CN113971582A - Method and system for generating advertisement putting plan, electronic device and storage medium - Google Patents

Method and system for generating advertisement putting plan, electronic device and storage medium Download PDF

Info

Publication number
CN113971582A
CN113971582A CN202110761489.4A CN202110761489A CN113971582A CN 113971582 A CN113971582 A CN 113971582A CN 202110761489 A CN202110761489 A CN 202110761489A CN 113971582 A CN113971582 A CN 113971582A
Authority
CN
China
Prior art keywords
network
advertisement putting
data
combination
sequence
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202110761489.4A
Other languages
Chinese (zh)
Inventor
张水佩
潘亮旭
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Beijing Minglue Zhaohui Technology Co Ltd
Original Assignee
Beijing Minglue Zhaohui Technology Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Beijing Minglue Zhaohui Technology Co Ltd filed Critical Beijing Minglue Zhaohui Technology Co Ltd
Priority to CN202110761489.4A priority Critical patent/CN113971582A/en
Publication of CN113971582A publication Critical patent/CN113971582A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/02—Marketing; Price estimation or determination; Fundraising
    • G06Q30/0241—Advertisements
    • G06Q30/0251—Targeted advertisements
    • G06Q30/0269—Targeted advertisements based on user profile or attribute
    • G06Q30/0271—Personalized advertisement
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/02—Marketing; Price estimation or determination; Fundraising
    • G06Q30/0241—Advertisements
    • G06Q30/0242—Determining effectiveness of advertisements
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/02—Marketing; Price estimation or determination; Fundraising
    • G06Q30/0241—Advertisements
    • G06Q30/0277—Online advertisement

Landscapes

  • Engineering & Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Strategic Management (AREA)
  • Finance (AREA)
  • Development Economics (AREA)
  • Accounting & Taxation (AREA)
  • General Physics & Mathematics (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Game Theory and Decision Science (AREA)
  • General Business, Economics & Management (AREA)
  • Marketing (AREA)
  • Economics (AREA)
  • Computational Linguistics (AREA)
  • Artificial Intelligence (AREA)
  • Software Systems (AREA)
  • Data Mining & Analysis (AREA)
  • General Health & Medical Sciences (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Mathematical Physics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Evolutionary Computation (AREA)
  • Health & Medical Sciences (AREA)
  • General Engineering & Computer Science (AREA)
  • Computing Systems (AREA)
  • Molecular Biology (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The embodiment of the application discloses a method, a system, an electronic device and a storage medium for generating an advertisement putting plan, which comprises the following steps: training according to the first historical advertisement putting combination sequence data, preset parameters and random vectors to generate a confrontation network time sequence model; inputting second historical advertisement putting combination sequence data and target parameters into the generated countermeasure network timing model to obtain a simulated target advertisement putting combination; and processing the simulated target advertisement putting combination to determine a target advertisement putting plan. The method can automatically generate a more scientific advertisement putting plan aiming at the condition of putting the advertisement across the platforms, replace the prior technical scheme of carrying out advertisement putting according to the thought experience, reduce the manpower consumption, improve the decision efficiency, be beneficial to more scientifically and reasonably allocating budget and optimize the specific advertisement putting condition every day in the future.

Description

Method and system for generating advertisement putting plan, electronic device and storage medium
Technical Field
The embodiment of the application relates to the technical field of advertisement putting, for example, to a method, a system, an electronic device and a storage medium for generating an advertisement putting plan.
Background
Due to the data barriers between information islands and platforms, advertising resource publishers have long faced the difficult problem of how to perform scientific budget allocation and cross-platform resource placement.
The related technology can only solve the problem of the delivery plan in a single platform, and the new delivery plan is planned by summarizing and analyzing the behavior rules of the user based on a statistical or machine learning model, so that accurate delivery can be realized in a data intercommunication platform. However, data barriers exist among different platforms, and aiming at cross-platform delivery, the existing scheme is stopped at the stage of effect evaluation and experience delivery, a general tuning direction is obtained according to an effect evaluation result, and delivery is carried out by matching with industry experience. Due to the dependence on human experience, the decision period is long, the delivery efficiency is low, and meanwhile, the specific day-to-day delivery scientific basis and the effect controllability are lacked, so that budget waste is caused.
Disclosure of Invention
The embodiment of the application provides an advertisement putting plan generating method, an advertisement putting plan generating system, electronic equipment and a storage medium, so that the condition that an advertisement putting plan cannot be reasonably formulated aiming at a cross-platform condition in the related technology is improved.
In a first aspect, an embodiment of the present application provides an advertisement delivery plan generating method, including:
training according to the first historical advertisement putting combination sequence data, preset parameters and random vectors to generate a countermeasure network timing model;
inputting second historical advertisement putting combination sequence data and target parameters into the generated countermeasure network timing model to obtain a simulated target advertisement putting combination;
and processing the simulated target advertisement putting combination to determine a target advertisement putting plan.
In a second aspect, an embodiment of the present application provides an advertisement placement plan generating system, including:
the system comprises a front-end module, a service processing module and a data processing module;
the front-end module is used for sending a request for generating a target advertisement putting plan and displaying the target advertisement putting plan;
the business processing module is used for receiving the target advertisement putting plan generating request, inputting training data to the data processing layer and generating a confrontation network time sequence model;
the data processing module is configured to receive the training data and the generated confrontation network time sequence model, obtain the target advertisement delivery plan, and return the target advertisement delivery plan to the service processing layer.
In a third aspect, an embodiment of the present application provides an electronic device, where the electronic device includes: one or more processors;
a storage device arranged to store one or more programs,
when the one or more programs are executed by the one or more processors, the one or more processors implement the advertisement delivery plan generating method according to any embodiment of the present application.
In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, on which a computer program is stored, where the computer program, when executed by a processor, implements an advertisement placement plan generating method according to any embodiment of the present application.
According to the embodiment of the invention, a confrontation network time sequence model is generated according to first historical advertisement putting combination sequence data, preset parameters and random vector training; inputting second historical advertisement putting combination sequence data and target parameters into the generated countermeasure network timing model to obtain a simulated target advertisement putting combination; and processing the simulated target advertisement putting combination to determine a target advertisement putting plan. The method can automatically generate a more scientific advertisement putting plan aiming at the condition of putting the advertisement across platforms, replaces the prior technical scheme of carrying out advertisement putting according to thought experience, can reduce manpower consumption, improves decision efficiency, is beneficial to more scientifically and reasonably distributing budget, and optimizes the specific advertisement putting condition every day in the future.
Drawings
Fig. 1 is a flowchart illustrating steps of a method for generating an advertisement placement plan according to an embodiment of the present application;
FIG. 2 is an architecture diagram of joint training for generating a countermeasure network timing model according to an embodiment of the present application;
FIG. 3 is a flow chart of a method for generating countermeasure network timing model processing data according to an embodiment of the present application;
FIG. 4 is a flow chart of a method for generating prediction model processing data included in a countermeasure network timing model verification module according to an embodiment of the present application;
fig. 5 is a block diagram illustrating an advertisement placement plan generating system according to an embodiment of the present application;
fig. 6 is a schematic structural diagram of an electronic device according to an embodiment of the present application.
Detailed Description
The present application will be described in detail with reference to the accompanying drawings and examples. It is to be understood that the exemplary embodiments described herein are for purposes of illustration only and are not limiting. It should be further noted that, for the convenience of description, only some of the structures related to the present application are shown in the drawings, not all of the structures. The embodiments and features of the embodiments in the present application may be combined with each other without conflict.
Fig. 1 is a flowchart of steps of an advertisement placement plan generating method according to an embodiment of the present application, where the embodiment of the present application is applicable to a case of generating an advertisement placement plan for a cross-platform case, and the method may be executed by an advertisement placement plan generating system according to an embodiment of the present application, where the advertisement placement plan generating system may be implemented by hardware or software and integrated in an electronic device according to an embodiment of the present application, and in an embodiment, as shown in fig. 1, the advertisement placement plan generating method according to an embodiment of the present application may include the following steps:
and S110, training and generating a confrontation network time sequence model according to the first historical advertisement putting combination sequence data, preset parameters and random vectors.
In this embodiment of the present application, the first historical advertisement delivery combination sequence data refers to historical data of a certain advertisement resource delivered in different platforms within a certain time period. The preset parameters refer to return data of the historical advertisement putting combined sequence data.
In one embodiment, training to generate a countermeasure network timing model according to the first historical advertisement delivery combination sequence data and preset parameters includes: training a characterization network in the generated confrontation network time sequence model according to the first historical advertisement putting combination sequence data and the preset parameters; training the countermeasure network in the countermeasure network timing model according to the random vector; training the supervising network according to the characterizing network and the countering network; wherein the characterization network, the countermeasure network, and the supervisory network make up the generative countermeasure network timing model.
In an embodiment, training the characterization network in the generated confrontation network timing model according to the first historical advertisement delivery combination sequence data and the preset parameters includes: determining a first placement characteristic (S, X) of the first historical advertisement placement portfolio sequence data1:T) And said first delivery feature (S, X) is delivered via an embedded network1:T) Mapping from original space to hidden space to obtain corresponding first hidden vector hS,h1:T＝e(s,x1:t) (ii) a Passing the first hidden vector h through a recovery networkS,h1:T＝e(s,x1:t) Second launch feature restored to the original space
Figure BDA0003150010180000031
Wherein the characterization network comprises the embedded network and the recovery network. For example, the embedded network maps the impression characteristics to a hidden space, making the competing network learn the potential temporal dynamics of the impression composition data through a low-dimensional representation. By HS,HXRepresenting the hidden vector space corresponding to the feature space S, X, embedding a function e to map the static and time sequence features to the hidden space hS,h1:T＝e(s,x1:t):hS＝eS(s)ht＝ ex(hS,ht-1,xt) Wherein e isSAnd eXAn embedded network of static and timing features, respectively. The recovery network then provides a mapping from the hidden space to the original space, r recovers the static and time-series hidden variables to their original characteristic representation:
Figure BDA0003150010180000041
we achieve this by a feed forward network:
Figure BDA0003150010180000042
there are many alternative structures for the embedding and recovery functions that implement autoregressive and the output of each step can only depend on the previous information.
In one embodiment, the countermeasure network includes the sequence generation network and the sequence discrimination network, and training the countermeasure network in the countermeasure network generation timing model according to the random vector includes: the sequence generation network generates a second hidden vector feature h in a hidden space according to the random vectorS,h1:T＝g(zS,z1:T) (ii) a The sequence discrimination network is used for the second implicit vector feature hS,h1:T＝g(zS,z1:T) Distinguishing to obtain sequence classification yS,y1:T＝d(hS,h1:T) (ii) a Wherein h isS＝gS(zS)，ht＝gX(hS,ht-1,zt)。
For example, generating a countermeasure network (GANs) includes a sequence generation network that takes as input a large number of random vectors generated with a known simple distribution and generates in hidden space H a sequence discrimination networkS,HXGenerating a second latent vector feature hS,h1:T＝ g(zS,z1:T)：hS＝gS(zS)，ht＝gX(hS,ht-1,zt). Wherein g isS:ZS→HSIs a static feature generator, gX:HS×HS×ZX→HXIs a timing feature cycle generator. Random vector zSCan be sampled from the distribution, ztA random process is followed. The sequence discrimination network is configured to discriminate a probability of authenticity of the second steganographic feature. Discriminating network d HS×∏tX→[0,1]×∏t[0,1]Accepting static and time-series characteristics, returning sequence classifications yS,y1:T＝d(hS,h1:T) The discriminator is implemented using a bi-directional loop network plus a feed forward output layer: y isS＝dS(hS)
Figure BDA0003150010180000043
Wherein
Figure BDA0003150010180000044
Figure BDA0003150010180000045
Representing a sequence of hidden states forward and backward respectively,
Figure BDA0003150010180000046
is the equation of circulation, dS,dXIs the output of the classification function. Also, the generator is autoregressive, and the architecture has a variety of options.
In one embodiment, the first hidden vector h is transformed by a recovery networkS,h1:T＝e(s,x1:t) Second launch feature restored to the original space
Figure BDA0003150010180000047
Then, the method further comprises the following steps: calculating reconstruction loss
Figure BDA0003150010180000048
Figure BDA0003150010180000049
In one embodiment, the sequence generation network is implemented by accepting artificial embedding hS,h1:t-1To generate the next htCalculating the gradient by unsupervised loss; classifying y according to said sequenceS,y1:T＝d(hS,h1:T) Calculating surveillant-free loss
Figure BDA00031500101800000410
Figure BDA00031500101800000411
According to the real data hS,h1:TAnd said calculated gradient, calculated supervised loss LS＝ Es,x1:T～p[∑t||ht-gX(hS,ht-1,zt)||2]。
And S120, inputting the second historical advertisement putting combination sequence data and the target parameters into the generation countermeasure network timing model to obtain a simulation target advertisement putting combination.
The second historical advertisement delivery combination sequence data is different from the first historical advertisement delivery combination sequence data used for training and generating the confrontation network time sequence model in the step S110, and the second historical advertisement delivery combination sequence data is historical advertisement delivery sequence data selected by any user in a certain time period according to the actual situation of the user when the user acquires the target advertisement delivery plan by adopting the advertisement delivery plan generating system. The target parameters comprise average unit price of the user for putting the advertisement on each platform when generating a new advertisement putting plan, budget threshold value of putting the advertisement, combination duration of advertisement putting on each platform and the like. The simulated target advertisement putting combination is a new advertisement putting combination and return data obtained by inputting second historical advertisement putting combination sequence data and target parameters input by a user into a trained generation pairing network timing model.
In an embodiment, the inputting the second historical advertisement placement combination sequence data and the target parameter into the generated countermeasure network timing model to obtain a simulated target advertisement placement combination includes: inputting second historical advertisement putting combination sequence data and synthetic data into the characterization network, inputting the synthetic data of the second historical advertisement putting combination sequence data into the countermeasure network for joint training and generating simulated target advertisement putting combination data when iteration exceeds a preset threshold.
S130, processing the simulation target advertisement putting combination and determining a target advertisement putting plan.
In the embodiment of the present application, the simulation target advertisement delivery combination is not the final target advertisement delivery plan, and after the generation of the confrontation network timing model and the output of the simulation target advertisement delivery combination, the simulation target advertisement delivery combination needs to be subjected to inspection processing and search processing. Verification process the authenticity of the simulated targeted ad placement portfolio data is tested by two types of sequence prediction models, ConvLSTM (as shown in FIG. 4) and Transformer (the specific data processing procedures for such models are not shown in the figures). And on the basis of confirming the authenticity of the simulated target advertisement delivery combination data, searching the simulated target advertisement delivery combination, wherein the searching process refers to traversing the combination return data of all the simulated target advertisement delivery combinations, calculating whether the budget of the advertisement delivery plan is within the budget threshold range set by the user according to target parameters input by the user, such as the average unit price of each platform advertisement resource, and searching the optimal advertisement delivery combination and the budget allocation scheme corresponding to the optimal advertisement delivery combination according to the combination duration input by the user and delivered by each platform advertisement resource.
In an embodiment, the processing the simulated targeted advertisement delivery combination to determine a targeted advertisement delivery plan includes: checking the simulation target advertisement putting combination to judge whether the simulation target advertisement putting combination is real or not; and searching the simulated target advertisement putting combination to output the optimal solution.
The generation method of the advertisement delivery plan comprises the steps of training and generating a confrontation network time sequence model according to first historical advertisement delivery combination sequence data, preset parameters and random vectors; inputting second historical advertisement putting combination sequence data and target parameters into the generated countermeasure network timing model to obtain a simulated target advertisement putting combination; the simulation target advertisement putting combination is processed, a target advertisement putting plan is determined, a more scientific advertisement putting plan can be automatically generated according to the condition of cross-platform advertisement putting, the technical scheme of carrying out advertisement putting according to thought experience in the prior art is replaced, manpower consumption can be reduced, decision efficiency is improved, budget can be more scientifically and reasonably distributed, and the specific advertisement putting condition in the future every day is optimized.
Fig. 2 is an architecture diagram of a joint training for generating a countermeasure network timing model according to an embodiment of the present disclosure, and as shown in fig. 2, the main process of the joint training for generating the countermeasure network timing model according to the embodiment of the present disclosure is as follows:
for the resource delivery sequence data, the variable dimension comprises a time variable, a delivery data variable of each platform and a final return variable. Each case of data comprises two types of characteristics, namely static characteristics (which do not change along with time, such as minimum visit quantity of a microblog) and time characteristics (which change along with time, such as visit quantity which changes along with the increase of a topic). And setting S as a vector space of static characteristics, setting X as a vector space with time sequence characteristics, and setting S as S and X as a random vector of a corresponding space. Suppose the release characteristics are (S, X)1:T) The tuples of the form satisfy some joint distribution p. The training set is represented as
Figure BDA0003150010180000061
The goal is to make the training data D learn to distribute
Figure BDA0003150010180000062
And can be closest to the true distribution p (S, X)1:T) Additionally using the pair distribution p (S, X)1:T)＝p(S)∏tp(Xt|S,X1:T-1) Learning distributions by autoregressive decomposition
Figure BDA0003150010180000063
In the closest distribution p (X)t|S,X1:T-1). The final goal can be described as two minimization problems,
1) minimizing the distribution difference:
Figure BDA0003150010180000064
d is calculating the difference between the two distributions.
2) Minimizing conditional probability distribution variance:
Figure BDA0003150010180000065
the generation countermeasure network timing model adopted in the embodiment of the application is composed of four networks in total: the system comprises an embedded network, a recovery network, a sequence generation network and a sequence discrimination network. The embedded network maps the release combination characteristic sequence S, X to the hidden space HS,HXCapturing static characteristics and dynamic characteristics of the release characteristic sequence, and restoring the network from the hidden space to the original space
Figure BDA0003150010180000066
The sequence generation network and the sequence discrimination network jointly form a countermeasure network, and the operation is carried out in a hidden space:
1) the potential dynamics of real data versus synthetic data is constrained by reconstruction and supervised losses:
Figure BDA0003150010180000067
where λ ≧ 0 is a hyperparameter that balances the two losses.
2) The confrontation network training comprises a generator and a discriminator, and is completed by minimizing unsupervised loss and maximizing supervised loss:
Figure BDA0003150010180000071
where η ≧ 0 is another superparameter which balances the two losses. L isSThe inclusion of an embedding process not only helps to reduce the dimensionality of the learning space, but also helps the generator learn the temporal relationships from the data.
The four networks are jointly trained and iterated more than 10000 times, so that the finally generated delivery combination data is enough to be like real delivery combination and the distribution is enough to cover the real data distribution.
The output module outputs a large number of simulated delivery combinations to the inspection module and the search module, and performs prediction effect inspection and optimal solution search on the generated delivery combinations respectively.
The test module of the application comprises two types of sequence prediction models, namely ConvLSTM and Transformer, for testing the prediction scores of the generated data: mean Square Error (MAPE), Root Mean Square Error (RMSE), and fitness R-squared.
Fig. 3 is a flowchart of generating the processing data of the countermeasure network timing model according to an embodiment of the present application, and as shown in fig. 3, the main process of generating the processing data of the countermeasure network timing model according to the embodiment of the present application is as follows:
the model learns jointly between the embedded network and the countermeasure network. Learning is primarily responsible for optimizing two loss functions:
1) loss due to reconstruction loss:
Figure BDA0003150010180000072
2) in a countermeasure network, a generator is embedded h by accepting an artificialS,h1:t-1To generate the next htThe gradient is then calculated by unsupervised loss. And then by providing the real data hS,h1:TAnd artificial data
Figure BDA0003150010180000073
Accurate classification y ofS,y1:TTo maximize (arbiter) or minimize (generator) likelihood to promote classification capability.
Figure BDA0003150010180000074
The excitation generator captures a stepwise conditional distribution in the data by constraining the learning by introducing additional penalties in addition to the antagonistic feedback of the generator and the arbiter. In the training process, studentsThe synthesizer receives real data h1:t-1To generate the next potential vector ht. The gradient can now be in the distribution p (H)t|HS,H1:t-1) And
Figure BDA0003150010180000075
the loss of difference between them is calculated. Supervision penalty with maximum likelihood:
Ls＝Es,x1:T～p[∑t||ht-gX(hS,ht-1,zt)||2]
wherein g isX(hs,ht-1,zt) Using a sample ztApproximation
Figure BDA0003150010180000081
LUThe LS further ensures that it produces a similar gradual transition effect as the push generator creates the sequence.
The method for generating the advertisement putting plan comprises the following steps:
in the first step, a user uploads sequence data, and economic indexes and macroscopic variables can be added according to an actual scene. And setting the combined time length of the impressions required to be generated.
Secondly, model operation: the model firstly carries out embedded coding on the data, then combines countercheck learning to generate time sequence data,
FIG. 3 is an execution flow after the model receives data: receiving model training data and parameters, and firstly selecting a normalization method according to the parameters; then constructing each piece of supervision training data according to the step length parameters; training the real input mapped data to generate a network under supervision; then training the whole network by combining confrontation, supervision and loss recovery; finally, by sampling a large number of samples from the known distribution, it is constrained to true investment sequence data by the upload network. Many corresponding operations can be performed on the output of the model to meet business requirements including verifying the authenticity of the generated sequences, searching for the highest selling sequences, visually presenting the final results, saving the results, and the like.
Fig. 4 is a flowchart of a method for generating prediction model processing data included in a countermeasure network timing model verification module according to an embodiment of the present application. The data processing of the ConvLSTM model output reward prediction is shown.
Fig. 5 is a block diagram of an advertisement placement plan generating system according to an embodiment of the present application, and as shown in fig. 5, the advertisement placement plan generating system according to the embodiment of the present application includes: the system comprises a front-end module, a service processing module and a data processing module;
the front-end module is used for sending a request for generating a target advertisement putting plan and displaying the target advertisement putting plan;
the business processing module is used for receiving the target advertisement putting plan generating request, inputting training data to the data processing layer and generating a confrontation network time sequence model;
the data processing module is configured to receive the training data and the generated confrontation network time sequence model, obtain the target advertisement delivery plan, and return the target advertisement delivery plan to the service processing layer.
The system is composed of three layers, namely a front-end layer, a service logic layer and a data processing layer. The three layers jointly realize the operations of creating, executing, storing, verifying and the like of the service. The concrete description is as follows:
