CN111178981A - Advertisement putting method and device, computer equipment and storage medium - Google Patents
Advertisement putting method and device, computer equipment and storage medium Download PDFInfo
- Publication number
- CN111178981A CN111178981A CN202010002156.9A CN202010002156A CN111178981A CN 111178981 A CN111178981 A CN 111178981A CN 202010002156 A CN202010002156 A CN 202010002156A CN 111178981 A CN111178981 A CN 111178981A
- Authority
- CN
- China
- Prior art keywords
- advertisement
- rate
- value
- marginal
- cost
- 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.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 55
- 238000006243 chemical reaction Methods 0.000 claims abstract description 114
- 230000008569 process Effects 0.000 claims abstract description 27
- 230000002860 competitive effect Effects 0.000 claims abstract description 17
- 238000012549 training Methods 0.000 claims description 46
- 230000004927 fusion Effects 0.000 claims description 30
- 230000006399 behavior Effects 0.000 claims description 16
- 238000004590 computer program Methods 0.000 claims description 14
- 238000004364 calculation method Methods 0.000 claims description 10
- 238000002716 delivery method Methods 0.000 claims description 10
- 230000006870 function Effects 0.000 claims description 9
- 238000012935 Averaging Methods 0.000 claims description 4
- 238000000605 extraction Methods 0.000 claims description 4
- 238000010586 diagram Methods 0.000 description 7
- 238000011156 evaluation Methods 0.000 description 7
- 230000008901 benefit Effects 0.000 description 6
- 238000013528 artificial neural network Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 230000010354 integration Effects 0.000 description 3
- 238000013526 transfer learning Methods 0.000 description 3
- 238000010276 construction Methods 0.000 description 2
- 238000012163 sequencing technique Methods 0.000 description 2
- 230000001360 synchronised effect Effects 0.000 description 2
- 230000001131 transforming effect Effects 0.000 description 2
- 206010044565 Tremor Diseases 0.000 description 1
- 230000009471 action Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000001186 cumulative effect Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000004091 panning Methods 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
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
- G06Q30/0242—Determining effectiveness of advertisements
- G06Q30/0244—Optimization
-
- 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
- G06Q30/0247—Calculate past, present or future revenues
Landscapes
- Business, Economics & Management (AREA)
- Engineering & Computer Science (AREA)
- Accounting & Taxation (AREA)
- Development Economics (AREA)
- Strategic Management (AREA)
- Finance (AREA)
- Game Theory and Decision Science (AREA)
- Entrepreneurship & Innovation (AREA)
- Economics (AREA)
- Marketing (AREA)
- Physics & Mathematics (AREA)
- General Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses an advertisement putting method, an advertisement putting device, computer equipment and a storage medium, belonging to the field of internet advertisements, wherein the method comprises the following steps: obtaining click rate pre-estimated values, conversion rate pre-estimated values and marginal cost rate pre-estimated values of the current user for all the advertisements in the candidate advertisement set through a pre-trained multi-task model; calculating thousands of display gains of each advertisement based on a click rate pre-estimated value, a conversion rate pre-estimated value, a marginal cost rate pre-estimated value and a competitive value of each advertisement of a current user; and selecting advertisements from the advertisement set according to the thousands of display profits of each advertisement and putting the advertisements to the current user. The embodiment of the invention can realize the accurate advertisement delivery in industries such as financial insurance and the like, and meet the requirement of the industries such as financial insurance and the like on accurate operation in the advertisement delivery process.
Description
Technical Field
The invention relates to the field of internet advertisements, in particular to an advertisement putting method, an advertisement putting device, computer equipment and a storage medium.
Background
Currently, the competitive advertising of internet information flow products has become a mainstream choice for advertisers, and the demand is continuously increasing, and the coverage industry is also more and more extensive. In the information flow bidding advertisement putting process, the display opportunity is not guaranteed, but competition is carried out in a bidding sorting mode, and the bidding logics of the mainstream flow platforms are basically consistent, namely the mode of sorting based on estimated thousands of display benefits (eCPM).
The most primitive eCMP computation mode is: the estimated thousand show revenues are equal to the advertiser bid multiplied by the estimated click through rate CTR multiplied by 1000. In this mode, the advertiser bids and the probability of the user clicking on the material are two major factors affecting the exposure of the advertisement, and ideally, a high CTR can bid lower to obtain greater traffic. However, clicks are often not the ultimate goal of ad placement, and click-through-not-conversion often means a reduction in investment profitability for the advertiser.
Therefore, traffic platforms like panning, trembling, today's first line, etc. propose an optimized CPM bidding model that considers not only click through rate CTR but also the client's conversion target CVR. This model is more friendly to advertisers due to the consideration of conversion factor, and if the estimated CVR is high, it can get the chance of advertisement presentation at a relatively low bid even if the CTR is not high, and this model requires the addition of a conversion CVR prediction model with high accuracy.
The inventor finds that the prior art has the following defects in the process of implementing the invention:
the highly accurate CVR predictive model can ensure that an advertiser's advertisements are more accurately shown to a group of potentially high conversion users (who are highly interested in advertised goods), reducing bid costs and thereby increasing ROI, but this is based on the assumption that the advertiser's revenue is achieved after users have converted in accordance with the advertiser's conversion goals. For example, if a user purchases an item through an advertisement click, the advertiser may obtain the price of the item, which revenue is fixed. For many industries, however, the actual gains after achieving the conversion goal vary widely among different populations. For example, in the game industry, the conversion target is usually to download and install apps, at most, login and register accounts, and consumption levels of different crowds in the game process are very different. For another example, in the financial field, especially in the insurance industry, risk levels of different users may be different, and a high-risk user means that a reimbursement cost is high, and for an advertiser, a display opportunity that wants to launch an advertisement can be accurately hooked with the risk level of a target user, so that the cost is effectively reduced, more profits are obtained, and for such problems, the optimized CPM bidding mode provided by the prior art does not consider dynamic costs of industries such as financial insurance and the like for different users, so that the advertisement of the industries such as financial insurance and the like is difficult to be launched accurately, and the requirement of the industries such as financial insurance and the like for accurate operation in the advertisement launching process cannot be met.
Disclosure of Invention
In order to solve at least one of the problems mentioned in the background above, the present invention provides an advertisement delivery method, an advertisement delivery apparatus, a computer device, and a storage medium.
The embodiment of the invention provides the following specific technical scheme:
in a first aspect, a method for advertisement delivery is provided, the method including:
obtaining click rate pre-estimated values, conversion rate pre-estimated values and marginal cost rate pre-estimated values of the current user for all the advertisements in the candidate advertisement set through a pre-trained multi-task model;
calculating thousands of display gains of each advertisement based on click rate pre-estimated values, conversion rate pre-estimated values and marginal cost rate pre-estimated values of the current user for each advertisement and the competitive value of each advertisement;
and selecting advertisements from the advertisement set according to the thousands of display profits of the advertisements and delivering the advertisements to the current user.
Further, the multitask model includes a representation extractor layer, a feature fusion layer, a click-through rate prediction model, a conversion rate prediction model and a cost prediction model, and the obtaining of the click-through rate prediction value, the conversion rate prediction value and the marginal cost rate prediction value of each advertisement in the candidate advertisement set by the current user through the pre-trained multitask model includes:
acquiring advertisement data of each advertisement and behavior data of the current user;
for each advertisement, inputting the behavior data of the current user and the advertisement data of the advertisement into the representation extractor layer for feature extraction to obtain a plurality of feature data;
inputting the plurality of feature data into the feature fusion layer to be respectively subjected to weighted fusion according to corresponding weight parameters to obtain a plurality of fusion features which are respectively used for click rate prediction, conversion rate prediction and marginal cost rate prediction;
and correspondingly inputting the plurality of fusion characteristics into the click rate prediction model, the conversion rate prediction model and the cost prediction model to obtain a click rate prediction value, a conversion rate prediction value and a marginal cost rate prediction value of the current user for the advertisement.
