Data Model Processing Method, Apparatus, Device, and Storage Medium

Abstract

The present application discloses a data model processing method, apparatus, device, and storage medium, relating to the technical field of data processing. The present application includes obtaining advertisement placement sample data, processing the advertisement placement sample data to obtain training data, determining the number of current GPUs, determining the number of groups of the training data according to the number of GPUs, grouping the training data according to the number of groups to obtain training grouped data, determining an initial placement model according to the advertisement placement sample data, determining the product correlation degree based on the initial placement model according to the advertisement placement sample data, performing parallel training on the initial placement model according to the product correlation degree to obtain target placement parameters, obtaining a target placement model according to the target placement parameters, and performing advertisement placement optimization based on the target placement model to obtain an advertisement placement plan, realizing data model processing by performing parallel training on user sample data, quickly learning the interests of users, updating model parameters, and obtaining a target placement model.

Description

Technical Field

[0001] This application relates to the technical field of data processing, and particularly relates to a method, apparatus, device, and storage medium for data model processing. Background Art

[0002] Advertising placement is a marketing method that can promote products to users and increase the popularity and influence of products. However, the usual advertising placement does not target the interests and hobbies of the placement objects, resulting in the inability of the placed advertisements to attract the attention of target users and even causing users' aversion, thereby increasing the marketing cost. To solve this problem, the currently used method is to only push advertisements based on the usage habits of target users and cannot adaptively adjust the placement strategy. Therefore, there will be a long-term push of advertisement information of the same product to target users, which further limits the product content in the advertisement push to target users and reduces the advertisement push effect of the product. Therefore, the current advertising placement method has the technical problem of being unable to adaptively adjust according to the interests of target users and having a poor placement effect.

[0003] The above content is only used to assist in understanding the technical solution of this application and does not represent an admission that the above content is prior art. Summary of the Invention

[0004] The main purpose of this application is to provide a method, apparatus, device, and storage medium for data model processing, aiming to solve the technical problem that the prior art cannot adaptively change the advertising placement strategy according to user interests, resulting in a reduction in the advertising placement effect.

[0005] To achieve the above purpose, this application proposes a data model processing method, and the data model processing method includes:

[0006] Obtain advertising placement sample data, and perform data processing on the advertising placement sample data to obtain training data;

[0007] Determine the number of current GPUs, determine the number of groups of the training data according to the number of GPUs, and group the training data according to the number of groups to obtain training grouped data;

[0008] Determine an initial placement model according to the advertising placement sample data, determine the product relevance according to the advertising placement sample data based on the initial placement model, and perform parallel training on the initial placement model according to the product relevance to obtain target placement parameters;

[0009] Obtain a target placement model according to the target placement parameters, and perform advertising placement optimization based on the target placement model to obtain an advertising placement plan.

[0010] In one embodiment, the steps of determining an initial placement model according to the advertisement placement sample data, determining a product relevance based on the initial placement model according to the advertisement placement sample data, and performing parallel training on the initial placement model according to the product relevance to obtain target placement parameters include:

[0011] Determine a product placement type, target user tags, average interest degree, and maximum interest degree according to the advertisement placement sample data, and match an initial placement model according to the product placement type;

[0012] Determine the target user tags according to the training grouping data, and determine a product relevance according to the target user tags and the product placement type;

[0013] Perform parallel training on the initial placement model according to the product relevance, the average interest degree, and the maximum interest degree to obtain target placement parameters.

[0014] In one embodiment, the steps of determining the target user tags according to the training grouping data and determining a product relevance according to the target user tags and the product placement type include:

[0015] Determine the interest circles of the associated users according to the product placement type and the target user tags;

[0016] Obtain the associated user tags of the associated users according to the interest circles of the associated users;

[0017] Obtain a product relevance based on the associated user tags, target user tags, and the product placement type.

[0018] In one embodiment, the steps of performing parallel training on the initial placement model according to the product relevance, the average interest degree, and the maximum interest degree to obtain target placement parameters include:

[0019] Obtain the number of initial placement models according to the number of groupings, and respectively deploy the initial placement models on GPUs, where each GPU corresponds to one initial placement model;

[0020] Input the relevance and the training grouping data into the corresponding initial placement models respectively to obtain initial model parameters;

[0021] Determine the fitness of each of the corresponding initial placement models according to the initial parameters;

[0022] Sort the fitness of each initial placement model to obtain the maximum fitness;

[0023] Obtain a maximum fitness threshold based on the average interest degree and the maximum interest degree;

[0024] When the maximum fitness is equal to the maximum fitness threshold, determine the initial parameter corresponding to the maximum fitness as the target delivery parameter.

[0025] In one embodiment, the step of sorting the fitness degrees of the respective initial delivery models to obtain the maximum fitness degree includes:

[0026] Summarize the fitness degrees of the respective initial delivery models and calculate the gradient value of the fitness degree;

[0027] Determine the error value of the initial delivery model according to the gradient value. When the error value is greater than a preset error value, update the initial delivery model corresponding to the gradient value to obtain an updated fitness degree;

[0028] Sort the updated fitness degree and the fitness degrees of the respective initial delivery models to obtain the maximum fitness degree.