the main functions of the front-end layer are: user registration and login, data uploading, data/service displaying, service managing, result checking/downloading and data initialization processing. Firstly, a user needs to register an account and then registers the account; if a new task is to be created, a user needs to click to create the task, then according to information such as a prompt uploading file, a model parameter and the like, after clicking submission, data and parameters are uploaded to a background through a business API of the created task, then a homepage is returned, and the business state is in execution. And when the service execution is finished, the server returns the execution state, the front end updates the execution state, and the model result can be displayed by clicking display, including the generation of each channel sequence with the highest prediction and the corresponding investment suggestion. The model results may be saved by click-to-download.
The service logic layer is responsible for receiving service requests, verifying identities, storing tasks and data, calling models, obtaining model results, analyzing returned results and returning task results. The service logic layer receives the task request of the front end, verifies the user identity according to the request type, and then performs corresponding processing; if the task is created and executed, firstly storing the model data and the parameters to a database after the identity authentication is successful, and simultaneously associating the task creation, management and use of a user account; then, the model and the parameters are used by calling API through the model, and simultaneously the task state is saved; and after receiving the model result, modifying the task state and reminding the front-end task of changing the task state. If the operation is other operations of the task, such as deleting or modifying the user, the operation is directly carried out, and then the execution result is returned.
The data processing layer is mainly responsible for receiving training data input by the business logic layer, training the model, storing the model result and returning the model result to the business logic layer. The model receives data and parameters transmitted by the business logic layer, and the data and the parameters are used for model training; saving the output result of the model to a database; and returning the result state of the model processing to the business logic layer.
Referring to fig. 6, a schematic structural diagram of an electronic device in an example of the present application is shown. As shown in fig. 6, the electronic device may include: a processor 801, a storage device 802, a display screen 803 with touch functionality, an input device 804, an output device 805, and a communication device 806. The number of the processors 801 in the device may be one or more, and one processor 801 is taken as an example in fig. 6. The processor 801, the storage means 802, the display 803, the input means 804, the output means 805 and the communication means 806 of the apparatus may be connected by a bus or in another way, as exemplified by the bus connection in fig. 6. The device is configured to execute the advertisement delivery plan generating method provided by any embodiment of the present application.
The embodiment of the present application further provides a computer-readable storage medium, where instructions in the storage medium, when executed by a processor of an electronic device, enable the electronic device to execute the advertisement delivery plan generating method according to the above method embodiment.
It should be noted that, for the embodiments of the electronic device and the storage medium, since they are basically similar to the embodiments of the method, the description is relatively simple, and for the relevant points, reference may be made to the partial description of the embodiments of the method.
The method and the device have the advantages that the generation countermeasure network in the image generation technology is applied to the field of delivery plan/budget allocation, compared with the traditional delivery plan/budget allocation mode, the generation countermeasure network time sequence model is adopted, long-time resource delivery combination data are used as the feature matrix, and a large amount of long-time resource delivery feature tensors are used as training input data by the encoder. The method for generating the confrontation network by the time sequence increases supervised learning and an embedded network on the basis of generating the confrontation network generally, so that various sequence data capable of simulating the throwing style and characteristics of people are generated by utilizing the unsupervised learning flexibility of the generated confrontation network and the capturing capability of the supervised network on the time sequence dynamics, and the future throwing can be guided more scientifically and efficiently. And the step-by-step condition distribution of the real data is captured by utilizing supervised learning and antagonistic network joint training, so that the generated new sequence can inherit the time dynamics of the real data.
The application provides an advertisement putting plan generating system, which is divided into three layers: the system comprises a bottom data operation layer, a service logic layer and a front-end application layer. The user inputs historical releasing and effect return data at the front end, the data is processed by the business layer and transmitted to the operation layer, the generated model returns the front end result after training is completed, the front end displays the releasing combination and the estimated effect return which are optimal in future time (for example, one month in the future), the background converts according to the latest average unit price of the resources, and the corresponding budget allocation scheme is displayed. And (4) creating a project, selecting a model type to be applied, and checking a final result by one-key operation.
In the description herein, reference to the description of the term "one embodiment," "some embodiments," "an example," "a specific example," or "some examples," etc., means that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the application. In this specification, the schematic representations of the terms used above do not necessarily refer to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.
It is noted that the foregoing is only illustrative of the embodiments of the present application and the technical principles employed. It will be understood by those skilled in the art that the present application is not limited to the particular embodiments described herein, but is capable of many obvious modifications, rearrangements and substitutions without departing from the scope of the application. Therefore, although the present application has been described in more detail with reference to the above embodiments, the present application is not limited to the above embodiments, and may include other equivalent embodiments without departing from the spirit of the present application, and the scope of the present application is determined by the scope of the appended claims.