Further, the plurality of feature data includes a task sharing feature, a differentiate if converted feature, a differentiate if clicked feature, and a differentiate user risk level feature.
Further, the training process of the multitask model comprises the following steps:
constructing a first sample data set according to guest group data provided by a traffic platform and user conversion data provided by an advertiser, wherein the user conversion data comprises positive sample data of advertisement touch and conversion and negative sample data of advertisement touch and conversion;
taking a click rate prediction task as a source task and a conversion rate prediction task as a target task, and training the representation extractor layer, the click rate prediction model and the conversion rate prediction model in the multi-task model based on the first sample data set to obtain the pre-trained multi-task model;
constructing a second sample data set, wherein the second sample data set is an intersection of user sample data which is provided by an advertiser and labeled with a cost tag and the guest group data after collision, and the cost tag is normalized cost data;
and training the cost prediction model in the pre-trained multitask model based on the second sample data set to obtain the trained multitask model.
Further, the training the cost prediction model in the pre-trained multitask model based on the second sample data set to obtain the trained multitask model includes:
iteratively training the cost prediction model in the pre-trained multitask model based on the second sample data set;
in each iterative training, calculating a loss value between a marginal cost rate estimated value output by the cost prediction model and a marginal cost rate true value of a corresponding sample by using a preset loss function;
and stopping iterative training of the cost prediction model when the loss value reaches a preset value, so as to obtain the trained multi-task model.
Further, the calculating the thousand-time display profit of each advertisement based on the click rate pre-evaluation value, the conversion rate pre-evaluation value and the marginal cost rate pre-evaluation value of the current user for each advertisement and the bid value of each advertisement comprises:
calculating a marginal rate of return estimate of the current user for each advertisement based on a click rate estimate, a conversion rate estimate and a marginal cost rate estimate of the current user for each advertisement;
and calculating thousands of display gains of each advertisement based on the click rate pre-estimated value, the conversion rate pre-estimated value, the marginal rate pre-estimated value and the competitive value of each advertisement of the current user.
Further, the method further comprises:
acquiring a marginal yield pre-estimation value adjusting coefficient corresponding to each advertisement;
calculating thousand display gains of each advertisement based on the click rate pre-estimated value, the conversion rate pre-estimated value, the marginal rate pre-estimated value and the competitive value of each advertisement of the current user, wherein the calculation comprises the following steps:
and calculating thousands of display gains of each advertisement based on the click rate pre-estimated value, the conversion rate pre-estimated value, the marginal rate pre-estimated value, the bid value of each advertisement and the marginal rate pre-estimated value adjusting coefficient corresponding to each advertisement of the current user.
Further, the obtaining a marginal rate of return estimate adjustment coefficient corresponding to each advertisement includes:
aiming at each sample user in a plurality of sample users, obtaining a click rate estimated value, a conversion rate estimated value and a marginal cost rate estimated value of the sample user for each advertisement in the candidate advertisement set through the pre-trained multitask model;
calculating a marginal rate-of-return estimate for each of the advertisements for each of the sample users based on the click-through rate estimate, the conversion rate estimate, and the marginal cost rate estimate for each of the advertisements for each of the sample users;
respectively averaging the marginal rate of return predicted values of all the sample users for each advertisement to obtain a marginal rate of return predicted value average value corresponding to each advertisement;
and respectively determining the average value of the marginal rate of return estimates corresponding to each advertisement as the adjustment coefficient of the marginal rate of return estimates corresponding to each advertisement.
In a second aspect, an advertisement delivery apparatus is provided, the apparatus comprising:
the acquisition module is used for acquiring click rate pre-estimated values, conversion rate pre-estimated values and marginal cost rate pre-estimated values of the current user for all the advertisements in the candidate advertisement set through a pre-trained multi-task model;
the calculation module is used for calculating thousand-time display income of each advertisement based on the click rate pre-estimated value, the conversion rate pre-estimated value and the marginal cost rate pre-estimated value of the current user for each advertisement and the competitive value of each advertisement;
and the releasing module is used for selecting the advertisements from the advertisement set and releasing the advertisements to the current user according to the thousands of display profits of the advertisements.
In a third aspect, a computer device is provided, comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the steps of the advertisement delivery method according to the first aspect when executing the computer program.
In a fourth aspect, a computer-readable storage medium is provided, on which a computer program is stored, which computer program, when being executed by a processor, carries out the steps of the advertisement delivery method according to the first aspect.
The technical scheme provided by the embodiment of the invention obtains the click rate pre-estimated value, the conversion rate pre-estimated value and the marginal cost rate pre-estimated value of the current user for each advertisement in the candidate advertisement set through a pre-trained multi-task model, calculates the thousand display gains of each advertisement based on the click rate pre-estimated value, the conversion rate pre-estimated value and the marginal cost rate pre-estimated value of the current user for each advertisement and the competitive value of each advertisement, selects and puts the advertisement from the advertisement set to the current user according to the thousand display gains of each advertisement, and not only considers the click rate pre-estimated value and the conversion rate pre-estimated value of the current user for each advertisement when predicting the potential thousand display gains (eCPM) in the advertisement putting process, but also increases the factor for predicting simple marginal cost (profit), thus carrying out advertisement sequencing based on the calculated thousand display gains of each advertisement, the advertisement of trades such as financial insurance can be made to put in more accurately, not only satisfies trades such as financial insurance and puts in the demand of the accurate operation of in-process at the advertisement, makes the advertiser can reduce cost effectively, acquires more profits, makes the advertisement of all advertisers put in more fairly moreover on the flow platform.
Drawings
In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings needed to be used in the description of the embodiments will be briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without creative efforts.
Fig. 1 is a flowchart of an advertisement delivery method according to an embodiment of the present invention;
FIG. 2 is a schematic structural diagram of a multitasking model according to an embodiment of the present invention;
FIG. 3 is a detailed flowchart of step S11 of the method shown in FIG. 1; (ii) a
FIG. 4 is a detailed flow chart of a training process of the multitask model in the method shown in FIG. 1; (ii) a
FIG. 5 is a detailed flowchart of step S12 of the method shown in FIG. 1; (ii) a
Fig. 6 is a flowchart of an advertisement delivery method according to an embodiment of the present invention;
FIG. 7 is a detailed flowchart of step S63 of the method shown in FIG. 6; (ii) a
Fig. 8 is a structural diagram of an advertisement delivery device according to an embodiment of the present invention;
fig. 9 is an internal structural diagram of a computer device according to an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
It is to be understood that, unless the context clearly requires otherwise, throughout the description and the claims, the words "comprise", "comprising", and the like are to be construed in an inclusive sense as opposed to an exclusive or exhaustive sense; that is, what is meant is "including, but not limited to".
Furthermore, in the description of the present invention, it is to be understood that the terms "first," "second," and the like are used for descriptive purposes only and are not to be construed as indicating or implying relative importance. In addition, in the description of the present invention, "a plurality" means two or more unless otherwise specified.
In the actual advertising process, for many industries, the actual revenue after the conversion goal is achieved varies greatly among different populations. For example, in the game industry, the conversion target is usually to download and install apps, at most, login and register accounts, and consumption levels of different crowds in the game process are very different. For another example, in the financial field, especially in the insurance industry, risk levels of different users may be different, and a high-risk user means that a payment cost is high, and for an advertiser, a display opportunity that wants to launch an advertisement can be accurately hooked with a risk level of a target user, so that cost is effectively reduced, more profits are obtained, and for such problems, in an actual advertisement launching process, an optimized CPM bidding mode (optimized CPM, opcpm) provided in the prior art only considers a click-through rate CTR and a conversion target CVR of a client, and does not consider dynamic costs of industries such as financial insurance and the like for different users, so that the advertisement of the industries such as financial insurance and the like is difficult to be launched accurately, and the demand of the industries such as financial insurance and the like for accurate operation in the advertisement launching process cannot be met. Therefore, the method increases a factor (eMR) for estimating simple Marginal cost (profit) in the definition for estimating potential thousand-time display profit (eCPM), and enables the quality of the current user to influence the probability of advertisement reach, thereby realizing accurate advertisement delivery in industries such as financial insurance and meeting the requirement of accurate operation in the advertisement delivery process in industries such as financial insurance and the like.
It should be noted that, in the description of the present invention, the marginal cost is referred to as "simple marginal cost", and the marginal benefit is referred to as "simple marginal benefit".
Example one
An execution subject of the method may be a server, and the server may adopt an independent server or a server cluster, as shown in fig. 1, the method may include:
and step S11, obtaining a click rate pre-estimated value, a conversion rate pre-estimated value and a marginal cost rate pre-estimated value of the current user for each advertisement in the candidate advertisement set through a pre-trained multitask model.
And step S12, calculating thousands of display gains of each advertisement based on the click rate pre-estimated value, the conversion rate pre-estimated value and the marginal cost rate pre-estimated value of the current user for each advertisement and the competitive value of each advertisement.
And step S13, selecting advertisements from the advertisement set according to the thousands of display profits of each advertisement and delivering the advertisements to the current user.
Through the steps, the embodiment of the invention can calculate the thousands of display yields of each advertisement in the candidate advertisement set and sort the display yields from high to low, thereby selecting the advertisement with the first sort to be delivered to the current user, since the click-through rate estimate and the conversion rate estimate of each advertisement of the current user are not only considered when estimating the potential thousand showing returns (eCPM) in the advertisement putting process, and the factor of estimating simple marginal cost (income) is increased, and the advertisement sequencing is carried out based on thousands of display income of each advertisement obtained by calculation, so that the advertisements of industries such as financial insurance can be put in more accurately, the requirements of industries such as financial insurance on accurate operation in the advertisement putting process are met, an advertiser can effectively reduce the cost, more income is obtained, and the advertisement putting of the whole advertiser on the flow platform is more fair.
In one embodiment, the multitasking model comprises a representation extractor layer, a feature fusion layer, a click-through rate prediction model, a conversion rate prediction model and a cost prediction model.
Fig. 2 is a schematic structural diagram of a multitask model provided in an embodiment of the present invention, and as shown in fig. 2, the entire architecture is divided into three layers, and the inside of each layer may be a multilayer neural network substructure:
a) the bottom layer is provided with a plurality of relatively weak sub-networks representing extractors, the sub-networks are not connected, a deep neural network structure (DNN) can be adopted as an extractor representing (representation), a specific network sub-structure does not need to be specified, and the network sub-structure is pluggable and replaceable in the whole framework;
b) the second layer is a sparse connection fusion layer, the output results of all weak extractors are integrated by adopting the idea of ensemble learning, the integration mode can be various, such as weighted sum, but the weight is learnable, and different weights can be learnt for different subtasks;
c) the result representing the output of the extractor subnetwork is then integrated and input to the third main layer, which comprises power networks with independent subtasks and is connected to the target output. Here, the power networks of the subtask 1, the subtask 2, and the subtask 3 are respectively used as a cost prediction model, a conversion rate prediction model, and a click rate prediction model to respectively output the marginal cost MR, the conversion rate CVR, and the click rate CTR, which can refer to a multi-task learning framework in the prior art and are not described herein again.
The advantages of adopting the multi-task model architecture are as follows: because each sub-model has limited capacity, the connection of the integration stage is sparse and simple, and the final training result can be that each small model is long, and the small models are good at distinguishing clicking and non-clicking passenger groups, good at distinguishing transforming and non-transforming passenger groups, good at distinguishing high-risk and low-risk passenger groups, and have some characteristics which show that the small models are useful for three tasks. Then, in an integration stage, the outputs of the weak models are combined and adapted to different tasks through a weighted fusion mechanism.
As shown in fig. 3, the implementation process of step S11 may include:
in step S31, advertisement data of each advertisement and behavior data of the current user are obtained.
Here, the current user refers to a user who refreshes an information stream provided by the traffic platform through a client. When refreshing information flow, a user can obtain a user identifier and/or a terminal identifier of the user through the flow platform, and behavior data of the user is obtained through the user identifier and/or the terminal identifier.
The behavior data of the user is data within a preset time range, for example, the behavior data of the last week or the last month, or the behavior data of the last half year, and the like, wherein the preset time range can be flexibly adjusted according to actual needs. The behavior data can sufficiently feed back online behaviors of the current user on the traffic platform within a preset time range, such as how many times a certain type of video is watched, which articles are browsed, videos are searched through which keywords, and the like, and the behavior data can analyze information such as gender, age, hobbies, shopping, occupation, activity geographic positions, attention points, and the like.
Wherein, each advertisement is an advertisement in the candidate advertisement set, and the advertisement data of the advertisement may include, but is not limited to, file data and picture data. Wherein, the file data can include but is not limited to advertisement keywords, and the picture data can include but is not limited to pixel size and the like.
Step S32 is to input the behavior data of the current user and the advertisement data of the advertisement to the presentation extractor layer for feature extraction for each advertisement, and to obtain a plurality of feature data.
The plurality of feature data may include a task sharing distinction feature, a distinction between conversion features, a distinction between click features, and a distinction between user risk level features. Here, each feature data may include one or more features, for example, the task sharing discrimination feature data includes a user gender, an age, and the like, and the discrimination feature data includes a behavior feature of z-type video viewed y times in the last x days.
It should be noted that in practical applications, it is difficult to distinguish four groups of feature data, and finally, the contribution degree of each feature to different tasks is controlled by weight at the finest granularity, and the grouping description is only used for facilitating understanding that the adoption of the structure training is helpful to adapt to the multiple tasks described in the present invention.
And step S33, inputting the plurality of feature data into the feature fusion layer, and performing weighted fusion according to the corresponding weight parameters to obtain a plurality of fusion features for click rate prediction, conversion rate prediction and marginal cost rate prediction respectively.
The characteristic fusion layer is preset with a plurality of different weight parameters, each weight parameter comprises a weight value corresponding to a plurality of characteristic data, the plurality of weight parameters respectively correspond to the click rate prediction task, the conversion rate prediction task and the marginal cost rate prediction task, and for convenience of description, the plurality of weight parameters are respectively marked as a first weight parameter, a second weight parameter and a third weight parameter.
Specifically, the plurality of feature data are weighted and summed according to a plurality of weight values in a first weight parameter to obtain a fusion feature for click rate prediction, the plurality of feature data are weighted and summed according to a plurality of weight values in a second weight parameter to obtain a fusion feature for conversion rate prediction, and the plurality of feature data are weighted and summed according to a plurality of weight values in a third weight parameter to obtain a fusion feature for marginal cost rate prediction.
For example, assuming that each weight parameter includes four weights, respectively, and the different weights correspond to different feature data, for example, the four weights in the first weight parameter are a1, a2, a3, and a4, then the fusion feature used for click rate prediction is the feature value a1 of the task sharing discrimination feature + the feature value a2 of whether to convert the feature + the feature value a3 of whether to discriminate the click feature + the feature value a4 of the user risk level feature; the four weight values in the second weight parameter are b1, b2, b3 and b4, then the fused feature used for conversion rate prediction is the feature value of the task sharing discrimination feature b1+ the feature value of whether to convert feature b2+ the feature value of whether to click feature b3+ the feature value of user risk level discrimination feature b4, and the four weight values in the third weight parameter are c1, c2, c3 and c4, then the fused feature used for marginal cost rate prediction is the feature value of the task sharing discrimination feature c1+ the feature value of whether to convert feature c2+ the feature value of whether to click feature c3+ the feature value of user risk level discrimination c 4.
It should be noted that the number of weights in each weight parameter is only an exemplary illustration, and in practical applications, each weight may further include more weights, for example, the weight a1 includes a1…anHere, n is equal to the number of features in the feature data corresponding to the weight a 1.
And step S34, correspondingly inputting the multiple fusion characteristics into the click rate prediction model, the conversion rate prediction model and the cost prediction model to obtain the click rate prediction value, the conversion rate prediction value and the marginal cost rate prediction value of the current user for the advertisement.
Specifically, inputting a plurality of fusion characteristics to a click rate prediction model, and obtaining a click rate predicted value of the current user for the advertisement, which is output by the click rate prediction model; inputting the fusion characteristics into a conversion rate prediction model to obtain a conversion rate predicted value of the current user for the advertisement, which is output by the conversion rate prediction model; and inputting the plurality of fusion characteristics into a click rate prediction model to obtain a marginal cost rate predicted value of the current user for the advertisement, which is output by the cost prediction model.
In one embodiment, as shown in fig. 4, the training process of the multitask model includes:
step S41, a first sample data set is constructed according to the guest group data provided by the traffic platform and the user conversion data provided by the advertiser, wherein the user conversion data comprises positive sample data of advertisement touch and conversion and negative sample data of advertisement touch and non-conversion.
And step S42, taking the click rate prediction task as a source task and the conversion rate prediction task as a target task, and training the representation extractor layer, the click rate prediction model and the conversion rate prediction model in the multi-task model based on the first sample data set to obtain a pre-trained multi-task model.
Step S43, a second sample data set is constructed, wherein the second sample data set is an intersection set of user sample data provided by the advertiser and marked with the cost label and guest group data after collision, and the cost label is normalized cost data.
The second sample data set is an intersection of cost-tagged customers provided by an advertiser and traffic platform customer groups after collision, and on the basis of the multi-task model after pre-training, even if the data volume of the second sample data set is relatively small, a good training effect can be obtained.
And step S44, training the cost prediction model in the pre-trained multitask model based on the second sample data set to obtain the trained multitask model.
Specifically, the implementation process of step S44 may include:
a. and performing iterative training on the cost prediction model in the pre-trained multi-task model based on the second sample data set.
b. In each iterative training, a preset loss function is used for calculating a loss value between a marginal cost rate estimated value output by the cost prediction model and a marginal cost rate true value of a corresponding sample.
Here, the preset loss function may adopt a cross entropy, where the cross entropy is a widely adopted loss function in the neural network loss functions, and the cross entropy is the prior art, and therefore is not described herein again.
c. And when the loss value reaches a preset value, stopping iteratively training the cost prediction model to obtain a trained multi-task model.
In this embodiment, since the relevance between the target task of the simple marginal cost prediction and the click rate prediction task is not high, if the classical MTL architecture is directly used, the shared bottom presentation layer needs to have the capability of accommodating the feature representation satisfying the discrimination between the two subtasks, so that the calculation amount and the training difficulty are indirectly improved, and the sharing efficiency is not high. Therefore, aiming at two subtasks with low relevance degree, the training efficiency of the shared representation layer can be improved by introducing an integrated model (ensemble) idea. That is, as described above, one large shared representation model is decomposed into several parallel small representation models, and then the outputs of the small models are fused by sparse connections before the Tower model. In the combined modeling scene of the advertiser and the traffic platform, CVR prediction is a main task, so that the model architecture which integrates the CTR prediction as a source task and the CVR and MR prediction as target tasks is provided by the invention.
The multi-task model adopts isomorphic transfer learning in transfer learning, namely a source domain and a target domain are the same and are characteristics of a customer group on a flow platform, and the target task is different from the source task. In the embodiment, the CTR prediction task is used as a source task, and the training data requirement of the MR prediction of the target task is reduced. Multi-Task Learning (Multi-Task Learning) belongs to one type of transfer Learning, a target Task and a source Task are placed in the same training Task and are trained by using the same data set, and constraint and assistance can be formed between the tasks, so that the problem of insufficient sample space of the target Task can be solved, and the deviation of an independent Task in sample selection can be avoided.
Optionally, after the trained multitask model is obtained, the trained multitask model may be stored to a cloud server, so as to implement safe storage of the model and use for subsequent online prediction.
In this embodiment, the training of the multi-task model is applicable to a combined modeling scenario, and includes a CVR prediction model training stage and an MR prediction model (i.e., a cost prediction model) training stage, where in the CVR prediction model training stage, implicit learning is performed on the CVR prediction task through the multi-task model by using platform full data, and a specific implementation manner may refer to an ESMM model in the prior art, which is not repeated herein. After the CVR prediction model training stage is completed, the multitask model is equivalent to a sub-model which well initializes an extractor layer and is related to conversion rate and click rate, and the pre-trained multitask model is obtained; in the training stage of the MR prediction model, a structure required by an MR prediction subtask is added to a multitask network structure, and the rest part uses the result of the first training step as an initialization parameter.
It should be noted that the fact Label (Label) information of the MR predictive model is a cost Label of the colliding customers provided by the advertiser, and is a cost other than the cost of the ad placement, the cost Label being normalized cost data, such as insurance claim payment cost.
In a specific embodiment, as shown in fig. 5, the implementation process of step S12 may include:
and step S51, calculating the marginal profit rate estimated value of the current user for each advertisement based on the click rate estimated value, the conversion rate estimated value and the marginal cost rate estimated value of the current user for each advertisement.
Specifically, the process may include:
and calculating the traffic cost of the current user for each advertisement based on the click rate pre-estimated value, the conversion rate pre-estimated value and the bid value of each advertisement of the current user, wherein the traffic cost refers to the advertisement cost required by the current user on a traffic platform, and the bid value of each advertisement can be approximately replaced by the historical average value of an advertiser.
Here, it should be noted that the marginal cost rate estimated value is a probability value between 0 and 1.
For each advertisement, based on the traffic cost and the marginal cost rate estimate of the current user for the advertisement, the marginal rate estimate eMR of the current user for the advertisement is calculated according to the following formula:
eMR＝Revenue-CostAD-CostLabel；
among them, CostLabelMeans the marginal Cost rate estimated value, Cost, of the current user for the advertisementADThe traffic cost of the current user to the advertisement is indicated, and the value of Revenue is 1.
And step S52, calculating thousands of display gains of each advertisement based on the click rate pre-estimated value, the conversion rate pre-estimated value, the marginal profit rate pre-estimated value and the competitive value of each advertisement of the current user.
Specifically, for each advertisement, based on a click rate pre-evaluation value, a conversion rate pre-evaluation value, a marginal rate of return pre-evaluation value and a bid value of the advertisement of a current user, a thousand-time display profit eCPM of the advertisement is calculated according to the following formula:
eCPM＝bid*eCTR*eCVR*eMR；
wherein bid refers to a bid value of the advertisement, eCTR refers to a click rate estimation value of the current user for the advertisement, eCTR refers to a conversion rate estimation value of the current user for the advertisement, and eMR refers to a marginal profitability estimation value of the current user for the advertisement.
Example two
During actual placement, the advertiser's assessment of the effectiveness of the ad placement is typically determined by the return on investment ROI, and the bidding (bid) strategy will be adjusted according to the expected ROI. In this mode, the advertiser's evaluation of the advertising effect can calculate the optimized version ROI based on simple marginal profit weighting by using the following formula, so as to adjust the bidding strategy more precisely:
wherein the oROI is an investment yield prediction value of the advertisement, the Revenue is Revenue of each user to the advertisement, the revnue is equal to actual Revenue after the user is converted, the revnue is equal to 0 when the user is not converted, eMR is a marginal cost rate prediction value of each user to the advertisement after the user is actually converted, and actual _ bids is a competitive value of the advertisement, namely an advertisement investment cost.
Clearly, the marginal rate of return ranges between 0 and 1, which if applied directly to the above formula, would result in a lower estimated CPM for all advertisers using joint modeling to add dynamic cost models than for customers who do not use this factor, which would result in their ads being at a disadvantage in traffic competition. Therefore, eMR needs to be normalized so that the value is greater than 1 for users with marginal profit above average and less than 1 for customers with marginal profit below average, and the cumulative impact of eMR on the large model should be close to 1 overall.
To this end, an embodiment of the present invention further provides an advertisement delivery method, as shown in fig. 6, where the method may include:
and step S61, obtaining a click rate pre-estimated value, a conversion rate pre-estimated value and a marginal cost rate pre-estimated value of the current user for each advertisement in the candidate advertisement set through a pre-trained multitask model.
Specifically, the implementation process of this step may refer to step S11 in the first embodiment, and details are not described here.
Step S62, obtaining the marginal rate of return estimate of the current user for each advertisement based on the click rate estimate, the conversion rate estimate and the marginal cost rate estimate of the current user for each advertisement;
specifically, the implementation process of this step may refer to step S12 in the first embodiment, and details are not described here.
Step S63, obtaining a marginal profitability prediction adjustment coefficient corresponding to each advertisement.
And step S64, calculating thousands of display gains of each advertisement based on the click rate pre-estimated value, the conversion rate pre-estimated value, the marginal rate pre-estimated value, the bid value of each advertisement and the marginal rate pre-estimated value adjusting coefficient corresponding to each advertisement of the current user.
Specifically, for each advertisement, multiplying the click rate pre-estimated value, the conversion rate pre-estimated value, the marginal rate of return pre-estimated value and the bid value of the advertisement of the current user, calculating to obtain a product value corresponding to the advertisement, dividing the product value corresponding to the advertisement by the marginal rate of return pre-estimated value adjustment coefficient corresponding to the advertisement, and calculating to obtain thousands of display benefits of the advertisement.
And step S65, selecting advertisements from the advertisement set according to the thousands of display profits of each advertisement and delivering the advertisements to the current user.
Specifically, the multiple advertisements are ranked from high to low according to respective thousands of display profits, and the first ranked target advertisement is selected from the multiple advertisements and delivered to the current user based on the ranking result.
In a specific embodiment, as shown in fig. 7, the implementation process of step S63 may include:
and step S71, aiming at each sample user in a plurality of sample users, obtaining a click rate pre-estimated value, a conversion rate pre-estimated value and a marginal cost rate pre-estimated value of the sample user for each advertisement in the candidate advertisement set through a pre-trained multitask model.
In this embodiment, after the MR prediction model is built, a sufficiently large sample set can be constructed for all users on the traffic platform, and the marginal cost rate estimates eMR for each person per ad can be calculated.
And step S72, calculating the marginal return rate estimated value of each sample user for each advertisement based on the click rate estimated value, the conversion rate estimated value and the marginal cost rate estimated value of each sample user for each advertisement.
And step S73, respectively averaging the marginal return rate predicted values of all the sample users for each advertisement to obtain a marginal return rate predicted value average value corresponding to each advertisement.
And step S74, respectively determining the average value of the marginal rate of return estimates corresponding to each advertisement as the adjustment coefficient of the marginal rate of return estimates corresponding to each advertisement.
It should be noted that the adjustment coefficient may be calculated by sampling the user at a certain time period.
In summary, the embodiment of the present invention avoids the problem of deviation of the overall oCPM model caused by adding a dynamic cost factor related to a guest group to the oCPM model by obtaining the marginal profitability prediction adjustment coefficient corresponding to each advertisement, and achieves the purpose of constraining the action thereof within a reasonable range.
EXAMPLE III
An embodiment of the present invention provides an advertisement delivery device, as shown in fig. 8, the device may include:
the obtaining module 81 is configured to obtain, through a pre-trained multitask model, a click rate pre-estimated value, a conversion rate pre-estimated value and a marginal cost rate pre-estimated value of a current user for each advertisement in a candidate advertisement set;
a calculating module 82, configured to calculate thousands of display yields of each advertisement based on a click rate pre-estimated value, a conversion rate pre-estimated value, a marginal cost rate pre-estimated value, and a bid value of each advertisement of a current user for each advertisement;
and the releasing module 83 is used for selecting the advertisements from the advertisement set and releasing the advertisements to the current user according to the thousands of display profits of each advertisement.
Further, the multitask model includes a representation extractor layer, a feature fusion layer, a click rate prediction model, a conversion rate prediction model, and a cost prediction model, and the obtaining module 81 is specifically configured to:
acquiring advertisement data of each advertisement and behavior data of a current user;
inputting the behavior data of the current user and the advertisement data of the advertisement into a representation extractor layer for feature extraction aiming at each advertisement to obtain a plurality of feature data;
inputting a plurality of feature data into a feature fusion layer, and performing weighted fusion according to corresponding weight parameters to obtain a plurality of fusion features which are used for click rate prediction, conversion rate prediction and marginal cost rate prediction respectively;
and correspondingly inputting the plurality of fusion characteristics into a click rate prediction model, a conversion rate prediction model and a cost prediction model to obtain a click rate prediction value, a conversion rate prediction value and a marginal cost rate prediction value of the current user for the advertisement.
Further, the plurality of feature data includes a task sharing feature, a distinguishing whether to translate feature, a distinguishing whether to click feature, and a distinguishing user risk level feature.
Further, the apparatus further comprises a training module, the training module comprising:
the first construction submodule is used for constructing a first sample data set according to the guest group data provided by the traffic platform and the user conversion data provided by the advertiser, wherein the user conversion data comprises positive sample data of advertisement touch and conversion and negative sample data of advertisement touch and non-conversion;
the first training submodule is used for training a representation extractor layer, a click rate prediction model and a conversion rate prediction model in a multi-task model on the basis of a first sample data set by taking a click rate prediction task as a source task and taking a conversion rate prediction task as a target task to obtain a pre-trained multi-task model;
the second construction submodule is used for constructing a second sample data set, wherein the second sample data set is an intersection of user sample data which is provided by an advertiser and is marked with a cost tag and guest group data after collision, and the cost tag is normalized cost data;
and the second training submodule is used for training the cost prediction model in the pre-trained multi-task model based on the second sample data set to obtain the trained multi-task model.
Further, the second training submodule is specifically configured to:
performing iterative training on a cost prediction model in the pre-trained multi-task model based on a second sample data set;
in each iterative training, calculating a loss value between a marginal cost rate estimated value output by the cost prediction model and a marginal cost rate true value of a corresponding sample by using a preset loss function;
and when the loss value reaches a preset value, stopping iteratively training the cost prediction model to obtain a trained multi-task model.
Further, the calculation module 82 includes:
the first calculation submodule is used for calculating the marginal profit rate predicted value of the current user for each advertisement based on the click rate predicted value, the conversion rate predicted value and the marginal cost rate predicted value of the current user for each advertisement;
and the second calculation submodule is used for calculating thousands of display gains of each advertisement based on the click rate pre-estimated value, the conversion rate pre-estimated value, the marginal rate pre-estimated value and the competitive value of each advertisement of the current user.
Further, the device also comprises an acquisition module;
the acquisition module is used for acquiring the marginal rate of return prediction value adjustment coefficient corresponding to each advertisement;
and the second calculation submodule is also used for calculating thousands of display gains of each advertisement based on the click rate pre-estimated value, the conversion rate pre-estimated value, the marginal rate pre-estimated value, the bid value of each advertisement and the marginal rate pre-estimated value adjustment coefficient corresponding to each advertisement of the current user.
Further, the obtaining module is specifically configured to:
aiming at each sample user in a plurality of sample users, obtaining a click rate pre-estimated value, a conversion rate pre-estimated value and a marginal cost rate pre-estimated value of each advertisement in a candidate advertisement set by the sample user through a pre-trained multi-task model;
calculating a marginal rate of return estimate of each sample user for each advertisement based on a click rate estimate, a conversion rate estimate and a marginal cost rate estimate of each sample user for each advertisement;
respectively averaging the marginal rate of return predicted values of all sample users for each advertisement to obtain a marginal rate of return predicted value average value corresponding to each advertisement;
and respectively determining the average value of the marginal rate of return prediction values corresponding to each advertisement as the adjustment coefficient of the marginal rate of return prediction values corresponding to each advertisement.
It should be noted that: in the advertisement delivery device provided in this embodiment, only the division of the functional modules is exemplified, and in practical applications, the functions may be distributed by different functional modules according to needs, that is, the internal structure of the device is divided into different functional modules, so as to complete all or part of the functions described above. In addition, the advertisement delivery device of the present embodiment and the advertisement delivery method embodiment in the above embodiments belong to the same concept, and specific implementation processes and beneficial effects thereof are described in detail in the advertisement delivery method embodiment and are not described herein again.
Fig. 9 is an internal structural diagram of a computer device according to an embodiment of the present invention. The computer device may be a server, and its internal structure diagram may be as shown in fig. 9. The computer device includes a processor, a memory, and a network interface connected by a system bus. Wherein the processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device comprises a nonvolatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of an operating system and computer programs in the non-volatile storage medium. The network interface of the computer device is used for communicating with an external terminal through a network connection. The computer program is executed by a processor to implement a method of advertisement delivery.
Those skilled in the art will appreciate that the configuration shown in fig. 9 is a block diagram of only a portion of the configuration associated with aspects of the present invention and is not intended to limit the computing devices to which aspects of the present invention may be applied, and that a particular computing device may include more or less components than those shown, or may combine certain components, or have a different arrangement of components.
In one embodiment, a computer device is provided, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, the processor implementing the following steps when executing the computer program:
obtaining click rate pre-estimated values, conversion rate pre-estimated values and marginal cost rate pre-estimated values of the current user for all the advertisements in the candidate advertisement set through a pre-trained multi-task model;
calculating thousands of display gains of each advertisement based on a click rate pre-estimated value, a conversion rate pre-estimated value, a marginal cost rate pre-estimated value and a competitive value of each advertisement of a current user;
and selecting advertisements from the advertisement set according to the thousands of display profits of each advertisement and putting the advertisements to the current user.
In one embodiment, a computer-readable storage medium is provided, having a computer program stored thereon, which when executed by a processor, performs the steps of:
obtaining click rate pre-estimated values, conversion rate pre-estimated values and marginal cost rate pre-estimated values of the current user for all the advertisements in the candidate advertisement set through a pre-trained multi-task model;
calculating thousands of display gains of each advertisement based on a click rate pre-estimated value, a conversion rate pre-estimated value, a marginal cost rate pre-estimated value and a competitive value of each advertisement of a current user;
and selecting advertisements from the advertisement set according to the thousands of display profits of each advertisement and putting the advertisements to the current user.
It will be understood by those skilled in the art that all or part of the processes of the methods of the embodiments described above can be implemented by hardware instructions of a computer program, which can be stored in a non-volatile computer-readable storage medium, and when executed, can include the processes of the embodiments of the methods described above. Any reference to memory, storage, databases, or other media used in embodiments provided herein may include non-volatile and/or volatile memory. Non-volatile memory can include read-only memory (ROM), Programmable ROM (PROM), Electrically Programmable ROM (EPROM), Electrically Erasable Programmable ROM (EEPROM), or flash memory. Volatile memory can include Random Access Memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms such as Static RAM (SRAM), Dynamic RAM (DRAM), Synchronous DRAM (SDRAM), Double Data Rate SDRAM (DDRSDRAM), Enhanced SDRAM (ESDRAM), Synchronous Link DRAM (SLDRAM), Rambus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).
The technical features of the above embodiments can be arbitrarily combined, and for the sake of brevity, all possible combinations of the technical features in the above embodiments are not described, but should be considered as the scope of the present specification as long as there is no contradiction between the combinations of the technical features.
The above-mentioned embodiments only express several embodiments of the present invention, and the description thereof is more specific and detailed, but not construed as limiting the scope of the invention. It should be noted that, for a person skilled in the art, several variations and modifications can be made without departing from the inventive concept, which falls within the scope of the present invention. Therefore, the protection scope of the present patent shall be subject to the appended claims.
Claims (10)
1. An advertisement delivery method, the method comprising:
obtaining click rate pre-estimated values, conversion rate pre-estimated values and marginal cost rate pre-estimated values of the current user for all the advertisements in the candidate advertisement set through a pre-trained multi-task model;
calculating thousands of display gains of each advertisement based on click rate pre-estimated values, conversion rate pre-estimated values and marginal cost rate pre-estimated values of the current user for each advertisement and the competitive value of each advertisement;
and selecting advertisements from the advertisement set according to the thousands of display profits of the advertisements and delivering the advertisements to the current user.
2. The method of claim 1, wherein the multitask model comprises a representation extractor layer, a feature fusion layer, a click-through rate prediction model, a conversion rate prediction model and a cost prediction model, and the obtaining of the click-through rate prediction value, the conversion rate prediction value and the marginal cost rate prediction value of the current user for each advertisement in the candidate advertisement set through a pre-trained multitask model comprises:
acquiring advertisement data of each advertisement and behavior data of the current user;
for each advertisement, inputting the behavior data of the current user and the advertisement data of the advertisement into the representation extractor layer for feature extraction to obtain a plurality of feature data;
inputting the plurality of feature data into the feature fusion layer to be respectively subjected to weighted fusion according to corresponding weight parameters to obtain a plurality of fusion features which are respectively used for click rate prediction, conversion rate prediction and marginal cost rate prediction;
and correspondingly inputting the plurality of fusion characteristics into the click rate prediction model, the conversion rate prediction model and the cost prediction model to obtain a click rate prediction value, a conversion rate prediction value and a marginal cost rate prediction value of the current user for the advertisement.
3. The method of claim 2, wherein the training process of the multitask model comprises:
constructing a first sample data set according to guest group data provided by a traffic platform and user conversion data provided by an advertiser, wherein the user conversion data comprises positive sample data of advertisement touch and conversion and negative sample data of advertisement touch and conversion;
taking a click rate prediction task as a source task and a conversion rate prediction task as a target task, and training the representation extractor layer, the click rate prediction model and the conversion rate prediction model in the multi-task model based on the first sample data set to obtain the pre-trained multi-task model;
constructing a second sample data set, wherein the second sample data set is an intersection of user sample data which is provided by an advertiser and labeled with a cost tag and the guest group data after collision, and the cost tag is normalized cost data;
and training the cost prediction model in the pre-trained multitask model based on the second sample data set to obtain the trained multitask model.
4. The method of claim 3, wherein training the cost prediction model in the pre-trained multitask model based on the second sample data set to obtain the trained multitask model comprises:
iteratively training the cost prediction model in the pre-trained multitask model based on the second sample data set;
in each iterative training, calculating a loss value between a marginal cost rate estimated value output by the cost prediction model and a marginal cost rate true value of a corresponding sample by using a preset loss function;
and stopping iterative training of the cost prediction model when the loss value reaches a preset value, so as to obtain the trained multi-task model.
5. The method of any one of claims 1 to 4, wherein said calculating the thousand showing returns of each of said advertisements based on the click-through rate estimate, the conversion rate estimate, the marginal cost rate estimate and the bid value of each of said advertisements of said current user comprises:
calculating a marginal rate of return estimate of the current user for each advertisement based on a click rate estimate, a conversion rate estimate and a marginal cost rate estimate of the current user for each advertisement;
and calculating thousands of display gains of each advertisement based on the click rate pre-estimated value, the conversion rate pre-estimated value, the marginal rate pre-estimated value and the competitive value of each advertisement of the current user.
6. The method of claim 5, further comprising:
acquiring a marginal yield pre-estimation value adjusting coefficient corresponding to each advertisement;
calculating thousand display gains of each advertisement based on the click rate pre-estimated value, the conversion rate pre-estimated value, the marginal rate pre-estimated value and the competitive value of each advertisement of the current user, wherein the calculation comprises the following steps:
and calculating thousands of display gains of each advertisement based on the click rate pre-estimated value, the conversion rate pre-estimated value, the marginal rate pre-estimated value, the bid value of each advertisement and the marginal rate pre-estimated value adjusting coefficient corresponding to each advertisement of the current user.
7. The method of claim 6, wherein obtaining the adjustment coefficient of the marginal rate of return estimate corresponding to each of the advertisements comprises:
aiming at each sample user in a plurality of sample users, obtaining a click rate estimated value, a conversion rate estimated value and a marginal cost rate estimated value of the sample user for each advertisement in the candidate advertisement set through the pre-trained multitask model;
calculating a marginal rate-of-return estimate for each of the advertisements for each of the sample users based on the click-through rate estimate, the conversion rate estimate, and the marginal cost rate estimate for each of the advertisements for each of the sample users;
respectively averaging the marginal rate of return predicted values of all the sample users for each advertisement to obtain a marginal rate of return predicted value average value corresponding to each advertisement;
and respectively determining the average value of the marginal rate of return estimates corresponding to each advertisement as the adjustment coefficient of the marginal rate of return estimates corresponding to each advertisement.
8. An advertising device, the device comprising:
the acquisition module is used for acquiring click rate pre-estimated values, conversion rate pre-estimated values and marginal cost rate pre-estimated values of the current user for all the advertisements in the candidate advertisement set through a pre-trained multi-task model;
the calculation module is used for calculating thousand-time display income of each advertisement based on the click rate pre-estimated value, the conversion rate pre-estimated value and the marginal cost rate pre-estimated value of the current user for each advertisement and the competitive value of each advertisement;
and the releasing module is used for selecting the advertisements from the advertisement set and releasing the advertisements to the current user according to the thousands of display profits of the advertisements.
9. A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the method of advertisement delivery according to any one of claims 1 to 7 when executing the computer program.
10. A computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, carries out the advertisement delivery method according to any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010002156.9A CN111178981B (en) | 2020-01-02 | 2020-01-02 | Advertisement putting method and device, computer equipment and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010002156.9A CN111178981B (en) | 2020-01-02 | 2020-01-02 | Advertisement putting method and device, computer equipment and storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111178981A true CN111178981A (en) | 2020-05-19 |
| CN111178981B CN111178981B (en) | 2023-11-28 |
Family
ID=70657804
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010002156.9A Active CN111178981B (en) | 2020-01-02 | 2020-01-02 | Advertisement putting method and device, computer equipment and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111178981B (en) |
Cited By (20)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111815373A (en) * | 2020-09-02 | 2020-10-23 | 北京悠易网际科技发展有限公司 | Conversion cost optimization method and device, electronic equipment and storage medium |
| CN111985971A (en) * | 2020-08-25 | 2020-11-24 | 北京达佳互联信息技术有限公司 | Advertisement screening method, device, equipment and storage medium |
| CN112150219A (en) * | 2020-10-22 | 2020-12-29 | 广州万唯邑众信息科技有限公司 | Online advertisement delivery system and use method thereof |
| CN112288146A (en) * | 2020-10-15 | 2021-01-29 | 北京沃东天骏信息技术有限公司 | Page display method, device, system, computer equipment and storage medium |
| CN112561589A (en) * | 2020-12-18 | 2021-03-26 | 成都新潮传媒集团有限公司 | Advertisement putting method, device and storage medium |
| CN113269595A (en) * | 2021-06-23 | 2021-08-17 | 南京网觉软件有限公司 | Advertisement delivery method and system based on real-time prediction ROI |
| CN113538020A (en) * | 2021-07-05 | 2021-10-22 | 深圳索信达数据技术有限公司 | Method and device for acquiring guest group feature association degree, storage medium and electronic device |
| CN113763035A (en) * | 2021-08-12 | 2021-12-07 | 苏宁易购集团股份有限公司 | Advertisement delivery effect prediction method and device, computer equipment and storage medium |
| CN113807888A (en) * | 2021-09-02 | 2021-12-17 | 支付宝(杭州)信息技术有限公司 | Marketing processing method and device |
| CN113822688A (en) * | 2020-06-23 | 2021-12-21 | 北京沃东天骏信息技术有限公司 | Advertisement conversion rate estimation method and device, storage medium and electronic equipment |
| CN113837811A (en) * | 2021-09-30 | 2021-12-24 | 成都新潮传媒集团有限公司 | Elevator advertisement point recommendation method and device, computer equipment and storage medium |
| CN113888233A (en) * | 2021-10-22 | 2022-01-04 | 北京明略软件系统有限公司 | Advertisement budget allocation method, system, equipment and storage medium based on time effect |
| CN114663155A (en) * | 2022-04-01 | 2022-06-24 | 广州华多网络科技有限公司 | Advertisement putting and selecting method and device, equipment, medium and product thereof |
| CN114862447A (en) * | 2022-04-08 | 2022-08-05 | 无线生活（杭州）信息科技有限公司 | Advertisement reordering method and device |
| WO2023116138A1 (en) * | 2021-12-21 | 2023-06-29 | 北京有竹居网络技术有限公司 | Modeling method for multi-task model, promotional content processing method, and related apparatuses |
| CN116485477A (en) * | 2023-06-19 | 2023-07-25 | 北京大学 | Advertisement valuation processing method and device and electronic equipment |
| CN117217828A (en) * | 2023-08-18 | 2023-12-12 | 上海数禾信息科技有限公司 | Method, device, computer equipment and storage medium for verifying conversion return data |
| CN117237022A (en) * | 2023-09-22 | 2023-12-15 | 杭州面朝信息科技有限公司 | Advertisement putting method and device based on social media account business value evaluation |
| CN117350792A (en) * | 2023-12-06 | 2024-01-05 | 南京智慧交通信息股份有限公司 | Adaptive advertisement putting method and system based on multitask learning |
| CN118410056A (en) * | 2024-07-01 | 2024-07-30 | 南通云尚找家纺电子商务有限公司 | Intelligent advertisement making feature library data updating system |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107657486A (en) * | 2017-10-19 | 2018-02-02 | 厦门美柚信息科技有限公司 | A kind of advertisement placement method and device |
| CN107977859A (en) * | 2017-11-14 | 2018-05-01 | 广州优视网络科技有限公司 | Advertisement placement method, device, computing device and storage medium |
| CN108280682A (en) * | 2018-01-16 | 2018-07-13 | 深圳市和讯华谷信息技术有限公司 | Advertisement placement method, terminal and computer readable storage medium |
| CN109447688A (en) * | 2018-09-27 | 2019-03-08 | 平安科技（深圳）有限公司 | Advertisement placement method, device and storage equipment, computer equipment |
| CN110033314A (en) * | 2019-03-18 | 2019-07-19 | 北京品友互动信息技术股份公司 | Advertisement data processing method and device |
| CN110570232A (en) * | 2019-08-05 | 2019-12-13 | 科大讯飞股份有限公司 | Internet advertisement putting method, device, server and storage medium |
-
2020
- 2020-01-02 CN CN202010002156.9A patent/CN111178981B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107657486A (en) * | 2017-10-19 | 2018-02-02 | 厦门美柚信息科技有限公司 | A kind of advertisement placement method and device |
| CN107977859A (en) * | 2017-11-14 | 2018-05-01 | 广州优视网络科技有限公司 | Advertisement placement method, device, computing device and storage medium |
| CN108280682A (en) * | 2018-01-16 | 2018-07-13 | 深圳市和讯华谷信息技术有限公司 | Advertisement placement method, terminal and computer readable storage medium |
| CN109447688A (en) * | 2018-09-27 | 2019-03-08 | 平安科技（深圳）有限公司 | Advertisement placement method, device and storage equipment, computer equipment |
| CN110033314A (en) * | 2019-03-18 | 2019-07-19 | 北京品友互动信息技术股份公司 | Advertisement data processing method and device |
| CN110570232A (en) * | 2019-08-05 | 2019-12-13 | 科大讯飞股份有限公司 | Internet advertisement putting method, device, server and storage medium |
Cited By (28)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113822688A (en) * | 2020-06-23 | 2021-12-21 | 北京沃东天骏信息技术有限公司 | Advertisement conversion rate estimation method and device, storage medium and electronic equipment |
| CN111985971A (en) * | 2020-08-25 | 2020-11-24 | 北京达佳互联信息技术有限公司 | Advertisement screening method, device, equipment and storage medium |
| CN111815373A (en) * | 2020-09-02 | 2020-10-23 | 北京悠易网际科技发展有限公司 | Conversion cost optimization method and device, electronic equipment and storage medium |
| CN112288146A (en) * | 2020-10-15 | 2021-01-29 | 北京沃东天骏信息技术有限公司 | Page display method, device, system, computer equipment and storage medium |
| CN112150219A (en) * | 2020-10-22 | 2020-12-29 | 广州万唯邑众信息科技有限公司 | Online advertisement delivery system and use method thereof |
| CN112561589A (en) * | 2020-12-18 | 2021-03-26 | 成都新潮传媒集团有限公司 | Advertisement putting method, device and storage medium |
| CN112561589B (en) * | 2020-12-18 | 2023-01-10 | 成都屏盟科技有限公司 | Advertisement delivery method, device and storage medium |
| CN113269595B (en) * | 2021-06-23 | 2022-01-11 | 南京网觉软件有限公司 | Advertisement delivery method and system based on real-time prediction ROI |
| CN113269595A (en) * | 2021-06-23 | 2021-08-17 | 南京网觉软件有限公司 | Advertisement delivery method and system based on real-time prediction ROI |
| CN113538020B (en) * | 2021-07-05 | 2024-03-26 | 深圳索信达数据技术有限公司 | Method and device for acquiring association degree of group of people features, storage medium and electronic device |
| CN113538020A (en) * | 2021-07-05 | 2021-10-22 | 深圳索信达数据技术有限公司 | Method and device for acquiring guest group feature association degree, storage medium and electronic device |
| CN113763035A (en) * | 2021-08-12 | 2021-12-07 | 苏宁易购集团股份有限公司 | Advertisement delivery effect prediction method and device, computer equipment and storage medium |
| CN113807888A (en) * | 2021-09-02 | 2021-12-17 | 支付宝(杭州)信息技术有限公司 | Marketing processing method and device |
| CN113807888B (en) * | 2021-09-02 | 2024-06-04 | 支付宝(杭州)信息技术有限公司 | Marketing processing method and device |
| CN113837811A (en) * | 2021-09-30 | 2021-12-24 | 成都新潮传媒集团有限公司 | Elevator advertisement point recommendation method and device, computer equipment and storage medium |
| CN113837811B (en) * | 2021-09-30 | 2023-10-10 | 成都屏盟科技有限公司 | Elevator advertisement point position recommending method and device, computer equipment and storage medium |
| CN113888233A (en) * | 2021-10-22 | 2022-01-04 | 北京明略软件系统有限公司 | Advertisement budget allocation method, system, equipment and storage medium based on time effect |
| WO2023116138A1 (en) * | 2021-12-21 | 2023-06-29 | 北京有竹居网络技术有限公司 | Modeling method for multi-task model, promotional content processing method, and related apparatuses |
| CN114663155A (en) * | 2022-04-01 | 2022-06-24 | 广州华多网络科技有限公司 | Advertisement putting and selecting method and device, equipment, medium and product thereof |
| CN114862447A (en) * | 2022-04-08 | 2022-08-05 | 无线生活（杭州）信息科技有限公司 | Advertisement reordering method and device |
| CN116485477B (en) * | 2023-06-19 | 2023-11-10 | 北京大学 | Advertisement valuation processing method and device and electronic equipment |
| CN116485477A (en) * | 2023-06-19 | 2023-07-25 | 北京大学 | Advertisement valuation processing method and device and electronic equipment |
| CN117217828A (en) * | 2023-08-18 | 2023-12-12 | 上海数禾信息科技有限公司 | Method, device, computer equipment and storage medium for verifying conversion return data |
| CN117217828B (en) * | 2023-08-18 | 2024-05-10 | 上海数禾信息科技有限公司 | Method, device, computer equipment and storage medium for verifying conversion return data |
| CN117237022A (en) * | 2023-09-22 | 2023-12-15 | 杭州面朝信息科技有限公司 | Advertisement putting method and device based on social media account business value evaluation |
| CN117350792A (en) * | 2023-12-06 | 2024-01-05 | 南京智慧交通信息股份有限公司 | Adaptive advertisement putting method and system based on multitask learning |
| CN117350792B (en) * | 2023-12-06 | 2024-03-08 | 南京智慧交通信息股份有限公司 | Adaptive advertisement putting method and system based on multitask learning |
| CN118410056A (en) * | 2024-07-01 | 2024-07-30 | 南通云尚找家纺电子商务有限公司 | Intelligent advertisement making feature library data updating system |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111178981B (en) | 2023-11-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111178981B (en) | Advertisement putting method and device, computer equipment and storage medium | |
| US12039564B2 (en) | Method and system for generation of at least one output analytic for a promotion | |
| Tang et al. | Automatic ad format selection via contextual bandits | |
| US11288709B2 (en) | Training and utilizing multi-phase learning models to provide digital content to client devices in a real-time digital bidding environment | |
| CN109345302A (en) | Machine learning model training method, device, storage medium and computer equipment | |
| CN111160983A (en) | Advertisement putting effect evaluation method and device, computer equipment and storage medium | |
| CN111340244B (en) | Prediction method, training method, device, server and medium | |
| US20220076299A1 (en) | Intelligent electronic advertisement generation and distribution | |
| CN111095330B (en) | Machine learning method and system for predicting online user interactions | |
| Ye et al. | Cold start to improve market thickness on online advertising platforms: Data-driven algorithms and field experiments | |
| CN111798280B (en) | Multimedia information recommendation method, device and equipment and storage medium | |
| WO2013067211A1 (en) | Individual-level modeling | |
| CN111052167A (en) | Method and system for intelligent adaptive bidding in automated online trading network | |
| US20160140577A1 (en) | Unified marketing model based on conduit variables | |
| CN111667024A (en) | Content pushing method and device, computer equipment and storage medium | |
| US10855786B1 (en) | Optimizing value of content items delivered for a content provider | |
| CN116109354A (en) | Content recommendation method, apparatus, device, storage medium, and computer program product | |
| CN113159809B (en) | Object processing method and device, electronic equipment and computer readable storage medium | |
| US20220374923A1 (en) | Computational methods and processor systems for predictive marketing analysis | |
| US20150339693A1 (en) | Determination of initial value for automated delivery of news items | |
| US11403668B2 (en) | Multitask transfer learning for optimization of targeted promotional programs | |
| Shariat et al. | Online evaluation of bid prediction models in a large-scale computational advertising platform: decision making and insights | |
| CN111899049B (en) | Advertisement putting method, device and equipment | |
| Kini et al. | Revenue maximization using multitask learning for promotion recommendation | |
| US20240289732A1 (en) | Method, apparatus, device and storage medium for delivery cost estimation |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |