CN113706211B - Advertisement click rate prediction method and system based on neural network - Google Patents
Advertisement click rate prediction method and system based on neural network Download PDFInfo
- Publication number
- CN113706211B CN113706211B CN202111016975.XA CN202111016975A CN113706211B CN 113706211 B CN113706211 B CN 113706211B CN 202111016975 A CN202111016975 A CN 202111016975A CN 113706211 B CN113706211 B CN 113706211B
- Authority
- CN
- China
- Prior art keywords
- advertisement
- user
- features
- target
- sample
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 71
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 37
- 238000012545 processing Methods 0.000 claims abstract description 40
- 238000007499 fusion processing Methods 0.000 claims abstract description 14
- 238000012549 training Methods 0.000 claims abstract description 14
- 230000004927 fusion Effects 0.000 claims description 40
- 230000006870 function Effects 0.000 claims description 21
- 238000013507 mapping Methods 0.000 claims description 19
- 238000004590 computer program Methods 0.000 claims description 18
- 238000004422 calculation algorithm Methods 0.000 claims description 14
- 238000005070 sampling Methods 0.000 claims description 11
- 238000003860 storage Methods 0.000 claims description 11
- 230000004913 activation Effects 0.000 claims description 10
- 238000004364 calculation method Methods 0.000 claims description 10
- 238000011176 pooling Methods 0.000 claims description 10
- 239000011159 matrix material Substances 0.000 claims description 8
- 238000005516 engineering process Methods 0.000 abstract description 21
- 238000013473 artificial intelligence Methods 0.000 abstract description 8
- 239000013598 vector Substances 0.000 description 23
- 230000008569 process Effects 0.000 description 22
- 238000012384 transportation and delivery Methods 0.000 description 14
- 238000010586 diagram Methods 0.000 description 11
- 230000008901 benefit Effects 0.000 description 6
- 239000000463 material Substances 0.000 description 6
- 238000003062 neural network model Methods 0.000 description 6
- 238000009826 distribution Methods 0.000 description 5
- 230000000694 effects Effects 0.000 description 5
- 238000007726 management method Methods 0.000 description 5
- 230000003993 interaction Effects 0.000 description 4
- 238000010606 normalization Methods 0.000 description 4
- RTZKZFJDLAIYFH-UHFFFAOYSA-N Diethyl ether Chemical compound CCOCC RTZKZFJDLAIYFH-UHFFFAOYSA-N 0.000 description 3
- 238000006243 chemical reaction Methods 0.000 description 3
- 238000005457 optimization Methods 0.000 description 3
- 238000004891 communication Methods 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 238000004880 explosion Methods 0.000 description 2
- 238000004519 manufacturing process Methods 0.000 description 2
- 230000001360 synchronised effect Effects 0.000 description 2
- 238000012935 Averaging Methods 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 238000007405 data analysis Methods 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 230000000977 initiatory effect Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003058 natural language processing Methods 0.000 description 1
- 238000004806 packaging method and process Methods 0.000 description 1
- 238000013439 planning Methods 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000003786 synthesis reaction Methods 0.000 description 1
- 230000001960 triggered effect Effects 0.000 description 1
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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- 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
- G06N3/045—Combinations of 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
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- 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
- G06Q30/0249—Advertisements based upon budgets or funds
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Strategic Management (AREA)
- Finance (AREA)
- Development Economics (AREA)
- Accounting & Taxation (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Economics (AREA)
- Molecular Biology (AREA)
- General Business, Economics & Management (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Marketing (AREA)
- Game Theory and Decision Science (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Entrepreneurship & Innovation (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention relates to an artificial intelligence technology, and provides an advertisement click rate prediction method and system based on a neural network, wherein the method comprises the following steps: acquiring historical advertisement characteristics and attribute characteristics of a target user, wherein the historical advertisement characteristics comprise basic information of the target user and historical purchase information between the target user and a target product, and the target advertisement is used for popularizing the target product; and inputting the historical advertisement features and the attribute features into a click prediction model, sequentially carrying out embedding dense processing and cross fusion processing, predicting the click rate of the target user on the target advertisement, and training the click prediction model to obtain a user sample, an advertisement sample and a score label. According to the embodiment of the invention, the user characteristics and the advertisement characteristics are fused, the user information and the advertisement information are mined from different levels, the problem that the prediction accuracy is reduced due to the lack of high-dimensional sparse characteristic data in the prior art is solved, and the prediction accuracy of the user on the advertisement is improved.
Description
Technical Field
The invention relates to the field of artificial intelligence, in particular to an advertisement click rate prediction method and system based on a neural network.
Background
With the rapid popularization and development of the internet, online advertising and popularization of products by using the internet has become a very important marketing mode. The advertisement click rate is the ratio of the advertisement click times to the advertisement display times, and the higher the advertisement click rate is, the more interested the user is in the advertisement, and the accurate prediction of the advertisement click rate is an important means for realizing accurate marketing of products, and is also an important index for reflecting the advertisement putting effect and evaluating the advertisement conversion rate. Therefore, the realization of the correct prediction of the click rate of the advertisement has very important practical significance.
The existing advertisement click prediction basically inputs detected data into a deep learning model to carry out probability prediction, the probability prediction is often related to a specific structure of deep learning, the model framework basically needs to map low-dimensional features into a high-dimensional space, data is always lost and inappropriately processed in the processing process, and the comprehensiveness of data feature information cannot be met, so that the reliability of advertisement click rate prediction is greatly reduced.
Therefore, there is a need for a highly reliable advertisement click rate prediction method.
Disclosure of Invention
The invention provides an advertisement click rate prediction method and system based on a neural network, which mainly aim to improve the prediction precision of advertisement click rate so that a user can reasonably plan advertisement putting strategies.
In a first aspect, an embodiment of the present invention provides a method for predicting an advertisement click rate based on a neural network, including:
acquiring historical advertisement characteristics of a target user and attribute characteristics of the target advertisement, wherein the historical advertisement characteristics comprise basic information of the target user and historical purchase information between the target user and a target product, and the attribute characteristics are acquired according to the basic information of the target advertisement;
and inputting the historical advertisement features and the attribute features into a click prediction model, respectively carrying out embedding dense processing on the historical advertisement features and the attribute features in sequence, carrying out cross fusion processing on the processed historical advertisement features and the processed attribute features, predicting the click rate of the target user on the target advertisement, and training a user sample, an advertisement sample and a score label by the click prediction model.
The step of inputting the historical advertisement features and the attribute features into a click prediction model, respectively carrying out embedding dense processing on the historical advertisement features and the attribute features in sequence, and carrying out cross fusion processing on the processed historical advertisement features and the processed attribute features to predict the click rate of the target user on the target advertisement, wherein the method specifically comprises the following steps:
Respectively carrying out embedding dense processing on the historical advertisement features and the attribute features to respectively obtain user embedding features and advertisement embedding features;
cross fusion and average pooling are carried out on the user embedded features and the advertisement embedded features, and fusion features are obtained;
and predicting the click rate of the target user on the target advertisement according to the fusion characteristics.