Claims (11)

1. An advertisement placement plan generating method includes:
training according to the first historical advertisement putting combination sequence data, preset parameters and random vectors to generate a confrontation network time sequence model;
inputting second historical advertisement putting combination sequence data and target parameters into the generated countermeasure network timing model to obtain a simulated target advertisement putting combination;
and processing the simulated target advertisement putting combination to determine a target advertisement putting plan.
2. The method of claim 1, wherein training the generation of the confrontational network timing model from the first historical advertising placement combined sequence data and the preset parameters comprises:
training a characterization network in the generated confrontation network time sequence model according to the first historical advertisement putting combination sequence data and the preset parameters;
training the countermeasure network in the countermeasure network timing model according to the random vector;
training the supervising network according to the characterizing network and the countering network;
wherein the characterization network, the countermeasure network, and the supervisory network comprise the generated countermeasure network timing model.
3. The method of claim 2, wherein training the characterization network in the generated confrontation network timing model according to the first historical advertisement placement combined sequence data and the preset parameters comprises:
determining a first placement characteristic (S, X) of the first historical ad placement combined sequence data1：T) And said first delivery feature (S, X) is delivered via an embedded network1：T) Mapping from original space to hidden space to obtain corresponding first hidden vector hS，h1：T＝e(s，x1：t)；
The first hidden vector h is transmitted through a recovery networkS，h1：T＝e(s，x1：t) Second launch feature restored to the original space
Figure RE-FDA0003402245650000011
Figure RE-FDA0003402245650000012
Wherein the characterization network comprises the embedded network and the recovery network.
4. The method of claim 3, wherein the countermeasure network comprises the sequence generation network and the sequence discrimination network, and training the countermeasure network in the generated countermeasure network timing model according to the random vector comprises:
the sequence generation network generates a second hidden vector feature h in a hidden space according to the random vectors，h1：T＝g(zS，z1：T)；
The sequence discrimination network is used for the second implicit vector feature hs，h1：T＝g(zS，z1：T) Distinguishing to obtain sequence classification ys，y1：T＝d(hS，h1：T)；
Wherein h isS＝gS(zS)，ht＝gX(hS，ht-1，zt)。
5. The method of claim 3, said first hidden vector h being passed through a recovery networkS，h1：T＝e(s，x1：t) Second launch feature restored to the original space
Figure RE-FDA0003402245650000021
Then, the method further comprises the following steps:
calculating reconstruction loss
Figure RE-FDA0003402245650000022
6. The method of claim 4, wherein,
the sequence generation network is built by accepting artificial embedding hS，h1：t-1To generate the next htCalculating the gradient by unsupervised loss;
classifying y according to said sequenceS，y1：T＝d(hS，h1：T) Calculating unsupervised losses
Figure RE-FDA0003402245650000023
Figure RE-FDA0003402245650000024
According to the real data hS，h1：TAnd said calculated gradient, calculated supervised loss LS＝Es，x1：T～p[∑t||ht-gX(hS，ht-1，zt)||2]。
7. The method of claim 2, wherein said inputting the second historical ad placement combination sequence data and targeting parameters into the generative confrontation network timing model results in a simulated targeted ad placement combination comprising:
inputting second historical advertisement putting combination sequence data and synthetic data into the characterization network, inputting the synthetic data of the second historical advertisement putting combination sequence data into the countermeasure network for joint training and generating simulated target advertisement putting combination data when iteration exceeds a preset threshold.
8. The method of claim 1, wherein processing the simulated targeted ad placement combination to determine a targeted ad placement plan comprises:
checking the simulation target advertisement putting combination to judge whether the simulation target advertisement putting combination is real or not;
and searching the simulated target advertisement putting combination to output the optimal solution.
9. An advertisement placement plan generation system comprising: the system comprises a front-end module, a service processing module and a data processing module;
the front-end module is used for sending a request for generating a target advertisement putting plan and displaying the target advertisement putting plan;
the business processing module is used for receiving the target advertisement putting plan generating request, inputting training data to the data processing layer and generating a confrontation network time sequence model;
the data processing module is configured to receive the training data and the generated confrontation network timing model to obtain the target advertisement delivery plan, and return the target advertisement delivery plan to the service processing layer.
10. An electronic device, the electronic device comprising:
one or more processors;
a storage device arranged to store one or more programs,
the one or more programs, when executed by the one or more processors, cause the one or more processors to implement the advertising placement plan generating method of any of claims 1-8.
11. A computer-readable storage medium, on which a computer program is stored which, when being executed by a processor, implements the advertisement placement plan generating method according to any one of claims 1 to 8.
CN202110761489.4A 2021-07-06 2021-07-06 Method and system for generating advertisement putting plan, electronic device and storage medium Pending CN113971582A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110761489.4A CN113971582A (en) 2021-07-06 2021-07-06 Method and system for generating advertisement putting plan, electronic device and storage medium

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110761489.4A CN113971582A (en) 2021-07-06 2021-07-06 Method and system for generating advertisement putting plan, electronic device and storage medium

Publications (1)

Publication Number Publication Date
CN113971582A true CN113971582A (en) 2022-01-25

Family

ID=79586198

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110761489.4A Pending CN113971582A (en) 2021-07-06 2021-07-06 Method and system for generating advertisement putting plan, electronic device and storage medium

Country Status (1)

Country Link
CN (1) CN113971582A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114581128A (en) * 2022-02-28 2022-06-03 北京沃东天骏信息技术有限公司 Advertisement putting method and device, electronic equipment and computer readable medium

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114581128A (en) * 2022-02-28 2022-06-03 北京沃东天骏信息技术有限公司 Advertisement putting method and device, electronic equipment and computer readable medium

Similar Documents

Publication Publication Date Title
Feng et al. Understanding dropouts in MOOCs
Gao et al. Deep leaf‐bootstrapping generative adversarial network for structural image data augmentation
CN112784994B (en) Block chain-based federated learning data participant contribution value calculation and excitation method
CN111291266A (en) Artificial intelligence based recommendation method and device, electronic equipment and storage medium
CN110462612A (en) The method and apparatus for carrying out machine learning using the network at network node with ageng and ranking then being carried out to network node
CN111973996B (en) Game resource release method and device
CN112052948B (en) Network model compression method and device, storage medium and electronic equipment
Rahman A deep learning framework for football match prediction
CN110766038A (en) Unsupervised landform classification model training and landform image construction method
CN109409739B (en) Crowdsourcing platform task allocation method based on POMDP model
CN113283948B (en) Generation method, device, equipment and readable medium of prediction model
CN102147727A (en) Method for predicting software workload of newly-added software project
Castilla‐Rho Groundwater Modeling with Stakeholders: Finding the Complexity that Matters.
CN109925718A (en) A kind of system and method for distributing the micro- end map of game
Arroyo et al. Re-thinking simulation: a methodological approach for the application of data mining in agent-based modelling
Schleier-Smith An architecture for agile machine learning in real-time applications
Zhou et al. Probabilistic graphical models parameter learning with transferred prior and constraints
CN113256335B (en) Data screening method, multimedia data delivery effect prediction method and device
CN113971582A (en) Method and system for generating advertisement putting plan, electronic device and storage medium
CN115049397A (en) Method and device for identifying risk account in social network
Muzy et al. Activity-based credit assignment heuristic for simulation-based stochastic search in a hierarchical model base of systems
Praynlin et al. Performance analysis of software effort estimation models using neural networks
CN112818241B (en) Content promotion method and device, computer equipment and storage medium
CN113762324A (en) Virtual object detection method, device, equipment and computer readable storage medium
Gemp et al. Developing, evaluating and scaling learning agents in multi-agent environments

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination