temperature flash. A dual-color temperature flash refers to a combination of a warm light flash and a cold light flash, which can be used for light compensation at different color temperatures.
[0177] The audio component 116 may include a microphone and a speaker. The microphone is used to collect sound waves from the user and the environment, and convert the sound waves into electrical signals and input them into the processor 102 for processing, or input them into the RF component 108 to achieve voice communication. For the purpose of stereo acquisition or noise reduction, there can be multiple microphones, which are respectively arranged at different parts of the electronic device 100. The microphone can also be an array microphone or an omnidirectional acquisition microphone. The speaker is used to convert the electrical signal from the processor 102 or the RF component 108 into sound waves. The speaker can be a traditional film speaker or a piezoelectric ceramic speaker. When the speaker is a piezoelectric ceramic speaker, it can not only convert the electrical signal into sound waves audible to humans, but also convert the electrical signal into sound waves inaudible to humans for purposes such as ranging. In some embodiments, the audio component 114 may also include a headphone jack.
[0178] The display screen 118 is used to display the UI (User Interface). The UI may include graphics, text, icons, videos, and any combination thereof. When the display screen 118 is a touch display screen, the display screen 118 also has the ability to collect touch signals on the surface or above the surface of the display screen 118. The touch signal can be input to the processor 102 as a control signal for processing. At this time, the display screen 118 can also be used to provide virtual buttons and / or virtual keyboards, also known as soft buttons and / or soft keyboards. In some embodiments, the display screen 118 can be one, and the front panel of the electronic device 100 is set; in other embodiments, the display screen 118 can be at least two, which are respectively set on different surfaces of the electronic device 100 or are folded; in some other embodiments, the display screen 118 can be a flexible display screen, which is set on the curved surface or folded surface of the electronic device 100. Even, the display screen 118 can also be set as a non-rectangular irregular shape, that is, a special-shaped screen. The display screen 118 can be made of materials such as LCD (Liquid Crystal Display), OLED (Organic Light-Emitting Diode, machine light-emitting diode).
[0179] The power supply 122 is used to power various components in the electronic device 100. The power supply 122 can be an alternating current, a direct current, a disposable battery, or a rechargeable battery. When the power supply 122 includes a rechargeable battery, the rechargeable battery can be a wired rechargeable battery or a wireless rechargeable battery. A wired rechargeable battery is a battery that is charged through a wired line, and a wireless rechargeable battery is a battery that is charged through a wireless coil. The rechargeable battery can also be used to support fast charging technology.
[0180] The embodiment of the present application also provides a structural block diagram of a computer-readable storage medium. The computer-readable medium stores program codes, which can be called by a processor to execute the method described in the above method embodiment.
[0181] The computer-readable storage medium may be an electronic memory such as a flash memory, an EEPROM (electrically erasable programmable read-only memory), an EPROM, a hard disk, or a ROM. Optionally, the computer-readable storage medium includes a non-transitory computer-readable storage medium. The computer-readable storage medium has storage space for program codes that execute any of the method steps in the above method. These program codes can be read from or written to one or more computer program products. The program code can be compressed, for example, in an appropriate form.
[0182] The embodiment of the present application also provides a computer program product or a computer program, which includes a computer instruction stored in a computer-readable storage medium. The processor of the computer device reads the computer instruction from the computer-readable storage medium, and the processor executes the computer instruction, so that the computer device performs the method described in the above various optional implementations.
[0183] In summary, the present application provides an advertising conversion parameter prediction model training method, prediction method, device and equipment. In the model training stage, a weight prediction model is used to obtain the importance weight of each feature vector, and a parameter prediction model is used to perform classification prediction based on multiple feature vectors included in the advertising sample data and the importance weight of each feature vector. By learning the feature importance between different features, the model's perception of feature importance can be enhanced. Subsequently, classification prediction is specifically used to obtain the advertising conversion parameters of each of the multiple display positions corresponding to the advertising sample data, calculate the model loss based on the advertising conversion label of the target display position corresponding to the advertising sample data and the advertising conversion parameters of each display position corresponding to the advertising sample data, and adjust the advertising conversion when the model parameters are adjusted based on the model loss. The parameter prediction model is used to predict the model parameters corresponding to the target display position of the advertising sample data. When the trained advertising conversion parameter prediction model is obtained, the model parameters corresponding to the target display position and the model parameters corresponding to other display positions can be physically decoupled, so that when the user selects the data to be viewed, the influence of the display position on the user's attention, subjective factors, etc. is eliminated, so that when the user selects the data to be viewed, the user's selection intention is only related to the user's interest, thereby effectively reducing the influence of the prediction deviation of the display position on the display effect (such as reducing the influence of the display position of goods or advertisements on the purchase rate, and reducing the influence of the display position of text, goods, advertisements or videos on the click-through rate), thereby improving the accuracy of the advertising conversion parameter prediction model.
[0184] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present application.
Claims
1. A method for training an advertising conversion parameter prediction model. It is characterized in that The advertisement conversion parameter prediction model includes a weight prediction model and a parameter prediction model. The method includes: Acquire advertisement sample data, wherein the advertisement sample data carries an advertisement conversion tag corresponding to a target display position of the advertisement sample data; Performing vector conversion on the user-side feature data and the advertisement-side feature data in the advertisement sample data to obtain a plurality of feature vectors included in the advertisement sample data; Obtaining the importance weight of each feature vector using the weight prediction model; Using the parameter prediction model to perform classification prediction based on multiple feature vectors included in the advertisement sample data and the importance weight of each feature vector, to obtain advertisement conversion parameters of each of the multiple display positions corresponding to the advertisement sample data; Calculate the model loss based on the advertisement conversion label corresponding to the target display position of the advertisement sample data and the advertisement conversion parameter corresponding to each display position of the advertisement sample data; The model parameters used to predict the advertisement sample data corresponding to the target display position in the advertisement conversion parameter prediction model are adjusted according to the model loss until the training end condition is reached.
2. The method according to claim 1, It is characterized in that The method of adjusting the model parameters in the advertisement conversion parameter prediction model for predicting the display position of the advertisement sample data as the target display position according to the model loss includes: Masking the target model parameters in the advertisement conversion parameter prediction model, wherein the target model parameters are other model parameters except the model parameters used in the classification layer of the parameter prediction model to predict the target display position corresponding to the advertisement sample data; Based on the model loss, the model parameters that are not masked in the advertising conversion parameter prediction model are adjusted.
3. The method according to claim 1, It is characterized in that The parameter prediction model includes a first sub-model and a second sub-model. The parameter prediction model is used to perform classification prediction based on a plurality of feature vectors included in the advertisement sample data and the importance weight of each feature vector to obtain advertisement conversion parameters corresponding to each display position of the advertisement sample data, including: For each feature vector, multiply the feature vector by the importance weight corresponding to the feature vector to obtain the target feature corresponding to each feature vector; Using the first sub-model to perform classification prediction based on multiple target features corresponding to the advertisement sample data, to obtain first advertisement conversion parameters corresponding to each display position of the advertisement sample data; Using the second sub-model to perform classification prediction based on multiple target features corresponding to the advertisement sample data, to obtain second advertisement conversion parameters corresponding to each display position of the advertisement sample data; The calculation of the model loss based on the advertisement conversion label corresponding to the display position of the advertisement sample data and the advertisement conversion parameter corresponding to each display position of the advertisement sample data includes: Calculate the model loss based on the advertisement conversion label of the display position corresponding to the advertisement sample data, the first advertisement conversion parameter of each display position corresponding to the advertisement sample data, and the second advertisement conversion parameter of each display position corresponding to the advertisement sample data; The method of adjusting the model parameters in the advertisement conversion parameter prediction model for predicting the advertisement sample data corresponding to the target display position according to the model loss includes: The model parameters of the weight prediction model are adjusted according to the model loss, the model parameters of the first sub-model used to predict the advertisement sample data corresponding to the target display position are adjusted, and the model parameters of the second sub-model are adjusted.
4. The method according to claim 3, It is characterized in that The first sub-model is a deep feature cross-network model, and the second sub-model is a linear regression model or a machine learning model based on factor decomposition.
5. The method according to claim 3, It is characterized in that The parameter prediction model further includes a third sub-model, wherein the parameter prediction model is used to perform classification prediction based on a plurality of feature vectors included in the advertisement sample data and the importance weights of each feature vector to obtain advertisement conversion parameters corresponding to each display position of the advertisement sample data, and further includes: Using the third sub-model to perform classification prediction based on multiple target features corresponding to the advertising sample data, obtaining a third advertising conversion parameter corresponding to each display position of the advertising sample data, wherein the second sub-model is a linear regression model, and the third sub-model is a machine learning model based on factor decomposition; The calculation of the model loss based on the first advertisement conversion label of the display position corresponding to the advertisement sample data, the advertisement conversion parameter of each display position corresponding to the advertisement sample data, and the second advertisement conversion parameter of each display position corresponding to the advertisement sample data includes: The model loss is calculated based on the advertisement conversion label corresponding to the display position of the advertisement sample data, the first advertisement conversion parameter corresponding to the advertisement sample data for each display position, the second advertisement conversion parameter corresponding to the advertisement sample data for each display position, and the third advertisement conversion parameter corresponding to the advertisement sample data for each display position.
6. The method according to any one of claims 1 to 5, It is characterized in that The method of adjusting the model parameters in the advertisement conversion parameter prediction model for predicting the display position of the advertisement sample data as the target display position according to the model loss includes: If the advertisement type in the advertisement sample data is the target advertisement type, based on the model loss, adjusting the model parameters in the advertisement conversion parameter prediction model for predicting the display position of the advertisement sample data to be the target display position; If the advertisement type in the advertisement sample data is not the target advertisement type, at least one model parameter of the advertisement conversion parameter prediction model is adjusted based on the model loss.
7. A method for predicting advertising conversion parameters, It is characterized in that The method further comprises: Acquire user-side features of a target user and advertisement-side features of a plurality of advertisement data to be published; Using the advertisement conversion parameter prediction model obtained as claimed in any one of claims 1 to 6, based on the user-side features and the advertisement-side features of each of the advertisement data to be published, the advertisement conversion parameters of each display position corresponding to the advertisement data to be published are obtained; Based on the advertisement conversion parameters of each advertisement data to be published corresponding to each display position, target advertisement data and the display position corresponding to the target advertisement data are determined from the plurality of advertisement data to be published.
8. The method according to claim 7, It is characterized in that The step of obtaining the user-side characteristics of the target user and the advertisement-side characteristics of the plurality of advertisement data to be published includes: Obtaining a content request, wherein the content request carries a user ID of a target user and a search keyword; Acquire user-side features of the target user according to the content request; A plurality of advertisement data to be published are acquired according to the search keyword, and advertisement-side features of each advertisement data to be published are extracted.
9. The method according to claim 7, It is characterized in that The step of determining target advertisement data and the display position corresponding to the target advertisement data from the plurality of advertisement data to be published based on advertisement conversion parameters corresponding to the display positions of the advertisements to be published comprises: Based on the advertising conversion parameters corresponding to each display position of each advertising data to be published, the advertising data to be published corresponding to the advertising conversion parameter with the largest value is selected as the target advertising data, and the display position corresponding to the advertising conversion parameter with the largest value in the target advertising data is used as the display position corresponding to the target advertising data.
10. An advertisement conversion parameter prediction model training device, It is characterized in that The device comprises: A data acquisition module, used to acquire advertisement sample data, wherein the advertisement sample data carries an advertisement conversion tag corresponding to a target display position of the advertisement sample data; A data conversion module, used for performing vector conversion on the advertisement sample data to obtain a plurality of feature vectors included in the advertisement sample data; A weight prediction module, used to obtain the importance weight of each feature vector using the weight prediction model; A parameter prediction module, configured to use the parameter prediction model to perform classification prediction based on a plurality of feature vectors included in the advertisement sample data and the importance weight of each feature vector, so as to obtain advertisement conversion parameters corresponding to each display position of the advertisement sample data; A loss acquisition module, used to calculate the model loss based on the advertisement conversion label corresponding to the display position of the advertisement sample data and the advertisement conversion parameter corresponding to each display position of the advertisement sample data; The model training module is used to adjust the model parameters in the advertisement conversion parameter prediction model used to predict the advertisement sample data corresponding to the target display position according to the model loss until the training end condition is reached.
11. An advertisement conversion parameter prediction device, It is characterized in that The device comprises: A feature acquisition module, used to acquire user-side features of a target user and advertisement-side features of a plurality of advertisement data to be published; An information prediction module, configured to obtain the advertisement conversion parameters of each display position corresponding to the advertisement data to be published, based on the user-side features and the advertisement-side features of each advertisement data to be published, using the advertisement conversion parameter prediction model obtained as claimed in claim 10; The information determination module is used to determine target advertisement data and the display position corresponding to the target advertisement data from the plurality of advertisement data to be published based on advertisement conversion parameters corresponding to each display position of each advertisement data to be published.
12. An electronic device, It is characterized in that include: one or more processors; Memory; One or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, and the one or more programs are configured to execute the method according to any one of claims 1-6 or claims 7-9.
13. A computer-readable storage medium, It is characterized in that The computer-readable storage medium stores program codes, and the program codes can be called by a processor to execute the method according to any one of claims 1 to 6 or claims 7 to 9.
14. A computer program product comprising a computer program / instructions, It is characterized in that When the computer program / instructions are executed by a processor, the steps of the method described in any one of claims 1 to 6 or claims 7 to 9 are implemented.
Citation Information
Cited By
Model training method, advertisement material prediction method and device
CN120146932A