Preferably, the embedding density processing is performed on the historical advertisement feature and the attribute feature respectively to obtain a user embedded feature and an advertisement embedded feature, which includes:
mapping the high-dimensional sparse features in the historical advertisement features into user dense features; and/or mapping the high-dimensional sparse features in the attribute features of the target advertisement to advertisement dense features;
normalizing the user continuous features in the historical advertisement features to obtain normalized user continuous features; and/or normalizing the advertisement continuous features in the attribute features to obtain normalized advertisement continuous features;
splicing the user dense features and the normalized user continuous features to obtain user spliced features; and/or splicing the advertisement dense features and the normalized advertisement continuous features to obtain advertisement splicing features;
And respectively carrying out multi-layer embedding on the user splicing characteristics and/or the advertisement splicing characteristics to obtain user embedding characteristics and advertisement embedding characteristics.
Preferably, the multi-layer embedding is performed on the user splicing feature and/or the advertisement splicing feature, so as to obtain a user embedding feature and the advertisement embedding feature, and a specific calculation formula is as follows:
wherein x is user Representing the user splice characteristics, x item Representing the characteristics of the splicing of the advertisements,representing the ith element in the user-embedded feature,/->Representing the i-th element in the advertisement embedding feature,/->Representing the ith element of the preset user's embedded matrix,/->Represents the ith element in the preset advertisement embedding matrix, and m represents the number of embedded subspaces.
Preferably, the predicting the click rate of the target user on the target advertisement according to the fusion feature includes:
and inputting the fusion characteristics into a full-connection layer, and obtaining the click rate through an activation function by outputting the full-connection layer.
The method for acquiring the user sample and the advertisement sample comprises the following steps:
acquiring an initial user sample and an initial advertisement sample;
and respectively processing the initial user sample and the initial advertisement sample by adopting a preset rule to obtain the user sample and the advertisement sample, wherein the preset rule comprises filling in missing values of a nearest neighbor algorithm and sampling in layers.
Preferably, the method further comprises:
acquiring a comprehensive score of the target advertisement according to the click rate and the putting price of the target advertisement;
and judging whether to throw the target advertisement according to the comprehensive score.
In a second aspect, an embodiment of the present invention provides an advertisement click rate prediction system based on a neural network, including:
the characteristic acquisition module is used for acquiring historical advertisement characteristics of a target user and attribute characteristics of the target advertisement, wherein the historical advertisement characteristics comprise basic information of the target user and historical purchase information between the target user and a target product, and the attribute characteristics are acquired according to the basic information of the target advertisement;
the prediction module is used for inputting the historical advertisement features and the attribute features into a click prediction model, respectively embedding the historical advertisement features and the attribute features into a dense process in sequence, carrying out cross fusion processing on the processed historical advertisement features and the processed attribute features, and predicting the click rate of the target user on the target advertisement, wherein the click prediction model is obtained by training a user sample, an advertisement sample and a score label.
In a third aspect, an embodiment of the present invention provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, where the steps of the above-mentioned neural network-based advertisement click rate prediction method are implemented when the processor executes the computer program.
In a fourth aspect, an embodiment of the present invention provides a computer storage medium storing a computer program, where the computer program when executed by a processor implements the steps of the above-described neural network-based advertisement click rate prediction method.
According to the advertisement click rate prediction method and system based on the neural network, the neural network model is adopted to conduct embedding dense processing on historical advertisement features and attribute features, the user embedded features and the advertisement embedded features represented by dense vectors are obtained through the embedding dense processing, and samples of most dimensions of the dense vectors are effective relative to high-dimensional sparse vectors, so that the problem that prediction accuracy is reduced due to high-dimensional sparse vector data loss in the prior art is solved; on the basis of carrying out dense processing on the historical advertisement features and the attribute features of the user independently, the user features and the advertisement features are fused and embedded in multiple layers, and the information of the user and the advertisement can be mined from different layers, so that the comprehensiveness of data processing is improved, the user and the advertisement information are fully fused, and the prediction precision of the user on the advertisement is further improved.
Drawings
FIG. 1 is an application scenario diagram of an advertisement click rate prediction method based on a neural network according to an embodiment of the present invention;
FIG. 2 is a flowchart of an advertisement click rate prediction method based on a neural network according to an embodiment of the present invention;
FIG. 3 is a schematic diagram illustrating a structure of a click prediction model according to an embodiment of the present invention;
FIG. 4 is a block diagram of an advertisement bidding delivery system based on a neural network-based advertisement click rate prediction method according to an embodiment of the present invention;
FIG. 5 is a schematic diagram of a system for predicting advertisement click rate based on a neural network according to an embodiment of the present invention;
fig. 6 is a schematic structural diagram of a computer device according to an embodiment of the present invention.
The achievement of the objects, functional features and advantages of the present invention will be further described with reference to the accompanying drawings, in conjunction with the embodiments.
Detailed Description
It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention.
Fig. 1 is an application scenario diagram of an advertisement click rate prediction method based on a neural network, where as shown in fig. 1, a client collects historical advertisement features and attribute features of a target user, sends the historical advertisement features and the attribute features of the target advertisement to a server, and after receiving the historical advertisement features and the attribute features of the target user, the server executes the advertisement click rate prediction method based on the neural network to predict the click rate of the target user on the target advertisement.
The embodiment of the application can acquire and process the related data based on the artificial intelligence technology. Among these, artificial intelligence (Artificial Intelligence, AI) is the theory, method, technique and application system that uses a digital computer or a digital computer-controlled machine to simulate, extend and extend human intelligence, sense the environment, acquire knowledge and use knowledge to obtain optimal results.
Artificial intelligence infrastructure technologies generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technologies, operation/interaction systems, mechatronics, and the like. The artificial intelligence software technology mainly comprises a computer vision technology, a robot technology, a biological recognition technology, a voice processing technology, a natural language processing technology, machine learning/deep learning and other directions.
It should be noted that, the server may be implemented by an independent server or a server cluster formed by a plurality of servers, or may be a cloud server that provides cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (Content Delivery Network, CDN), and basic cloud computing services such as big data and artificial intelligence platforms.
The client may be, but is not limited to, a smart phone, a tablet computer, a notebook computer, a desktop computer, etc. The client and the server may be connected by bluetooth, USB (Universal Serial Bus ) or other communication connection, which is not limited in this embodiment of the present invention.