[0029] In one embodiment, the step of obtaining advertisement delivery sample data and performing data processing on the advertisement delivery sample data to obtain training data includes:

[0030] Obtain advertisement delivery sample data and extract the feature fields of the advertisement delivery sample data;

[0031] Determine the duplicate data in the advertisement delivery sample data according to the feature fields, remove the duplicate data, and perform normalization processing on the data after duplicate removal to obtain training data.

[0032] In one embodiment, after the step of obtaining a target delivery model according to the target delivery parameter and performing advertisement delivery optimization based on the target delivery model to obtain an advertisement delivery plan includes:

[0033] Detect the output data of the target delivery model. When abnormal data is detected, detect the abnormal data;

[0034] When the abnormal data is preset abnormal data, perform data processing on the abnormal data and the advertisement delivery sample data to obtain training data;

[0035] Return to the step of determining the number of current GPUs, determining the number of groups of the training data according to the number of GPUs, and grouping the training data according to the number of groups to obtain training grouped data.

[0036] In addition, to achieve the above object, the present application also proposes a data model processing device, and the data model processing device includes:

[0037] A data preparation module, configured to obtain advertisement placement sample data and perform data processing on the advertisement placement sample data to obtain training data;

[0038] A data loading module, configured to determine the number of current GPUs, determine the number of groups of the training data according to the number of the GPUs, and group the training data according to the number of groups to obtain grouped training data;

[0039] A parameter fitting module, configured to determine an initial placement model according to the advertisement placement sample data, determine a product correlation degree based on the initial placement model according to the advertisement placement sample data, and perform parallel training on the initial placement model according to the product correlation degree to obtain target placement parameters;

[0040] A model generation module, configured to obtain a target placement model according to the target placement parameters, and perform advertisement placement optimization based on the target placement model to obtain an advertisement placement plan.

[0041] In addition, to achieve the above object, the present application further provides a data model processing device, where the device includes: a memory, a processor, and a computer program stored on the memory and executable on the processor, and the computer program is configured to implement the steps of the data model processing method as described above.

[0042] In addition, to achieve the above object, the present application further provides a storage medium, where the storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium, and when the computer program is executed by a processor, the steps of the data model processing method as described above are implemented.

[0043] In addition, to achieve the above object, the present application further provides a computer program product, where the computer program product includes a computer program, and when the computer program is executed by a processor, the steps of the data model processing method as described above are implemented.

[0044] One or more technical solutions proposed in this application have at least the following technical effects: obtaining advertising placement sample data, processing the advertising placement sample data to obtain training data, determining the number of current GPUs, determining the number of groups for the training data according to the number of GPUs, grouping the training data according to the number of groups to obtain training grouped data, determining an initial placement model according to the advertising placement sample data, determining the product correlation degree according to the advertising placement sample data based on the initial placement model, performing parallel training on the initial placement model according to the product correlation degree to obtain target placement parameters, obtaining a target placement model according to the target placement parameters, and optimizing the advertising placement based on the target placement model to obtain an advertising placement plan, so as to realize parallel training on user sample data, quickly learn the interests of users, update model parameters, and perform data model processing on the obtained target placement model. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] The accompanying drawings herein are incorporated into and constitute a part of this specification, showing embodiments consistent with this application and, together with the specification, are used to explain the principles of this application.

[0046] In order to more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the following will briefly introduce the accompanying drawings required for use in the description of the embodiments or the prior art. Obviously, for those of ordinary skill in the art, other drawings can also be obtained based on these drawings without creative efforts.

[0047] Figure 1 It is a schematic flowchart provided for the first embodiment of the data model processing method of this application;

[0048] Figure 2 It is a schematic diagram of the training of the initial placement model provided for the first embodiment of the data model processing method of this application;

[0049] Figure 3 It is a schematic block diagram of the module structure of the data model processing device according to the embodiment of this application;

[0050] Figure 4 It is a schematic diagram of the device structure of the hardware operating environment involved in the data model processing method according to the embodiment of this application.

[0051] The implementation, functional features, and advantages of the purpose of this application will be further described in conjunction with the embodiments with reference to the accompanying drawings. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0052] It should be understood that the specific embodiments described herein are only used to explain the technical solutions of this application and are not used to limit this application.

[0053] In order to better understand the technical solutions of this application, the following will be described in detail in conjunction with the accompanying drawings of the specification and the specific embodiments.

[0054] The main solution of the embodiment of the present application is as follows: Obtain advertising placement sample data, perform data processing on the advertising placement sample data to obtain training data, determine the number of current GPUs, determine the number of groups of the training data according to the number of GPUs, group the training data according to the number of groups to obtain training grouped data, determine an initial placement model according to the advertising placement sample data, determine the product relevance based on the initial placement model according to the advertising placement sample data, perform parallel training on the initial placement model according to the product relevance to obtain target placement parameters, obtain a target placement model according to the target placement parameters, and perform advertising placement optimization based on the target placement model to obtain an advertising placement plan.

[0055] In this embodiment, for the convenience of description, the following will be described with the data model processing device as the execution subject.

[0056] Since the prior art only performs advertisement pushing based on the usage habits of target users and cannot adaptively adjust the placement strategy, the advertisement information of the same product will be pushed to the target users for a long time. As a result, the product content in the advertisement pushing to the target users is limited, reducing the advertisement pushing effect of the product. Therefore, the current advertisement placement method has the technical problem of being unable to adaptively target the interests of target users and having a poor placement effect.

[0057] The present application provides a solution that enables advertisement placement to quickly learn the interests of users, update model parameters, and obtain a target placement model for data model processing through parallel training of user sample data, thereby achieving the purpose of accurate advertisement placement.

[0058] As can be seen from the above embodiment, the present application obtains advertising placement sample data, performs data processing on the advertising placement sample data to obtain training data, determines the number of current GPUs, determines the number of groups of the training data according to the number of GPUs, groups the training data according to the number of groups to obtain training grouped data, determines an initial placement model according to the advertising placement sample data, determines the product relevance based on the initial placement model according to the advertising placement sample data, performs parallel training on the initial placement model according to the product relevance to obtain target placement parameters, obtains a target placement model according to the target placement parameters, and performs advertising placement optimization based on the target placement model to obtain an advertising placement plan, thereby achieving parallel training of user sample data, quickly learning the interests of users, updating model parameters, and obtaining a target placement model for data model processing.

[0059] It should be noted that the execution subject of this embodiment can be a computing service device with data processing, network communication, and program running functions, such as a tablet computer, a personal computer, a mobile phone, etc., or an electronic device, a data model processing device, etc. that can implement the above functions. Hereinafter, the data model processing device will be taken as an example to illustrate this embodiment and the following embodiments.

[0060] Based on this, the embodiment of the present application provides a data model processing method. Refer to Figure 1 , Figure 1 which is a schematic flowchart of the first embodiment of the data model processing method of the present application.

[0061] In this embodiment, the data model processing method includes steps S10 to S40:

[0062] Step S10, obtain advertising placement sample data, and perform data processing on the advertising placement sample data to obtain training data.

[0063] It should be noted that the advertising placement sample data refers to the user tags including the target population of the avatar object, the product type, and the statistical data of the user's interest in the advertising placement obtained by sorting out the feedback data of the advertising placement in previous advertising placements.

[0064] In a specific implementation, when processing the advertisement placement sample data, it is possible to perform data processing on the advertisement placement sample data, such as data screening and filtering operations on the advertisement placement sample data. When processing the advertisement placement sample data, it is possible to obtain the advertisement placement sample data and extract the feature fields of the advertisement placement sample data; determine the duplicate data in the advertisement placement sample data according to the feature fields, remove the duplicates from the duplicate data, and perform normalization processing on the data after duplicate removal to obtain training data. Specifically, by obtaining the advertisement placement sample data, the advertisement placement sample data is obtained through statistics and is saved and transmitted in the form of a database file, which specifically includes information such as the type of the placed product, the role label of the target user, the degree of interest of the target object in the current product, the average degree of interest of the user group in the current product, and the maximum degree of interest of the user group in the current product. By extracting the feature fields in the advertisement placement sample data, for example, in this embodiment, the ID information of the user can be extracted as a feature field for data processing. At this time, it is possible to screen for duplicate values in the advertisement placement data according to the ID information of the user, use the ID information of the user as the primary key value, sort and / or traverse the advertisement placement sample data, determine the duplicate values in the user ID information, and delete the advertisement placement sample data corresponding to the duplicate ID information to remove the duplicate values from the data. After removing the duplicates from the data, in order to facilitate the model's processing of the data and reduce the size of the data volume, it is possible to normalize the data after duplicate removal to unify the dimension of the data after duplicate removal to obtain training data.

[0065] Step S20: Determine the number of current GPUs, determine the number of groups for the training data according to the number of GPUs, and group the training data according to the number of groups to obtain training grouped data.

[0066] It should be noted that GPU stands for Graphics Processing Unit, which is a highly parallel processor that can execute a large number of calculations simultaneously, and its data processing speed is more efficient compared to the central processing unit CPU.

[0067] In a specific implementation, the hardware information of the current data model processing device is read to obtain the number of GPUs in the current device, and the number of groups for training data is determined according to the number of GPUs. For example, when the number of GPUs in this embodiment is 4, the current training data can be evenly distributed according to the data volume or the number of data volume entries to obtain 4 groups of training grouped data with equal or similar numbers of data volume entries or data volumes. At the same time, the current 4 GPUs can be respectively labeled as GPU1, GPU2, GPU3, and GPU4, and the corresponding 4 grouped training data are labeled as Data1, Data2, Data3, and Data4. And Data1 is respectively loaded into GPU1, Data2 is loaded into GPU2, and according to a similar loading method, the grouped training data are respectively loaded into the corresponding GPUs. When grouping the training data, the original sample data set Data = {X 1 , X 2 ,..., X n} of the advertising placement sample data can be defined as a set of M independently generated Bootstrap sample sets, denoted as BSP(Data) = {Data1, Data2,..., Datan}. At this time, the obtained data set can be considered as a large data set, denoted as an HDFS file directory BSP-HDFS / .

[0068] Step S30: Determine an initial placement model according to the advertising placement sample data, and determine the product association degree based on the initial placement model according to the advertising placement sample data. Parallel training is performed on the initial placement model according to the product association degree to obtain target placement parameters.

[0069] It should be noted that the initial placement model is a neural network model, which includes a model input layer, a hidden layer, a model output layer, a loss function, and an optimizer, and is used to train the initial placement model according to the advertising placement sample data and optimize the parameters in the initial placement model.

[0070] In a specific implementation, when training the initial placement model, the initial placement model can be copied and loaded into each GPU respectively, referring to Figure 2 , Figure 2 which is a schematic diagram for training the initial placement model. In Figure 2 , the obtained data set Data is divided according to the number of GPUs, and the grouped training data is loaded. The loaded grouped training data is sent to the input layer of the model, the data is calculated through the hidden layer, the product association degree is output from the model output layer, and the product association degree is trained through the loss function and the optimizer to obtain the target placement parameters. A target placement parameter will be generated in each GPU.

[0071] Exemplarily, the steps of determining an initial placement model according to the advertisement placement sample data, determining a product relevance based on the initial placement model according to the advertisement placement sample data, and performing parallel training on the initial placement model according to the product relevance to obtain target placement parameters include:

[0072] Determine a product placement type, target user tags, average interest degree, and maximum interest degree according to the advertisement placement sample data, and match an initial placement model according to the product placement type;

[0073] Determine the target user tags according to the training grouping data, and determine a product relevance according to the target user tags and the product placement type;

[0074] Perform parallel training on the initial placement model according to the product relevance, the average interest degree, and the maximum interest degree to obtain target placement parameters.

[0075] In a specific implementation, interpret the advertisement placement sample data to determine a product placement type, target user tags, average interest degree, and maximum interest degree. At this time, it is possible to determine the matching between the product type to be placed and the target user tags, and it is possible to determine the relevance between the product type to be placed and the user tags, and determine the matching degree between the currently placed product type and the user. The higher the matching degree, the higher the current product relevance. Before training, it is necessary to match an initial placement model according to the placement type of the product. There are differences in the model parameters of the initial placement models corresponding to different placement types. And according to the product relevance, average interest degree, and maximum interest degree, where the average interest degree is the average of the interest degrees in all sample data, and the maximum interest degree is the maximum interest degree in the sample data, representing the maximum tolerance level. Perform parallel training on the initial placement model with the obtained product relevance, average interest degree, and maximum interest degree to obtain target placement parameters.

[0076] Exemplarily, the steps of determining the target user tags according to the training grouping data and determining a product relevance according to the target user tags and the product placement type include:

[0077] Determine the interest circle of the associated users according to the product placement type and the target user tags;

[0078] Obtain the associated user tags of the associated users according to the interest circle of the associated users;

[0079] Obtain a product relevance based on the associated user tags, target user tags, and the product placement type.

[0080] In a specific implementation, after determining the product placement type and target user tags, an interest circle can be determined based on the product placement type and target user tags, where the interest circle refers to a collection of tags corresponding to the user tags. It can be understood that a target user can have multiple user tags, and correspondingly, a person can have multiple interest circles, where all people in an interest circle have the same identity tag. Suppose there are currently 3 target users, the tags of user A are (photography, travel, food...), the tags of user B are (fitness, music, photography...), and the tags of user C are (music, photography, e-sports...). At this time, it can be obtained that user A corresponds to interest circles such as the photography interest circle, the travel interest circle, and the food interest circle. From another perspective, in the photography interest circle, there are user A, user B, and user C. At this time, determining the interest circle of the associated user according to the product placement type and the target user tags can be understood as determining the interest circle where the current user is located based on the user's tags, determining the identity tags of other users in each interest circle, and determining the interest circle corresponding to the identity tags of the obtained other users as the interest circle of the associated user. Still taking the above example, for user A, his interest circles are photography, travel, and food, and user B who is in the same photography interest circle as user A has fitness and music in addition to photography. At this time, user A can be associated, and it can be determined that user A may also be interested in fitness and music. Therefore, the associated user tags of user A can be determined as fitness and music. At this time, advertising can be carried out based on fitness and music, and the placement ratio can be determined. The specific implementation method is realized through a knowledge graph, and then the product correlation degree is obtained based on the associated user tags, target user tags, and product placement type.

[0081] Exemplarily, the step of parallel training of the initial placement model according to the product correlation degree, the average interest degree, and the maximum interest degree to obtain the target placement parameter includes:

[0082] Obtain the number of initial placement models according to the number of groups, and respectively deploy the initial placement models on the GPU, where each GPU corresponds to one initial placement model;

[0083] Respectively input the correlation degree and the training grouped data into the corresponding initial placement model to obtain the initial model parameters;

[0084] Determine the fitting degree of each corresponding initial placement model according to the initial parameters;

[0085] Sort the fitting degrees of the initial placement models to obtain the maximum fitting degree;

[0086] Obtain the maximum fitting degree threshold according to the average interest degree and the maximum interest degree;

[0087] When the maximum fitness is equal to the maximum fitness threshold, determine the initial parameters corresponding to the maximum fitness as the target delivery parameters.

[0088] In a specific implementation, during training, since the grouped training data has been loaded into the corresponding GPUs, at this time, the initial delivery model also needs to be copied and loaded into each GPU according to the number of groups, where each GPU corresponds to an initial delivery model. The correlation degree and the training grouped data are respectively input into the corresponding initial delivery model to obtain the initial model parameters. Through the ReLU activation function f(x) = max(0, x). Calculate the fitness according to the model training formula, and the formula is:

[0089]

[0090] where f(t) is the fitness, α is the correlation degree, x i is the sample data, i is the i-th sample data, and f(x) is the activation function value.

[0091] When calculating the fitness, it can be based on:

[0092]

[0093] where M is the number of categories, y o,c is the one-hot encoding of the true label, and p o,c is the predicted value of the model.

[0094] When the fitness is greater than 0, it indicates that the current delivery intensity is too large and the delivery intensity needs to be reduced. At this time, the fitness value can be reduced. Correspondingly, when the fitness is less than 0, the fitness value needs to be increased.

[0095] After obtaining the fitness of all the initial delivery models, the fitness values of each model can be sorted to obtain the maximum fitness value. At the same time, the maximum fitness threshold is obtained according to the average interest degree and the maximum interest degree. When the maximum fitness is equal to the maximum fitness threshold, determine the initial parameters corresponding to the maximum fitness as the target delivery parameters.

[0096] Exemplarily, the step of sorting the fitness of the initial delivery models to obtain the maximum fitness includes:

[0097] Summarize the fitness of the initial delivery models and calculate the gradient value of the fitness;

[0098] Determine the error value of the initial delivery model according to the gradient value. When the error value is greater than the preset error value, update the initial delivery model corresponding to the gradient value to obtain the updated fitness;

[0099] Sort the fitness of the updated fitness and each initial placement model to obtain the maximum fitness.

[0100] In a specific implementation, summarize the fitness of each initial placement model, and obtain the gradient value according to the formula:

[0101]

[0102] Where L is the loss function, θ is the model parameter, and k is the number of parameters.

[0103] And determine the error value of the initial placement model according to the gradient value, where the error value is When the error value is greater than the preset error value, update the initial placement model corresponding to the gradient value to obtain the updated fitness. When the parameters are updated, it can be passed through Where ω is the learning rate. Sort the fitness of the updated fitness and each initial placement model to obtain the maximum fitness.

[0104] Step S40: Obtain a target placement model according to the target placement parameters, and optimize the advertisement placement based on the target placement model to obtain an advertisement placement plan.

[0105] In a specific implementation, after obtaining the target placement model according to the target placement parameters, it is possible to optimize the advertisement placement of the target placement model and output the corresponding advertisement placement plan. After obtaining the advertisement placement plan, it is also possible to detect the output data of the target placement model. When abnormal data is detected, detect the abnormal data. When the abnormal data is preset abnormal data, perform data processing on the abnormal data and the advertisement placement sample data to obtain training data, return to the step of determining the number of current GPUs, determine the number of groups of the training data according to the number of GPUs, and group the training data according to the number of groups to obtain the training grouped data, and retrain the model.

[0106] This embodiment provides a data model processing method. By obtaining advertising placement sample data, performing data processing on the advertising placement sample data to obtain training data, determining the number of current GPUs, determining the number of groups for the training data based on the number of GPUs, grouping the training data according to the number of groups to obtain training grouped data, determining an initial placement model based on the advertising placement sample data, determining the product relevance based on the advertising placement sample data for the initial placement model, performing parallel training on the initial placement model according to the product relevance to obtain target placement parameters, obtaining a target placement model based on the target placement parameters, and performing advertising placement optimization based on the target placement model to obtain an advertising placement plan, it realizes data model processing by performing parallel training on user sample data, quickly learning the interests of users, updating model parameters, and obtaining a target placement model.

[0107] This application also provides a data model processing device. Please refer to Figure 3 , and the data model processing device includes:

[0108] A data preparation module 10, configured to obtain advertising placement sample data and perform data processing on the advertising placement sample data to obtain training data;

[0109] A data loading module 20, configured to determine the number of current GPUs, determine the number of groups for the training data based on the number of GPUs, and group the training data according to the number of groups to obtain training grouped data;

[0110] A parameter fitting module 30, configured to determine an initial placement model based on the advertising placement sample data, determine the product relevance based on the advertising placement sample data for the initial placement model, and perform parallel training on the initial placement model according to the product relevance to obtain target placement parameters;

[0111] A model generation module 40, configured to obtain a target placement model based on the target placement parameters and perform advertising placement optimization based on the target placement model to obtain an advertising placement plan.

[0112] The data model processing device provided by this application adopts the data model processing method in the above embodiment, and can solve the technical problem that the prior art cannot adaptively change the advertising placement strategy according to user interests, resulting in a decrease in advertising placement effect. Compared with the prior art, the beneficial effects of the data model processing device provided by this application are the same as those of the data model processing method provided by the above embodiment, and other technical features in the data model processing device are the same as the features disclosed in the method of the above embodiment, and will not be elaborated here.

[0113] This application provides a data model processing device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein, the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the data model processing method in the first embodiment above.

[0114] Refer to the following Figure 4 , which shows a schematic structural diagram of a data model processing device suitable for implementing the embodiments of this application. The data model processing device in the embodiments of this application may include, but is not limited to, mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (Personal Digital Assistant), PADs (Portable Application Description: tablet computers), PMPs (Portable Media Player: portable multimedia players), in-vehicle terminals (such as in-vehicle navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc. Figure 4 The data model processing device shown is only an example and should not impose any limitation on the functions and usage scope of the embodiments of this application.

[0115] As Figure 4As shown, the data model processing device may include a processing device 1001 (such as a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM: Read Only Memory) 1002 or a program loaded from a storage device 1003 into a random access memory (RAM: Random Access Memory) 1004. In the RAM 1004, various programs and data required for the operation of the data model processing device are also stored. The processing device 1001, the ROM 1002, and the RAM 1004 are connected to each other through a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Generally, the following systems can be connected to the I / O interface 1006: an input device 1007 including, for example, a touch screen, a touchpad, a keyboard, a mouse, an image sensor, a microphone, an accelerometer, a gyroscope, etc.; an output device 1008 including, for example, a liquid crystal display (LCD: Liquid Crystal Display), a speaker, a vibrator, etc.; a storage device 1003 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 1009. The communication device 1009 can allow the data model processing device to communicate with other devices wirelessly or wiredly to exchange data. Although the figure shows a data model processing device having various systems, it should be understood that it is not required to implement or have all the systems shown. Instead, more or fewer systems can be implemented or had.

[0116] Particularly, according to the embodiments disclosed in the present application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, the embodiments disclosed in the present application include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program contains program codes for executing the methods shown in the flowcharts. In such an embodiment, the computer program can be downloaded and installed from a network through the communication device, or installed from the storage device 1003, or installed from the ROM 1002. When the computer program is executed by the processing device 1001, the above functions defined in the methods of the embodiments disclosed in the present application are executed.

[0117] The data model processing device provided by the present application adopts the data model processing method in the above embodiments, and can solve the technical problem that the prior art cannot adaptively change the advertising delivery strategy according to the user's interest, resulting in a reduction in the advertising delivery effect. Compared with the prior art, the beneficial effects of the data model processing device provided by the present application are the same as those of the data model processing method provided by the above embodiments, and other technical features in the data model processing device are the same as those disclosed in the method of the previous embodiment, and will not be elaborated here.

[0118] It should be understood that each part disclosed in this application can be implemented by hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in a suitable manner in any one or more embodiments or examples.

[0119] As described above, it is only the specific implementation manner of this application, but the protection scope of this application is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed in this application, and all should be covered by the protection scope of this application. Therefore, the protection scope of this application should be subject to the protection scope of the claims.

[0120] This application provides a computer-readable storage medium with computer-readable program instructions (i.e., computer programs) stored thereon, and the computer-readable program instructions are used to execute the data model processing method in the above embodiments.

[0121] The computer-readable storage medium provided by this application can be, for example, a USB flash drive, but is not limited to electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or any combination of the above. More specific examples of computer-readable storage media can include, but are not limited to: electrical connections with one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM) or flash memory, optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the above. In this embodiment, the computer-readable storage medium can be any tangible medium that contains or stores a program, and this program can be used by or combined with an instruction execution system, device, or device. The program code contained on the computer-readable storage medium can be transmitted by any appropriate medium, including but not limited to: wires, optical cables, RF (radio frequency), etc., or any suitable combination of the above.

[0122] The above computer-readable storage medium can be included in the data model processing device; it can also exist separately without being assembled into the data model processing device.

[0123] The above computer-readable storage medium carries one or more programs. When the above one or more programs are executed by the data model processing device, the data model processing device is caused to:

[0124] Obtain advertisement placement sample data, and perform data processing on the advertisement placement sample data to obtain training data;

[0125] Determine the number of current GPUs, determine the number of groups of the training data according to the number of the GPUs, and group the training data according to the number of groups to obtain training grouped data;

[0126] Determine an initial placement model according to the advertisement placement sample data, determine the product relevance according to the advertisement placement sample data based on the initial placement model, and perform parallel training on the initial placement model according to the product relevance to obtain target placement parameters;

[0127] Obtain a target placement model according to the target placement parameters, and perform advertisement placement optimization based on the target placement model to obtain an advertisement placement plan.

[0128] Computer program code for performing the operations of the present application can be written in one or more programming languages or combinations thereof. The above programming languages include object-oriented programming languages - such as Java, Smalltalk, C++, and also include conventional procedural programming languages - such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, executed as an independent software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer can be connected to the user's computer through any type of network - including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computer (for example, by using an Internet service provider to connect through the Internet).

[0129] The flowcharts and block diagrams in the accompanying drawings illustrate the possible architectures, functions, and operations of systems, methods, and computer program products according to various embodiments of the present application. In this regard, each block in the flowchart or block diagram may represent a module, a segment of a program, or a part of code that contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than that marked in the accompanying drawings. For example, two consecutive blocks shown may actually be executed substantially in parallel, and they may sometimes be executed in the reverse order, depending on the functions involved. It should also be noted that each block in the block diagram and / or flowchart, as well as combinations of blocks in the block diagram and / or flowchart, can be implemented by a dedicated hardware-based system that performs the specified functions or operations, or can be implemented by a combination of dedicated hardware and computer instructions.

[0130] The modules involved in the embodiments described in the present application can be implemented in software or in hardware. In some cases, the name of the module does not constitute a limitation on the unit itself.

[0131] The readable storage medium provided by the present application is a computer-readable storage medium, and the computer-readable storage medium stores computer-readable program instructions (i.e., computer programs) for executing the above data model processing method, which can solve the technical problem that the prior art cannot adaptively change the advertising placement strategy according to user interests, resulting in a decrease in the advertising placement effect. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided by the present application are the same as those of the data model processing method provided by the above embodiments, and will not be elaborated here.

[0132] The present application also provides a computer program product, including a computer program, and when the computer program is executed by a processor, it implements the steps of the data model processing method as described above.

[0133] The computer program product provided by the present application can solve the technical problem that the prior art cannot adaptively change the advertising placement strategy according to user interests, resulting in a decrease in the advertising placement effect. Compared with the prior art, the beneficial effects of the computer program product provided by the present application are the same as those of the data model processing method provided by the above embodiments, and will not be elaborated here.

[0134] The above are only some embodiments of the present application, and thus do not limit the patent scope of the present application. Any equivalent structural transformation made under the technical concept of the present application by using the content of the specification and drawings of the present application, or any direct / indirect application in other related technical fields, is included in the patent protection scope of the present application.

Claims

1. A data model processing method, characterized in that: The data model processing method comprises: Acquire advertisement delivery sample data, and perform data processing on the advertisement delivery sample data to obtain training data; Determine the number of current GPUs, determine the number of groups of the training data according to the number of GPUs, and group the training data according to the number of groups to obtain training group data; Determine an initial delivery model according to the advertising delivery sample data, determine product relevance based on the initial delivery model according to the advertising delivery sample data, and perform parallel training on the initial delivery model according to the product relevance to obtain target delivery parameters; Obtaining a target delivery model according to the target delivery parameters, and performing data model processing based on the target delivery model to obtain an advertisement delivery plan; The steps of determining an initial delivery model according to the sample data of advertisement delivery, determining product relevance according to the sample data of advertisement delivery based on the initial delivery model, and training the initial delivery model in parallel according to the product relevance to obtain target delivery parameters include: Determine the product delivery type, target user label, average interest and maximum interest according to the advertisement delivery sample data, and match the initial delivery model according to the product delivery type; Determine the target user tag according to the training group data, and determine the product association degree according to the target user tag and the product delivery type; Parallel training of the initial delivery model is performed according to the product association, the average interest and the maximum interest to obtain target delivery parameters; The step of determining the target user tag according to the training group data, and determining the product association degree according to the target user tag and the product delivery type comprises: Determine the interest circle of the associated user according to the product delivery type and the target user tag; Obtaining an associated user tag of the associated user according to the associated user's interest circle; Obtaining product relevance based on the associated user tag, the target user tag and the product launch type; The step of training the initial delivery model in parallel according to the product association, the average interest and the maximum interest to obtain target delivery parameters comprises: Obtaining the number of the initial delivery models according to the number of groups, and deploying the initial delivery models on GPUs respectively, wherein each GPU corresponds to one initial delivery model; Inputting the correlation degree and the training group data into the corresponding initial delivery model respectively to obtain initial model parameters; The degree of fit of each of the corresponding initial delivery models is determined according to the initial model parameters, and the degree of fit calculation formula is: in, is the degree of fit, is the correlation, is the sample data, For the Sample data, is the activation function value, the correlation The calculation formula is , where M is the number of categories, is the one-hot encoding of the true label, is the predicted value of the model; Sort the fit of each initial launch model to obtain the maximum fit; Obtaining a maximum fitting degree threshold value according to the average interest degree and the maximum interest degree; When the maximum degree of fit is equal to the maximum degree of fit threshold, the initial model parameters corresponding to the maximum degree of fit are determined as target delivery parameters.

2. The method according to claim 1, characterized in that The step of sorting the fitness of each initial delivery model to obtain the maximum fitness comprises: Summarizing the fitness of each initial delivery model and calculating the gradient value of the fitness; Determining an error value of the initial delivery model according to the gradient value, and when the error value is greater than a preset error value, updating the initial delivery model corresponding to the gradient value to obtain an updated degree of fit; The updated fitness degree and the fitness degrees of each initial delivery model are sorted to obtain the maximum fitness degree.

3. The method according to claim 1, characterized in that The step of obtaining the sample data for advertisement delivery and performing data processing on the sample data for advertisement delivery to obtain training data includes: Acquire advertisement delivery sample data, and extract characteristic fields from the advertisement delivery sample data; The duplicate data in the advertisement delivery sample data is determined according to the characteristic field, the duplicate data is deduplicated, and the deduplicated data is normalized to obtain training data.

4. The method according to any one of claims 1 to 3, characterized in that The step of obtaining a target delivery model according to the target delivery parameters, and performing data model processing based on the target delivery model to obtain an advertisement delivery plan includes: Detecting the output data of the target delivery model, and when abnormal data is detected, detecting the abnormal data; When the abnormal data is preset abnormal data, data processing is performed on the abnormal data and the advertisement delivery sample data to obtain training data; Return to the step of determining the number of current GPUs, determining the number of groups of the training data according to the number of GPUs, and grouping the training data according to the number of groups to obtain training group data.

5. A data model processing device, characterized in that: The device comprises: A data preparation module is used to obtain sample data of advertisement delivery and process the sample data of advertisement delivery to obtain training data; A data loading module, used to determine the number of current GPUs, determine the number of groups of the training data according to the number of GPUs, and group the training data according to the number of groups to obtain training group data; A parameter fitting module, used to determine an initial delivery model according to the advertisement delivery sample data, and to perform parallel training on the initial delivery model according to the training group data to obtain target delivery parameters; A model generation module, used to obtain a target delivery model according to the target delivery parameters, and perform data model processing based on the target delivery model to obtain an advertisement delivery plan; The steps of determining an initial delivery model according to the sample data of advertisement delivery, determining product relevance according to the sample data of advertisement delivery based on the initial delivery model, and training the initial delivery model in parallel according to the product relevance to obtain target delivery parameters include: Determine the product delivery type, target user label, average interest and maximum interest according to the advertisement delivery sample data, and match the initial delivery model according to the product delivery type; Determine the target user tag according to the training group data, and determine the product association degree according to the target user tag and the product delivery type; Parallel training of the initial delivery model is performed according to the product association, the average interest and the maximum interest to obtain target delivery parameters; The step of determining the target user tag according to the training group data, and determining the product association degree according to the target user tag and the product delivery type comprises: Determine the interest circle of the associated user according to the product delivery type and the target user tag; Obtaining an associated user tag of the associated user according to the associated user's interest circle; Obtaining product relevance based on the associated user tag, the target user tag and the product launch type; The step of training the initial delivery model in parallel according to the product association, the average interest and the maximum interest to obtain target delivery parameters comprises: Obtaining the number of the initial delivery models according to the number of groups, and deploying the initial delivery models on GPUs respectively, wherein each GPU corresponds to one initial delivery model; Inputting the correlation degree and the training group data into the corresponding initial delivery model respectively to obtain initial model parameters; The degree of fit of each of the corresponding initial delivery models is determined according to the initial model parameters, and the degree of fit calculation formula is: in, is the degree of fit, is the correlation, is the sample data, For the Sample data, is the activation function value, the correlation The calculation formula is , where M is the number of categories, is the one-hot encoding of the true label, is the predicted value of the model; Sort the fit of each initial launch model to obtain the maximum fit; Obtaining a maximum fitting degree threshold value according to the average interest degree and the maximum interest degree; When the maximum degree of fit is equal to the maximum degree of fit threshold, the initial model parameters corresponding to the maximum degree of fit are determined as target delivery parameters.

6. A data model processing device, characterized in that: The device comprises: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program is configured to implement the steps of the data model processing method according to any one of claims 1 to 4.

7. A storage medium, characterized in that: The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, the steps of the data model processing method according to any one of claims 1 to 4 are implemented.

Images