Fig. 2 is a flowchart of an advertisement click rate prediction method based on a neural network according to an embodiment of the present invention, as shown in fig. 2, where the method includes:
s1, acquiring historical advertisement characteristics of a target user and attribute characteristics of the target advertisement, wherein the historical advertisement characteristics comprise basic information of the target user and historical purchase information between the target user and a target product, and the attribute characteristics are obtained according to the basic information of the target advertisement;
for a target user and a target advertisement, when the probability that the target user clicks the target advertisement needs to be judged, basic information of the target user and the target advertisement is firstly input, then information related to the target user and attribute information related to the advertisement are extracted, the information is called historical advertisement characteristics of the target user, the historical advertisement characteristics describe some basic information of the target user and some information related to the advertisement in the past, the basic information of the user comprises information such as age, sex, occupation, region and the like of the user, and the information related to the advertisement is particularly different according to advertisement types, for example, if a target product of the advertisement is a car insurance, the historical purchase information related to the car insurance of the user comprises information such as brand, car age, historical car claim information, car insurance, insurance amount, application time, car mileage and the like of a owned car; if the target product of the advertisement is an insurance, the historical purchase information about the user's insurance includes information about the user's marital status, the product being applied, the premium, the amount of the insurance, the expiration date of the insurance, and the expiration date of the insurance. It can be seen that the historical advertisement features of the target users corresponding to different target advertisements are different, and can be specifically determined according to actual situations, and the embodiments of the present invention are not described here one by one.
The attribute characteristics of the target advertisement comprise some basic information of the target advertisement, such as advertisement duration, advertisement belonging category, advertisement quality, advertisement audience group, selling price of advertisement products, historical click rate of the advertisement products and the like, and can be determined according to practical situations.
S2, inputting the historical advertisement features and the attribute features into a click prediction model, respectively embedding the historical advertisement features and the attribute features into a dense process in sequence, carrying out cross fusion processing on the processed historical advertisement features and the processed attribute features, and predicting the click rate of the target user on the target advertisement, wherein the click prediction model is obtained by training a user sample, an advertisement sample and a score label.
Specifically, the historical advertisement features and the attribute features of the target advertisement are input into a click prediction model, the click prediction model performs embedding dense processing and cross fusion processing on the historical advertisement features and the attribute features, the embedding dense processing refers to mapping high-dimensional sparse features in the historical advertisement features and the attribute features into dense vectors respectively and splicing the dense vectors with continuous features, and therefore the problem that prediction accuracy is reduced due to high-dimensional sparse vector data missing in the prior art is solved.
The cross fusion processing is to fuse the user characteristics and the advertisement characteristics on the basis of carrying out dense processing on the historical advertisement characteristics and the attribute characteristics of the user independently, so that the user information and the advertisement information can be mined from different layers, and the user information and the advertisement information can be fully fused, thereby further improving the prediction precision of the user on the advertisement.
The click prediction model is a neural network model, and the common neural network model has a structure of a BP neural network model, a convolutional neural network, and the like, and may be specifically selected according to actual needs, which is not specifically limited in this embodiment.
Generally, before the click prediction model is used, training is needed, the structure of the neural network and the forward propagation output result are defined first, then a loss function is defined and a backward propagation optimization algorithm is selected, and the backward propagation optimization algorithm is repeatedly operated on training data, wherein the training data comprise historical advertisement features of a user sample, attribute features of the advertisement sample and labels, and the labels refer to the click rate of the user sample in the advertisement sample.
The training process is to determine the most suitable parameters of the click prediction model, and after the parameters of each structure in the click prediction model are determined, the click prediction model has the capability of predicting the click rate of the advertisement, and when the prediction is needed, the historical advertisement characteristics of the target user and the attribute characteristics of the target advertisement are directly input into the click prediction model, so that the click rate of the target user on the target advertisement can be obtained.
The embodiment of the invention provides an advertisement click prediction method, which adopts a neural network model to embed and densify historical advertisement features and attribute features, obtains user embedded features and advertisement embedded features represented by dense vectors through embedding and densifying, wherein the dense vectors are effective for samples with most dimensions relative to high-dimensional sparse vectors, thereby solving the problem of prediction accuracy reduction caused by high-dimensional sparse vector data loss in the prior art; on the basis of carrying out dense processing on the historical advertisement features and the attribute features of the user independently, the user features and the advertisement features are fused and embedded in multiple layers, and the information of the user and the advertisement can be mined from different layers, so that the comprehensiveness of data processing is improved, the user and the advertisement information are fully fused, and the prediction precision of the user on the advertisement is further improved.
The step of inputting the historical advertisement features and the attribute features into a click prediction model, respectively carrying out embedding dense processing on the historical advertisement features and the attribute features in sequence, and carrying out cross fusion processing on the processed historical advertisement features and the processed attribute features to predict the click rate of the target user on the target advertisement, wherein the method specifically comprises the following steps:
Respectively carrying out embedding dense processing on the historical advertisement features and the attribute features to respectively obtain user embedding features and advertisement embedding features;
cross fusion and average pooling are carried out on the user embedded features and the advertisement embedded features, and fusion features are obtained;
and predicting the click rate of the target user on the target advertisement according to the fusion characteristics.
Specifically, embedding dense processing is carried out on the historical advertisement characteristics, and user embedded characteristics are obtained; performing embedding dense processing on the attribute features to obtain advertisement embedding features; cross fusion and average pooling are carried out on the user embedded features and the advertisement embedded features, and fusion features are obtained; and predicting the click rate of the target user on the target advertisement according to the fusion characteristics.
And carrying out embedding dense processing on the historical advertisement characteristics to obtain user embedding characteristics, and carrying out embedding dense processing on the attribute characteristics to obtain advertisement embedding characteristics.
It should be noted that, embedding means mapping the feature into a dense vector, embedding means converting (dimension-reducing) the data into a feature representation (vector) of a fixed size, so as to facilitate processing and calculation (such as distance calculation), the user embedded feature and the advertisement embedded feature obtained after the embedding dense process are a dense vector, the dense vector is a sparse vector, the sample value of most dimensions in the sparse vector is 0, and the effective sample value is only a small part, and in contrast, the sample value of most dimensions in the dense vector is effective, so that the problem of data missing in the prior art cannot exist when the feature is mapped into the dense vector. The general category features are sparse after single-heat encoding, and dense embedding treatment is carried out to obtain dense user embedded features and advertisement embedded features, so that the problem of dimensional explosion caused by direct input of sparse features is avoided.
And then carrying out cross fusion and average pooling treatment on the user embedded features and the advertisement embedded features to obtain fusion features, wherein the cross fusion of the features can fully combine the advantages of the user embedded features and the advertisement embedded features, mine user information and advertisement information from different layers, fully fuse the user information and the advertisement information, and improve the prediction precision of the user on the advertisement.
And then predicting the click rate of the target user on the target advertisement according to the fusion characteristics.
According to the embodiment of the invention, the original historical advertisement features and attribute features are subjected to embedding and dense processing, and the category features are subjected to independent thermal encoding and sparse, so that the user embedded features and the advertisement embedded features which are relatively dense are obtained through embedding and dense processing, the problem of data missing in the process of mapping the low-dimensional vector to the high-dimensional space in the prior art is avoided, and the problem of dimensional explosion caused by directly inputting sparse features is also overcome; in addition, through carrying out cross fusion processing on the user embedded features and the advertisement embedded features, the advantages of the user information and the advertisement information can be fully combined, and the prediction accuracy is improved.
Specifically, mapping high-dimensional sparse features in the historical advertisement features into user dense features; normalizing the user continuous features in the advertisement features to obtain normalized user continuous features; splicing the user dense features and the normalized user continuous features to obtain user spliced features; and carrying out multi-layer embedding on the user splicing characteristics to obtain the user embedding characteristics.
High-dimensional sparse features are understood to mean that low-dimensional dense features are mapped to a high-dimensional space, the number of 0, 0's in which many are present is much greater than the number of other values, and high-dimensional sparse features are typically class features, typically expressed in one-hot, and mathematically are a vector.
Firstly, mapping high-dimensional sparse features in historical advertisement features into user dense features, normalizing continuous features in the historical advertisement features, and splicing the user dense features and the normalized user continuous features to obtain user splicing features, wherein the user splicing features are marked as x user 。
It should be noted that, the dense user feature may be obtained by multiplying the high-dimensional sparse feature with a preset conversion matrix. Taking 1024 independent words as an example for illustration, the high-dimensional sparse feature is expressed by one-hot, namely 1024 dimensions, and the advantage is that all dimensions are mutually orthogonal, and cosine similarity is 0. The dense vector representation is less than 1024 dimensions in dimension, and a minimum of 10 dimensions, i.e., a binary coded representation.
In order to further mine the characteristics of the user, a multi-layer embedding is needed to be carried out on the user splicing characteristics, the main idea is that the user embedding characteristics are projected into m different subspaces to obtain different representations of the user embedding, and the calculation formula is as follows:
Wherein x is user Representing the characteristics of the user splice,representing the ith element in the user-embedded feature,representing the i-th element of the preset user's embedded matrix.
By carrying out multi-layer embedding processing on the historical advertisement characteristics of the user, the information contained in the user is fully mined, and the user information existing in the historical advertisement characteristics is mined from different layers, so that the prediction accuracy of the click rate is improved.
Specifically, mapping high-dimensional sparse features in attribute features of the target advertisement to advertisement dense features; normalizing the advertisement continuous features in the attribute features to obtain normalized advertisement continuous features; splicing the advertisement dense features and the normalized advertisement continuous features to obtain advertisement splicing features; and carrying out multi-layer embedding on the advertisement splicing characteristics to obtain advertisement embedding characteristics.
Similarly, mapping the high-dimensional sparse features in the attribute features into advertisement dense features, normalizing continuous features in the attribute features, and splicing the advertisement dense features and the normalized advertisement continuous features to obtain advertisement splicing features, wherein the advertisement splicing features are marked as x item The feature is represented by a vector.
In order to further mine the characteristics of the advertisement, a multi-layer embedding is needed to be carried out on the advertisement splicing characteristics, the main idea is to project the advertisement embedding characteristics into m different subspaces to obtain different representations of advertisement embedding, and the calculation formula is as follows:
wherein x is item Representing the characteristics of the splicing of the advertisements,representing the ith element in the ad embedding feature,representing the ith element of the preset advertisement embedded matrix.
By carrying out multi-layer embedding processing on the attribute characteristics of the advertisement, advertisement information existing in the attribute characteristics is mined from different layers, and information contained in the advertisement is fully mined, so that the prediction accuracy of click rate is improved.
After the user embedded feature and the advertisement embedded feature are obtained by the method, the user embedded feature is embedded into the userAnd advertisement embedding feature->The fusion is carried out in the following specific way:
wherein,representing the i-th element of the fused user-embedded feature,/->The i element of the embedded feature of the advertisement after fusion is represented, j, m and k are positive integers, e represents an identity matrix, and +.>Representing the ith element in the user embedded feature,representing the ith element in the ad embedding feature.
Next, an average pooling (i.e., averaging) is performed on the fused features to obtain fused features:
Wherein,representing the user fusion feature after mean pooling, +.>Representing the average pooled ad fusion features.
On the basis of the foregoing embodiment, preferably, the predicting, according to the fusion feature, a click rate of the target advertisement by the target user includes:
and inputting the fusion characteristics into a full-connection layer, and obtaining the click rate through an activation function by outputting the full-connection layer.
In the implementation process, the fusion characteristics are input into an output layer, the output layer is a fully connected network, then an activation function is connected, and the purpose of introducing the activation function is as follows: because the data distribution is mostly nonlinear, but the calculation of a general neural network is linear, the neural network can simulate the nonlinear data distribution by introducing an activation function, and the learning capacity of the network is enhanced. The greatest feature of the activation function is non-linearity. Several activation functions are common: the Sigmoid function, the tanh function and the ReLU function are specifically selected according to the distribution characteristics of the analog data, and in the embodiment of the invention, the Sigmoid function is selected as an activation function.
And obtaining the click rate of the user on the advertisement by using the data output by the output layer through a sigmoid activation function, wherein the calculation formula of the output layer is as follows:
Wherein y represents click rate, FCN represents full convolution network, x user Representing user splice characteristics, x item Representing the characteristics of the advertisement splice,representing the user fusion feature after mean pooling, +.>Representing the average pooled ad fusion features.
Fig. 3 is a schematic structural diagram of a click prediction model according to an embodiment of the present invention, as shown in fig. 3, in a specific implementation process, the click prediction model is composed of three layers, namely an embedding layer 310, an interaction layer 320 and an output layer 330, where the embedding layer further includes two parallel independent user embedding units 311 and an advertisement embedding unit 312, the user embedding units are used for performing embedding processing on historical advertisement features of users, and the advertisement embedding units are used for performing embedding processing on attribute features of advertisements. The embedded layer, the interaction layer and the output layer are sequentially connected in sequence, and finally are connected and transferred to the sigmoid layer.
Specifically, the embedded layer comprises a user embedded unit and an advertisement embedded unit, the historical advertisement characteristics are input into the user embedded unit to obtain user embedded characteristics, the attribute characteristics of the target advertisement are input into the advertisement embedded unit to obtain advertisement embedded characteristics, and the embedded layer processes two parts of a user and an advertisement separately so as to fully mine user information and advertisement information.
And inputting the user embedded features and the advertisement embedded features into an interaction layer to obtain fusion features, and inputting the fusion features into an output layer to obtain the click rate of the target user on the target advertisement.
On the basis of the above embodiment, preferably, the method for obtaining the user sample and the advertisement sample includes:
acquiring an initial user sample and an initial advertisement sample;
and respectively processing the initial user sample and the initial advertisement sample by adopting a preset rule to obtain the user sample and the advertisement sample, wherein the preset rule comprises filling in missing values of a nearest neighbor algorithm and sampling in layers.
The obtained initial user sample and initial advertisement sample are the most original data samples, and the sample data have a lot of noise, and the problems of partial characteristic data missing and extremely unbalanced positive and negative samples are also faced in the subsequent application of click prediction model training, so that the training of the subsequent click prediction model is influenced very poorly, and the trained click prediction model has low precision.
Adopting a missing value filling technology of a nearest neighbor algorithm and a hierarchical sampling technology to sequentially process an initial user sample to obtain a user sample;
And adopting a missing value filling technology of a nearest neighbor algorithm and a layered sampling technology to sequentially process the initial advertisement sample to obtain the advertisement sample.
In the embodiment of the invention, a nearest neighbor algorithm (KNN) missing value filling technology is adopted to fill data in initial user samples and initial advertisements, an auxiliary variable (namely a variable without missing values) is utilized to define a distance function among samples, K non-missing value samples closest to the missing value samples are searched, and an average value or weighted average value of the K non-missing value samples is utilized to fill missing data.
And then clustering the initial user samples and the initial advertisement sample data by adopting a hierarchical sampling technology, clustering a plurality of class samples (negative samples), randomly extracting a certain number of samples from each class according to a proportion to form a new negative sample set, wherein the sampling can keep the balanced distribution of the samples to a great extent, and the prediction effect of the model is ensured on the premise of balancing the positive and negative samples.
If a photo classification of an automobile is made, a positive sample is a correct picture of the automobile, and a negative sample is a picture of the automobile. By model training, the click prediction model can be told which are pairs, which are wrong, pairs are positive samples, and errors are negative samples.
In the sample acquisition stage, the embodiment of the invention fills in the missing value through the nearest neighbor algorithm, and ensures the equalization degree of the positive sample and the negative sample in the sample through the layered sampling technology, thereby ensuring the prediction precision of the click prediction model.
On the basis of the above embodiment, it is preferable that the method further includes:
acquiring a comprehensive score of the target advertisement according to the click rate and the putting price of the target advertisement;
and judging whether to throw the target advertisement according to the comprehensive score.
Specifically, according to the putting price of the target advertisement and other advertisement bids, the price weight of the target advertisement is determined, the putting price of the target advertisement is expressed in a weight form, and the score of each advertisement is conveniently unified.
And multiplying the putting price of the target advertisement by the click rate to obtain the comprehensive score of the target advertisement.
Fig. 4 is a schematic diagram of an advertisement bidding delivery system based on a neural network-based advertisement click rate prediction method according to an embodiment of the present invention, as shown in fig. 4, the system includes an OBM410 (Own Branding & Manufacturing), a DSP420 (Demand-Side Platform), a DPP430 (Data Product Platform, data product service Platform) and an AEther440.
Wherein, the OBM comprises 4 advertisement strategies for users to select, wherein, strategy 1 is a default advertisement, namely, a default advertisement interface is opened; policy 2 is a model advertisement, i.e. the advertisement most likely to be clicked by the user is selected according to the user characteristics and advertisement characteristics, policy 3 is a corporate customization policy, which refers to an interface showing customization for a specific enterprise or a specific user.
The model advertisement strategy and the company customization strategy (the default advertisements or the customized advertisements are prepared in advance and are directly displayed) are directly adopted for specific clients (including new users), the strategy 2 is adopted for general users, the general users need to enter a DSP advertisement delivery platform subsequently, and the proper advertisements are selected through the model by combining the user characteristics and the advertisement characteristics.
The DSP platform is used for managing advertisements and transmitting corresponding information to the DPP and the OBM according to the selection of an advertiser.
The DPP is used for consulting a crowd pack to which the user belongs, acquiring user ID, material information and advertisement information transmitted by the DSP, and packaging input parameters of a prediction model;
the Ather modeling platform is used for predicting the click rate of the advertisement according to the user ID, the material information and the advertisement information.
The terminal 1 is a certain vehicle management APP, the terminal 2 is a certain advertisement management APP, and the description is given by taking the vehicle insurance as an example, and the specific implementation process and the algorithm model related in the process in the advertisement process of the accurate pushing are described in detail below.
1. User accesses vehicle management APP, and system sends out advertisement putting request
When a user accesses the vehicle management APP, the OBM is triggered to select an advertisement strategy, generally, the OBM selects the strategy 2, and executes two behavior instructions simultaneously, and simultaneously, the user communicates with the DPP system to inquire the historical advertisement characteristics of the crowd represented by the logged-in user while sending the advertisement request of the strategy 2 to the DSP.
2. Request prediction result of DSP advertisement delivery platform
An advertiser logs in an advertisement management APP and triggers a DSP platform to acquire attribute characteristics of a target advertisement; and after receiving the OBM policy 2 advertisement request, the DSP packages advertisement and material information to the DPP, namely, the attribute features are combined with the historical advertisement features and sent to the DPP, and the DPP requests a click prediction model deployed by the Aether data analysis modeling platform to obtain a prediction result.
Advertisement material: it means that something like pictures, videos are needed when advertising, that is, one advertisement is composed of a plurality of advertisement materials;
advertisement operation: overall process of advertisement initiation, planning and execution
Bidding and delivering: the novel network advertisement form paid according to the advertisement effect is automatically put in, managed by the user and ranked by adjusting the price.
3. And the AEther obtains a real-time click rate prediction result according to the prediction request and by combining information such as user ID, advertisement materials, crowd characteristics and the like.
4. Combining advertiser bids for placement
And the DSP receives the click rate and the advertiser bids and performs weight calculation to obtain the comprehensive score of the target advertisement.
The advertisement bidding delivery system provided by the embodiment has the following advantages:
1. in the existing advertising industry, a plurality of advertisement delivery platforms currently support advertisement owners to select advertisement positions, support advertisement position bidding, but do not support advertisement click rate prediction and selection of advertisement audience groups, so that the advertisement delivery effect is not good enough. The advertisement bidding system provided by the embodiment of the invention supports the advertisement as bidding on the basis of the prediction of the advertisement click rate, and can select the audience group suitable for the advertisement according to the comprehensive score, thereby realizing accurate advertisement delivery.
2. In the utilization of the existing advertising resources, the use of the same time interval and the same resource position cannot be scientifically and reasonably allocated, and the release effect cannot be scientifically estimated, so that the utilization efficiency of marketing resources is low; the advertisement marketing resource position is completely planned to be put in, and is used according to the resource requirement, so that the market competition is lacking, and the resource position can not be used as commodity to maximize the benefit. The advertisement bidding and delivering system comprehensively considers the rights and interests of advertisers, users and flow parties, takes a thousand-person and thousand-face advertisement click rate prediction model as a core, and pushes proper advertisements to interested people on the premise of ensuring the delivery demands of the flow parties and the advertisers as much as possible.
3. The advertisement click rate is not estimated by combining a scientific algorithm model, and advertisement audience users are not selected by combining crowd images, so that the advertisement requirements seen by the users are low in matching, the click rate is low, the conversion rate is low, and a large amount of exposure resource positions are wasted and invalid information is caused. The advertisement bidding delivery system builds a DSP tag system through the project, so that the accurate recommendation capability of thousands of people and thousands of faces is realized for users; constructing a click rate prediction model and a development model of the risk production system, and providing effect-oriented intelligent throwing capability for advertisers; and constructing a calculation advertisement data report system and an optimization strategy, and providing an automatic advertisement operation decision service capability for the platform.
Fig. 5 is a schematic structural diagram of an advertisement click rate prediction system based on a neural network according to an embodiment of the present invention, as shown in fig. 5, where the system includes a feature obtaining module 510 and a prediction module 520, where:
the feature acquisition module 510 is configured to acquire a historical advertisement feature of a target user and an attribute feature of the target advertisement, where the historical advertisement feature includes basic information of the target user and historical purchase information between the target user and a target product, and the attribute feature is acquired according to the basic information of the target advertisement;
The prediction module 520 is configured to input the historical advertisement feature and the attribute feature into a click prediction model, sequentially embed the historical advertisement feature and the attribute feature into a dense process, and perform a cross fusion process on the processed historical advertisement feature and the processed attribute feature, so as to predict the click rate of the target user on the target advertisement, where the click prediction model is obtained by training a user sample, an advertisement sample and a score tag.
The embodiment is a system embodiment corresponding to the advertisement click prediction method, the implementation process of the system embodiment is the same as that of the method embodiment, and the detailed description of the system embodiment is omitted herein with reference to the method embodiment.
According to the advertisement click rate prediction system based on the neural network, the neural network model is adopted to conduct embedding dense processing and cross fusion processing on the historical advertisement features and the attribute features, and the historical advertisement features and the attribute features are spliced with the continuous features, so that the problem that prediction accuracy is reduced due to the fact that high-dimensional sparse feature data are missing in the prior art is solved; on the basis of performing dense processing on the historical advertisement features of the user and the attribute features of the target advertisement independently, the user features and the advertisement features are fused, the information of the user and the advertisement can be mined from different layers, the information of the user and the advertisement is fully fused, and the prediction precision of the user on the advertisement is further improved.
Specifically, the prediction module further comprises a user embedding module, an advertisement embedding module, a fusion module and an output module, wherein:
the user embedding module is used for carrying out embedding dense processing on the historical advertisement characteristics to obtain user embedding characteristics;
the advertisement embedding module is used for carrying out embedding dense processing on the attribute characteristics to obtain advertisement embedding characteristics;
the fusion module is used for carrying out cross fusion and average pooling on the user embedded features and the advertisement embedded features to obtain fusion features;
and the output module is used for predicting the click rate of the target user on the target advertisement according to the fusion characteristics.
Specifically, the user embedding module includes a user mapping unit, a user normalization unit, a user splicing unit, and a user embedding unit, wherein:
the user mapping unit is used for mapping the high-dimensional sparse features in the historical advertisement features into user dense features;
the user normalization unit is used for normalizing the user continuous features in the historical advertisement features to obtain normalized user continuous features;
the user splicing unit is used for splicing the user dense features and the normalized user continuous features to obtain user splicing features;
The user embedding unit is used for carrying out multi-layer embedding on the user splicing characteristics to obtain user embedding characteristics.
Specifically, the advertisement embedding module comprises an advertisement mapping unit, an advertisement normalization unit, an advertisement splicing unit and an advertisement embedding unit, wherein:
the advertisement mapping unit is used for mapping the high-dimensional sparse features in the attribute features of the target advertisement into advertisement dense features;
the advertisement normalization unit is used for normalizing the advertisement continuous characteristics in the attribute characteristics to obtain normalized advertisement continuous characteristics;
the advertisement splicing unit is used for splicing the advertisement dense features and the normalized advertisement continuous features to obtain advertisement splicing features;
the advertisement embedding unit is used for carrying out multi-layer embedding on the advertisement splicing characteristics to obtain advertisement embedding characteristics.
Specifically, the output module includes:
and inputting the fusion characteristics into a full-connection layer, and obtaining the click rate through a sigmoid function by outputting the full-connection layer.
Specifically, the user sample and the advertisement sample in the prediction module are obtained by:
acquiring an initial user sample and an initial advertisement sample;
Adopting a missing value filling technology of a nearest neighbor algorithm and a hierarchical sampling technology to sequentially process the initial user sample to obtain the user sample;
and adopting a missing value filling technology and a layered sampling technology of a nearest neighbor algorithm to sequentially process the initial advertisement sample to obtain the advertisement sample.
Specifically, the advertisement click prediction system further comprises a synthesis module and a delivery module, wherein:
the comprehensive module is used for obtaining the comprehensive score of the target advertisement according to the click rate and the putting price of the target advertisement;
and the delivery module is used for judging whether to deliver the target advertisement according to the comprehensive score.
The specific implementation process of the embodiment is consistent with the embodiment of the method, and the details of the embodiment of the system are referred to the embodiment of the method and are not repeated herein.
The modules in the advertisement click rate prediction system based on the neural network can be all or partially realized by software, hardware and a combination thereof. The above modules may be embedded in hardware or may be independent of a processor in the computer device, or may be stored in software in a memory in the computer device, so that the processor may call and execute operations corresponding to the above modules.
Fig. 6 is a schematic structural diagram of a computer device according to an embodiment of the present invention, where the computer device may be a server, and an internal structure diagram of the computer device may be as shown in fig. 6. The computer device includes a processor, a memory, a network interface, and a database 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 includes a computer storage medium, an internal memory. The computer storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the computer storage media. The database of the computer equipment is used for storing data such as a flow node number and a target service node generated or acquired in the process of executing the advertisement click rate prediction method based on the neural network. The network interface of the computer device is used for communicating with an external terminal through a network connection. The computer program when executed by a processor implements a neural network-based advertisement click rate prediction method.
In one embodiment, a computer device is provided that includes a memory, a processor, and a computer program stored on the memory and executable on the processor, the processor implementing the steps of the neural network-based advertisement click rate prediction method of the above embodiments when the computer program is executed by the processor. Alternatively, the processor, when executing the computer program, implements the functions of the modules/units in this embodiment of a neural network-based advertisement click rate prediction system.
In one embodiment, a computer storage medium is provided, and a computer program is stored on the computer storage medium, where the computer program is executed by a processor to implement the steps of the advertisement click rate prediction method based on a neural network in the above embodiment. Alternatively, the computer program, when executed by a processor, implements the functions of the modules/units in the embodiment of the neural network-based advertisement click rate prediction system described above.
Those skilled in the art will appreciate that implementing all or part of the above described methods may be accomplished by way of a computer program stored on a non-transitory computer readable storage medium, which when executed, may comprise the steps of the embodiments of the methods described above. Any reference to memory, storage, database, or other medium used in the various embodiments provided herein may include non-volatile and/or volatile memory. The nonvolatile 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), memory bus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), among others.
It will be apparent to those skilled in the art that, for convenience and brevity of description, only the above-described division of the functional units and modules is illustrated, and in practical application, the above-described functional distribution may be performed by different functional units and modules according to needs, i.e. the internal structure of the apparatus is divided into different functional units or modules to perform all or part of the above-described functions.
The above embodiments are only for illustrating the technical solution of the present invention, and not for limiting the same; although the invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some technical features thereof can be replaced by equivalents; such modifications and substitutions do not depart from the spirit and scope of the technical solutions of the embodiments of the present invention, and are intended to be included in the scope of the present invention.
Claims (7)
1. The advertisement click rate prediction method based on the neural network is characterized by comprising the following steps of:
s1, acquiring historical advertisement characteristics of a target user and attribute characteristics of the target advertisement, wherein the historical advertisement characteristics comprise basic information of the target user and historical purchase information between the target user and a target product, and the attribute characteristics are obtained according to the basic information of the target advertisement;
S2, inputting the historical advertisement features and the attribute features into a click prediction model, respectively carrying out embedding dense processing on the historical advertisement features and the attribute features in sequence to respectively obtain user embedding features and advertisement embedding features, wherein the method comprises the following steps:
mapping the high-dimensional sparse features in the historical advertisement features into user dense features, and mapping the high-dimensional sparse features in the attribute features of the target advertisement into advertisement dense features;
normalizing the user continuous features in the historical advertisement features to obtain normalized user continuous features, normalizing the advertisement continuous features in the attribute features to obtain normalized advertisement continuous features;
splicing the user dense features and the normalized user continuous features to obtain user splicing features, and splicing the advertisement dense features and the normalized advertisement continuous features to obtain advertisement splicing features;
respectively carrying out multi-layer embedding on the user splicing characteristics and the advertisement splicing characteristics to obtain the user embedding characteristics and the advertisement embedding characteristics;
s3, carrying out cross fusion processing on the user embedded feature and the advertisement embedded feature to predict the click rate of the target user on the target advertisement, wherein the cross fusion processing comprises the following steps:
Cross fusion and average pooling are carried out on the user embedded features and the advertisement embedded features, and fusion features are obtained;
predicting the click rate of the target user on the target advertisement according to the fusion characteristics;
the click prediction model is obtained by training a user sample, an advertisement sample and a score label;
the method for acquiring the user sample and the advertisement sample comprises the following steps:
acquiring an initial user sample and an initial advertisement sample;
and respectively processing the initial user sample and the initial advertisement sample by adopting a preset rule to obtain the user sample and the advertisement sample, wherein the preset rule comprises filling in missing values and sampling in layers of a nearest neighbor algorithm.
2. The neural network-based advertisement click rate prediction method according to claim 1, wherein the user mosaic feature and the advertisement mosaic feature are respectively embedded in multiple layers to obtain a user embedding feature and the advertisement embedding feature, and a specific calculation formula is as follows:
，
，
wherein,representing the user's spellingConnection feature(s)>Representing the advertisement concatenation feature,/->Representing the ith element in the user-embedded feature,/- >Representing the i-th element in the advertisement embedding feature,/->Representing the ith element of the preset user's embedded matrix,/->Represents the ith element in the preset advertisement embedding matrix, and m represents the number of embedded subspaces.
3. The method for predicting the click-through rate of an advertisement based on a neural network according to claim 1, wherein predicting the click-through rate of the target user on the target advertisement according to the fusion feature comprises:
and inputting the fusion characteristics into a full-connection layer, and obtaining the click rate through an activation function by outputting the full-connection layer.
4. A method of predicting advertisement click-through rate based on a neural network as set forth in any one of claims 1 to 3, further comprising:
acquiring a comprehensive score of the target advertisement according to the click rate and the putting price of the target advertisement;
and judging whether to throw the target advertisement according to the comprehensive score.
5. An advertising click-through rate prediction system based on a neural network, comprising:
the characteristic acquisition module is used for acquiring historical advertisement characteristics of a target user and attribute characteristics of the target advertisement, wherein the historical advertisement characteristics comprise basic information of the target user and historical purchase information between the target user and a target product, and the attribute characteristics are acquired according to the basic information of the target advertisement;
The prediction module is configured to input the historical advertisement feature and the attribute feature into a click prediction model, and sequentially perform embedding dense processing on the historical advertisement feature and the attribute feature to obtain a user embedded feature and an advertisement embedded feature, respectively, where the prediction module includes:
mapping the high-dimensional sparse features in the historical advertisement features into user dense features, and mapping the high-dimensional sparse features in the attribute features of the target advertisement into advertisement dense features;
normalizing the user continuous features in the historical advertisement features to obtain normalized user continuous features, normalizing the advertisement continuous features in the attribute features to obtain normalized advertisement continuous features;
splicing the user dense features and the normalized user continuous features to obtain user splicing features, and splicing the advertisement dense features and the normalized advertisement continuous features to obtain advertisement splicing features;
respectively carrying out multi-layer embedding on the user splicing characteristics and the advertisement splicing characteristics to obtain the user embedding characteristics and the advertisement embedding characteristics;
performing cross fusion processing on the user embedded feature and the advertisement embedded feature to predict the click rate of the target user on the target advertisement, including:
Cross fusion and average pooling are carried out on the user embedded features and the advertisement embedded features, and fusion features are obtained;
predicting the click rate of the target user on the target advertisement according to the fusion characteristics;
the click prediction model is obtained by training a user sample, an advertisement sample and a score label;
the method for acquiring the user sample and the advertisement sample comprises the following steps:
acquiring an initial user sample and an initial advertisement sample;
and respectively processing the initial user sample and the initial advertisement sample by adopting a preset rule to obtain the user sample and the advertisement sample, wherein the preset rule comprises filling in missing values and sampling in layers of a nearest neighbor algorithm.
6. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that the processor implements the steps of the neural network based advertisement click rate prediction method of any one of claims 1 to 4 when the computer program is executed.
7. A computer storage medium storing a computer program, wherein the computer program when executed by a processor implements the steps of the neural network-based advertisement click rate prediction method of any one of claims 1 to 4.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111016975.XA CN113706211B (en) | 2021-08-31 | 2021-08-31 | Advertisement click rate prediction method and system based on neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111016975.XA CN113706211B (en) | 2021-08-31 | 2021-08-31 | Advertisement click rate prediction method and system based on neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113706211A CN113706211A (en) | 2021-11-26 |
| CN113706211B true CN113706211B (en) | 2024-04-02 |
Family
ID=78658430
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111016975.XA Active CN113706211B (en) | 2021-08-31 | 2021-08-31 | Advertisement click rate prediction method and system based on neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113706211B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20230229736A1 (en) * | 2022-01-19 | 2023-07-20 | Lemon Inc. | Embedding optimization for a machine learning model |
| CN117522479A (en) * | 2023-11-07 | 2024-02-06 | 北京创信合科技有限公司 | Accurate Internet advertisement delivery method and system |
| CN117455366A (en) * | 2023-12-19 | 2024-01-26 | 杭州网易云音乐科技有限公司 | Inventory allocation method, device, medium and equipment based on neural network model |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110619540A (en) * | 2019-08-13 | 2019-12-27 | 浙江工业大学 | Click stream estimation method of neural network |
| CN110825957A (en) * | 2019-09-17 | 2020-02-21 | 中国平安人寿保险股份有限公司 | Deep learning-based information recommendation method, device, equipment and storage medium |
| CN111325579A (en) * | 2020-02-25 | 2020-06-23 | 华南师范大学 | Advertisement click rate prediction method |
| CN111339415A (en) * | 2020-02-25 | 2020-06-26 | 中国科学技术大学 | Click rate prediction method and device based on multi-interactive attention network |
| CN111506811A (en) * | 2020-03-19 | 2020-08-07 | 上海理工大学 | Click rate prediction method based on deep residual error network |
| CN112365297A (en) * | 2020-12-04 | 2021-02-12 | 东华理工大学 | Advertisement click rate estimation method |
| CN112381581A (en) * | 2020-11-17 | 2021-02-19 | 东华理工大学 | Advertisement click rate estimation method based on improved Transformer |
| KR102232393B1 (en) * | 2020-03-26 | 2021-03-25 | 한국해양대학교 산학협력단 | Method and apparatus for developing economic operation solution of autonomous ship |
| CN112700274A (en) * | 2020-12-29 | 2021-04-23 | 华南理工大学 | Advertisement click rate estimation method based on user preference |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7631008B2 (en) * | 2005-11-16 | 2009-12-08 | Yahoo! Inc. | System and method for generating functions to predict the clickability of advertisements |
| US20140108094A1 (en) * | 2012-06-21 | 2014-04-17 | Data Ventures, Inc. | System, method, and computer program product for forecasting product sales |
| BR112021017318A2 (en) * | 2019-04-10 | 2021-11-16 | Climate Corp | Leverage Feature Engineering to Increase Placement Predictability for Seed Product Selection and Per-Field Recommendation |
-
2021
- 2021-08-31 CN CN202111016975.XA patent/CN113706211B/en active Active
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110619540A (en) * | 2019-08-13 | 2019-12-27 | 浙江工业大学 | Click stream estimation method of neural network |
| CN110825957A (en) * | 2019-09-17 | 2020-02-21 | 中国平安人寿保险股份有限公司 | Deep learning-based information recommendation method, device, equipment and storage medium |
| CN111325579A (en) * | 2020-02-25 | 2020-06-23 | 华南师范大学 | Advertisement click rate prediction method |
| CN111339415A (en) * | 2020-02-25 | 2020-06-26 | 中国科学技术大学 | Click rate prediction method and device based on multi-interactive attention network |
| CN111506811A (en) * | 2020-03-19 | 2020-08-07 | 上海理工大学 | Click rate prediction method based on deep residual error network |
| KR102232393B1 (en) * | 2020-03-26 | 2021-03-25 | 한국해양대학교 산학협력단 | Method and apparatus for developing economic operation solution of autonomous ship |
| CN112381581A (en) * | 2020-11-17 | 2021-02-19 | 东华理工大学 | Advertisement click rate estimation method based on improved Transformer |
| CN112365297A (en) * | 2020-12-04 | 2021-02-12 | 东华理工大学 | Advertisement click rate estimation method |
| CN112700274A (en) * | 2020-12-29 | 2021-04-23 | 华南理工大学 | Advertisement click rate estimation method based on user preference |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113706211A (en) | 2021-11-26 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113706211B (en) | Advertisement click rate prediction method and system based on neural network | |
| CN109919316B (en) | Method, device and equipment for acquiring network representation learning vector and storage medium | |
| WO2020135535A1 (en) | Recommendation model training method and related apparatus | |
| CN107330731B (en) | Method and device for identifying click abnormity of advertisement space | |
| CN109783730A (en) | Products Show method, apparatus, computer equipment and storage medium | |
| US20210056458A1 (en) | Predicting a persona class based on overlap-agnostic machine learning models for distributing persona-based digital content | |
| US10748195B2 (en) | Database image matching using machine learning with output estimation | |
| CN111008335B (en) | Information processing method, device, equipment and storage medium | |
| US10395313B1 (en) | Image analysis and identification using machine learning with output personalization | |
| US10325315B1 (en) | Database image matching using machine learning with output personalization | |
| CN112348592A (en) | Advertisement recommendation method and device, electronic equipment and medium | |
| US20240104159A1 (en) | Creating an effective product using an attribute solver | |
| CN114331495A (en) | Multimedia data processing method, device, equipment and storage medium | |
| US11962817B2 (en) | Machine learning techniques for advanced frequency management | |
| CN113570422B (en) | Creative guide information generation method and device, computer equipment and storage medium | |
| CN115222112A (en) | Behavior prediction method, behavior prediction model generation method and electronic equipment | |
| CN113779116A (en) | Object sorting method, related equipment and medium | |
| US20140143019A1 (en) | Managing modeled audience extension information | |
| CN113239879A (en) | Federal model training and certificate detection method, device, equipment and medium | |
| CN114417944B (en) | Recognition model training method and device, and user abnormal behavior recognition method and device | |
| WO2024051492A1 (en) | Content pushing method and apparatus, device, and medium | |
| KR102637198B1 (en) | Method, computing device and computer program for sharing, renting and selling artificial intelligence model through artificial intelligence model production platform | |
| US20240040202A1 (en) | Adaptive placement of audiovisual content on user devices | |
| CN115687593A (en) | Intelligent question answering method and device, computer equipment and storage medium | |
| CN116127363A (en) | Customer classification method, apparatus, device, medium, and program product |
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 |