CN112036963A - Webpage advertisement putting device and method based on multilayer random hidden feature model - Google Patents
Webpage advertisement putting device and method based on multilayer random hidden feature model Download PDFInfo
- Publication number
- CN112036963A CN112036963A CN202011012586.5A CN202011012586A CN112036963A CN 112036963 A CN112036963 A CN 112036963A CN 202011012586 A CN202011012586 A CN 202011012586A CN 112036963 A CN112036963 A CN 112036963A
- Authority
- CN
- China
- Prior art keywords
- advertisement
- matrix
- user
- characteristic matrix
- target
- 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 27
- 239000011159 matrix material Substances 0.000 claims abstract description 197
- 230000004913 activation Effects 0.000 claims abstract description 27
- 238000012163 sequencing technique Methods 0.000 claims abstract description 6
- 238000006243 chemical reaction Methods 0.000 claims abstract description 4
- 238000013480 data collection Methods 0.000 claims abstract description 4
- 230000008569 process Effects 0.000 claims description 11
- 238000005457 optimization Methods 0.000 claims description 6
- 238000002716 delivery method Methods 0.000 claims description 3
- 238000009795 derivation Methods 0.000 claims description 3
- 230000006399 behavior Effects 0.000 description 59
- 230000006870 function Effects 0.000 description 36
- 239000013598 vector Substances 0.000 description 32
- 238000004458 analytical method Methods 0.000 description 3
- 238000011161 development Methods 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 2
- 238000011156 evaluation Methods 0.000 description 2
- 230000003993 interaction Effects 0.000 description 2
- 230000008685 targeting Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 1
- 230000005484 gravity Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000012360 testing method Methods 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
- G06Q30/0251—Targeted advertisements
- G06Q30/0269—Targeted advertisements based on user profile or attribute
- G06Q30/0271—Personalized advertisement
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/15—Correlation function computation including computation of convolution operations
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/16—Matrix or vector computation, e.g. matrix-matrix or matrix-vector multiplication, matrix factorization
-
- 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
- G06Q30/0201—Market modelling; Market analysis; Collecting market data
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0241—Advertisements
- G06Q30/0277—Online advertisement
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Accounting & Taxation (AREA)
- Development Economics (AREA)
- Finance (AREA)
- Strategic Management (AREA)
- Theoretical Computer Science (AREA)
- Mathematical Physics (AREA)
- Data Mining & Analysis (AREA)
- Mathematical Optimization (AREA)
- Entrepreneurship & Innovation (AREA)
- Pure & Applied Mathematics (AREA)
- Mathematical Analysis (AREA)
- Computational Mathematics (AREA)
- Economics (AREA)
- Game Theory and Decision Science (AREA)
- Marketing (AREA)
- General Business, Economics & Management (AREA)
- Computing Systems (AREA)
- Algebra (AREA)
- Databases & Information Systems (AREA)
- Software Systems (AREA)
- General Engineering & Computer Science (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a webpage advertisement putting device and method based on a multilayer random hidden feature model; the utility model provides a webpage advertisement puts in device based on hidden feature model at random of multilayer which characterized in that: the device includes: the advertisement data collection module is used for collecting and storing advertisement behavior data of the user; the data conversion module is used for converting the advertisement behavior data into a target matrix; the characteristic updating module is used for initializing and updating relevant parameters related in the user behavior characteristic matrix and the advertisement characteristic matrix; randomly generating weights and bias based on the target matrix, and updating the behavior characteristics of each user by using an activation function to form a user behavior characteristic matrix; combining the user behavior characteristic matrix and the target matrix to obtain an advertisement characteristic matrix; the advertisement recommendation module is used for obtaining advertisement weight by utilizing the user behavior characteristic matrix and the advertisement characteristic matrix and sequencing the advertisement recommendation sequence of each user according to the advertisement weight; the invention can be widely applied to various internet platforms.
Description
Technical Field
The invention relates to the field of advertisements, in particular to a webpage advertisement putting device and method based on a multilayer random hidden feature model.
Background
Advertisements are a media means for persuading potential consumers to purchase or pay attention to certain goods, and from newspapers and magazines to televisions, movies and websites, advertisements are ubiquitous in our lives. Meanwhile, with the rapid development of mobile internet technology, the center of gravity begins to shift from traditional media advertisements to mobile internet advertisements. By means of the good digital media environment of the modern society, the development of the advertising media industry is quicker. However, many advertisements are randomly delivered at present, and dynamic adjustment is difficult to be made according to the attributes and behaviors of users, that is, a user selects an advertisement to deliver to the user at random, which results in low advertisement delivery efficiency and cannot well meet the requirements of mobile terminal development. Meanwhile, random advertisement putting may also cause the user to feel the objections, and in the face of such a situation, when webpage advertisements are put in real time, accurate putting of the webpage advertisements is guaranteed, which is a problem that needs to be solved urgently.
At present, some mainstream advertisement delivery platforms, such as a hundredth CPA advertisement platform, a Google AdSense personalized advertisement delivery platform developed by Google corporation, and a social network advertisement delivery system developed by Facebook all have very high advertisement recommendation precision for users, and due to the existence of a large amount of advertisements, the advertisement pushing efficiency needs to be improved. Meanwhile, some existing advertisement recommendation technologies match advertisements by establishing a user interest model based on user browsing records and behaviors, or recommend advertisements by calculating user similarity according to social network relationships of users. However, due to the existence of a large number of users and a large number of advertisements, under the condition of low advertisement click rate, for example, the advertisement click rate is in the proportion of ten-thousandth to one-thousandth, the problem of data sparseness is caused, so that the advertisement recommendation efficiency is low. How to efficiently and accurately recommend advertisements has become a major research and urgent problem in the industry.
Disclosure of Invention
The invention aims to provide a webpage advertisement putting device and method based on a multilayer random hidden feature model.
In order to solve the technical problems, the technical scheme of the invention is as follows:
the utility model provides a webpage advertisement puts in device based on hidden feature model at random of multilayer which characterized in that: the device includes:
and the advertisement data collection module is used for collecting and storing advertisement behavior data of the user.
And the data conversion module is used for converting the advertisement behavior data into a target matrix.
The characteristic updating module is used for initializing and updating relevant parameters involved in the user behavior characteristic matrix and the advertisement characteristic matrix. And randomly generating weights and bias based on the target matrix, and updating the behavior characteristics of each user by using an activation function to form a user behavior characteristic matrix. And combining the user behavior characteristic matrix and the target matrix to obtain an advertisement characteristic matrix.
The advertisement recommending module is used for obtaining advertisement weights by utilizing the user behavior characteristic matrix and the advertisement characteristic matrix, sequencing the advertisement recommending sequence of each user according to the advertisement weights, and selecting the advertisement with the highest weight from the advertisement recommending sequence to launch the user.
According to the preferable scheme of the webpage advertisement putting device based on the multilayer random hidden feature model, the feature updating module comprises:
and the parameter initialization unit is used for initializing and updating relevant parameters related in the user behavior characteristic matrix and the advertisement characteristic matrix.
And the user random characteristic updating unit randomly generates weights and bias based on the target matrix, and updates the behavior characteristic of each user by using the activation function to form a user behavior characteristic matrix.
And the advertisement characteristic updating unit is used for combining the user behavior characteristic matrix and the target matrix to obtain an advertisement characteristic matrix.
The second technical scheme of the invention is that the webpage advertisement putting method based on the multilayer random hidden feature model is characterized in that: the method comprises the following steps:
and S1, collecting and storing the advertisement behavior data of the user.
And S2, converting the advertisement behavior data into a target matrix and storing for later use.
And S3, randomly generating weights and biases based on the target matrix, generating a user behavior characteristic matrix by using an activation function, and then obtaining an advertisement characteristic matrix by using the derivation of the user behavior characteristic matrix.
S4, obtaining advertisement weight by using the user behavior characteristic matrix and the advertisement characteristic matrix, sequencing the advertisement recommendation sequence of each user according to the advertisement weight, and selecting the advertisement with the highest weight from the advertisement recommendation sequences to launch to the user.
According to the preferable scheme of the webpage advertisement putting method based on the multilayer random hidden feature model, the step S3 includes:
and S3-1, receiving an instruction for placing the advertisement sent by the server.
S3-2, parameter initialization: and initializing relevant parameters involved in the updating process of the user behavior characteristic matrix and the advertisement characteristic matrix.
These parameters include:
and controlling variable l by the number of layers of the multilayer structure. And the highest layer number L of the multilayer random hidden feature model. Advertising behavior targeting matrix R, element R in matrix Ru,iAnd the element of the ith row and the ith column in the representative matrix is a high-dimensional sparse matrix with the row number of M and the column number of N, wherein M represents the number of users, and N represents the number of advertisements. User behavior feature matrix PlThe matrix PlVector p in (1)uRepresenting the eigenvector of the u-th user in the matrix. Advertisement characteristic matrix QlThe matrix QlElement q in (1)iFeatures representing ith advertisement in matrixAnd (5) sign vectors. The number f of nodes of the hidden layer is the spatial dimension of the hidden feature. The activation function top layer weight matrix a. The first layer bias vector b. The activation function is a multi-level weight matrix W. A multi-layer bias vector d. A regularization penalty term factor λ. The convergence termination threshold is updated to be tau, etc.
The initial value of the number f of the hidden layer nodes is a positive integer.
User behavior feature matrix PlIs a matrix with a number of rows M and a number of columns f, and each element parameter is initialized to a very small random number.
Advertisement characteristic matrix QlIs a matrix with a number of rows f and a number of columns N, and each element parameter is initialized to a very small random number.
The size of the activation function top layer weight matrix a is a matrix with a number of rows f and a number of columns N.
The size of the activation function multi-level weight matrix W is a matrix with a number of rows f and a number of columns f.
The top layer offset vector b and the multi-layer offset vector d are vectors of length f.
The upper limit of the multilayer structure layer number control variable L is the highest layer number L of the multilayer random hidden feature model, and the parameters are initialized to positive integers.
The regularization penalty term factor lambda is a regularization constant that prevents overfitting during the update process and is initialized to a very small positive number.
The updated convergence termination threshold τ is a threshold parameter for judging whether the iteration process has converged, and is initialized to a minimum positive number.
S3-3, known data set R based on target matrix RKConstructing a target loss function by using a user behavior feature matrix PlAnd advertisement characteristic matrix QlThe inner product of (a) approximates the target loss function. And establishing a distance function corresponding to the approximation value as an optimization target. The optimization process is constrained using regularization.
S3-4, judging whether the control variable L of the multilayer structure layer number reaches the upper limit L, if so, executing S4, and if not, executing S3-5.
In this step, 1 is added to the control variable L of the number of layers of the multilayer structure, and then it is determined whether the control variable L of the number of layers of the multilayer structure is greater than the upper limit L.
S3-5, judging the target loss function in the known data set RKIf the above has converged, if yes, execute step S4, if not, execute steps S3-6.
In this step, the device judges that R isKThe basis of the upper convergence is that the value of the target loss function before the start of the current iteration is compared with the value of the target loss function before the start of the previous iteration, and whether the absolute value of the difference is smaller than the convergence judgment threshold tau or not is determined. If the convergence rate is less than the preset convergence rate, the convergence is judged, otherwise, the non-convergence is judged.
And S3-6, randomly generating weights and biases according to the target matrix and the initialized related parameters, and updating the user behavior characteristic matrix by using the randomly generated weights and biases and combining with the activation function.
And S3-7, combining the user behavior characteristic matrix and the target matrix to obtain an advertisement characteristic matrix. And returns to S3-4.
S3-5, judging the target loss function in the known data set RKIf yes, go to S4, if no, go to S3-6.
And S3-6, randomly generating weights and biases according to the target matrix and the initialized related parameters, and updating the user behavior characteristic matrix by using the randomly generated weights and biases and combining with the activation function.
And S3-7, updating the advertisement characteristic matrix by combining the user behavior characteristic matrix and the target matrix, and returning to the step S3-4.
The webpage advertisement releasing device and method based on the multilayer random hidden feature model have the advantages that the multilayer random hidden feature model is utilized, the webpage advertisement is released accurately to users while the webpage advertisement is released in real time, the advertisement releasing efficiency is high, efficient and accurate advertisement recommendation is achieved, the advertisement releasing cost is reduced, and the webpage advertisement releasing device and method can be widely applied to various internet platforms.
Drawings
Fig. 1 is a schematic structural diagram of a web advertisement delivery device based on a multilayer random hidden feature model.
Fig. 2 is a flowchart of a web advertisement delivery method based on a multilayer random hidden feature model.
Fig. 3a is a graph comparing the accuracy of advertisement placement before and after the implementation of the present invention.
FIG. 3b is a graph comparing the efficiency of advertisement placement before and after the practice of the present invention.
Detailed Description
The present invention will be described in further detail with reference to test examples and specific embodiments. It should be understood that the scope of the above-described subject matter is not limited to the following examples, and any techniques implemented based on the disclosure of the present invention are within the scope of the present invention.
Example 1: referring to fig. 1, a web advertisement delivery device based on a multilayer random hidden feature model includes:
and the advertisement data collection module 110 is used for collecting and storing advertisement behavior data of the user. Wherein the advertisement behavior data refers to a record of interactions between the user and the advertisement.
The data conversion module 120 is configured to convert the advertisement behavior data into a target matrix R, where the target matrix R is a high-dimensional sparse matrix with M rows and N columns, where M represents the number of users and N represents the number of advertisements. The target matrix R refers to a matrix converted by using interaction records between users and advertisements, and an element R in the matrixu,iRepresenting the element in the ith row and column of the matrix, r if the ith user browses the ith advertisementu,iOtherwise, the value is the missing value.
The feature update module 130 is used to initialize the relevant parameters involved in updating the user behavior feature matrix and the advertisement feature matrix. Randomly generating weights and biases based on the target matrix, updating the behavior characteristics of each user by using an activation function, and forming a user behavior characteristic matrix Pl. Combining the user behavior characteristic matrix and the target matrix to obtain an advertisement characteristic matrix Ql。
The weights comprise a first-layer weight matrix A and a multi-layer weight matrix W and are used for adding weighted items to the user behavior characteristic matrix in the activation function.
The bias comprises a first-layer bias vector b and a multi-layer bias vector d, and is used for adding a bias item to the user behavior feature matrix in the activation function.
User behavior feature matrix PlThe user behavior feature matrix obtained by updating in the l-th layer is used for representing the feature preference of the user. Where l is a multilayer structure layer number control variable. Vector p in the matrixuRepresenting the eigenvector of the u-th user in the matrix.
Advertisement characteristic matrix QlThe advertisement feature matrix obtained by updating in the ith layer is used for representing the characteristics of the advertisement, such as advertisement duration, advertisement type, advertisement applicable crowd and the like. Where l is a multilayer structure layer number control variable. Element q in the matrixiRepresenting the feature vector of the ith advertisement in the matrix.
The advertisement recommendation module 140 is configured to obtain advertisement weights by using the user behavior feature matrix and the advertisement feature matrix, sort the advertisement recommendation sequences of each user according to the advertisement weights, and select an advertisement with the highest weight from the advertisement recommendation sequences to be delivered to the user. The advertisement recommendation sequence of the user mainly comprises advertisements to be recommended, the order of the advertisements in the advertisement recommendation sequence is determined by advertisement weight, and the higher the weight is, the more the rank is, the more possible the advertisements are recommended to the user.
In a specific embodiment, the feature update module 130 includes:
the parameter initialization unit 131 is configured to initialize relevant parameters involved in updating the user feature matrix and the advertisement feature matrix.
The user random feature updating unit 132 randomly generates weights and biases based on the target matrix, and updates the behavior feature of each user by using the activation function to form a user behavior feature matrix.
The advertisement characteristic updating unit 133 combines the user behavior characteristic matrix and the target matrix to obtain an advertisement characteristic matrix.
Example 2: referring to fig. 2, a webpage advertisement delivery method based on a multilayer random hidden feature model is characterized in that: the method comprises the following steps:
and S1, collecting and storing the advertisement behavior data of the user.
And S2, converting the advertisement behavior data into a target matrix and storing for later use.
And S3, randomly generating weights and biases based on the target matrix, generating a user behavior characteristic matrix by using an activation function, and then obtaining an advertisement characteristic matrix by using the derivation of the user behavior characteristic matrix.
The weights comprise a first-layer weight matrix A and a multi-layer weight matrix W and are used for adding weighted items to the user behavior characteristic matrix in the activation function.
The bias comprises a first-layer bias vector b and a multi-layer bias vector d, and is used for adding a bias item to the user behavior feature matrix in the activation function.
User behavior feature matrix PlThe user behavior feature matrix obtained by updating in the l-th layer is used for representing the feature preference of the user. Where l is a multilayer structure layer number control variable. Vector p in the matrixuRepresenting the eigenvector of the u-th user in the matrix.
Advertisement characteristic matrix QlThe advertisement feature matrix obtained by updating in the ith layer is used for representing the characteristics of the advertisement, such as advertisement duration, advertisement type, advertisement applicable crowd and the like. Where l is a multilayer structure layer number control variable. Element q in the matrixiRepresenting the feature vector of the ith advertisement in the matrix.
In a particular embodiment, step S3 includes
And S3-1, receiving an instruction for placing the advertisement sent by the server.
S3-2, parameter initialization: and initializing relevant parameters involved in the updating process of the user behavior characteristic matrix and the advertisement characteristic matrix. These parameters include:
and controlling variable l by the number of layers of the multilayer structure. And the highest layer number L of the multilayer random hidden feature model. Advertising behavior targeting matrix R, element R in matrix Ru,iAnd the element of the ith row and the ith column in the representative matrix is a high-dimensional sparse matrix with the row number of M and the column number of N, wherein M represents the number of users, and N represents the number of advertisements. User behavior feature momentsArray PlThe matrix PlVector p in (1)uRepresenting the eigenvector of the u-th user in the matrix. Advertisement characteristic matrix QlThe matrix QlElement q in (1)iRepresenting the feature vector of the ith advertisement in the matrix. The number f of nodes of the hidden layer is the spatial dimension of the hidden feature. The activation function top layer weight matrix a. The first layer bias vector b. The activation function is a multi-level weight matrix W. A multi-layer bias vector d. A regularization penalty term factor λ. The convergence termination threshold is updated to be tau, etc.
The initial value of the number f of the hidden layer nodes is a positive integer.
User behavior feature matrix PlIs a matrix with a number of rows M and a number of columns f, and each element parameter is initialized to a very small random number.
Advertisement characteristic matrix QlIs a matrix with a number of rows f and a number of columns N, and each element parameter is initialized to a very small random number.
The size of the activation function top layer weight matrix a is a matrix with a number of rows f and a number of columns N.
The size of the activation function multi-level weight matrix W is a matrix with a number of rows f and a number of columns f.
The top layer offset vector b and the multi-layer offset vector d are vectors of length f.
The upper limit of the multilayer structure layer number control variable L is the highest layer number L of the multilayer random hidden feature model, and the parameters are initialized to positive integers.
The regularization penalty term factor lambda is a regularization constant that prevents overfitting during the update process and is initialized to a very small positive number.
The updated convergence termination threshold τ is a threshold parameter for judging whether the iteration process has converged, and is initialized to a minimum positive number.
S3-3, known data set R based on target matrix RKConstructing a target loss function by using a user behavior feature matrix PlAnd advertisement characteristic matrix QlThe inner product of (a) approximates the target loss function. And establishing a distance function corresponding to the approximation value as an optimization target. Using regularization, the optimization process is constrained, with the objective loss function as follows:
wherein:
Plrepresenting a matrix of user behavior characteristics, QlMatrix Q representing advertisement characteristicsl，ru,iRepresenting the element of the ith row and column in the object matrix R, RKRepresenting a set of known elements in the object matrix R,
represents PlThe feature vector of the u-th user,
represents QlThe feature vector of the ith advertisement;
to represent
The Frobenius norm of (A),
to represent
The frobenius norm of (a).
S3-4, judging whether the control variable L of the multilayer structure layer number reaches the upper limit L, if so, executing S4, and if not, executing S3-5.
In this step, 1 is added to the control variable L of the number of layers of the multilayer structure, and then it is determined whether the control variable L of the number of layers of the multilayer structure is greater than the upper limit L.
S3-5, judging the target loss function in the known data set RKIf the above has converged, if yes, execute step S4, if not, execute steps S3-6.
In this step, the device judges that R isKThe basis of the upper convergence is that the value of the target loss function before the start of the current iteration is compared with the value of the target loss function before the start of the previous iteration, and whether the absolute value of the difference is smaller than the convergence judgment threshold tau or not is determined. If the convergence rate is less than the preset convergence rate, the convergence is judged, otherwise, the non-convergence is judged.
S3-6, randomly generating weights and biases according to the target matrix and the initialized related parameters, and updating the user behavior characteristic matrix by using the randomly generated weights and biases and combining with the activation function, wherein the updating formula is as follows:
g(x)＝1/(1+e-x) In the form of an activation function.
PlA matrix of characteristics of the behavior of the user is represented,
represents PlThe feature vector of the u-th user,
representing the transpose of the u-th row vector in the object matrix R, aiRepresents the ith vector in the first-layer weight matrix a,
representing the ith vector in the multi-layer weight matrix W, biRepresenting the ith element in the top layer bias vector b,
representing the ith element in the multilayer offset vector d. Wherein u is a positive integer from 1 to M, i is a positive integer from 1 to f, and l is a positive integer from 1 to n.
And S3-7, updating the advertisement characteristic matrix by combining the user random characteristic matrix and the target matrix, and returning to the step S3-4. The update formula is as follows:
wherein
Represents PlThe feature vector of the u-th user,
to represent
The transpose of (a) is performed,
represents QlFeature vector of the ith advertisement, ru,iRepresenting the element of the ith row and column in the object matrix R, RKRepresents a set of known elements in the target matrix R, λ represents a regularization penalty factor, U (i) represents a set of users that clicked on advertisement i, RK (i) represents a set of known users associated with advertisement i, | RK(i) I represents the number of users having clicked on advertisement i, RKRepresenting the set of known elements in the target matrix R and I representing the identity matrix.
At RKAbove, the above updating process is repeated using the multi-layer structure until at RKFor the above convergence, the convergence determination condition is that the number of the layer number control variable l of the multilayer structure reaches the upper limit, or the value before the iteration of the layer starts, compared with the value before the update of the previous layer starts, the absolute value of the difference is already smaller than the convergence termination threshold value tau.
And S4, calculating by using the user behavior characteristic matrix and the advertisement characteristic matrix to obtain advertisement weight, sequencing the advertisement recommendation sequence of each user according to the advertisement weight, and selecting the advertisement with the highest weight from the advertisement recommendation sequences to launch the user. The advertisement recommendation sequence of the user mainly comprises advertisements to be recommended, the order of the advertisements in the advertisement recommendation sequence is determined by advertisement weight, and the higher the weight is, the more the rank is, the more possible the advertisements are recommended to the user.
In order to verify the performance of the webpage advertisement putting method and device based on the multilayer random hidden feature model, the device is installed on a server (configuration: Intel Xeon E5-2650 v4, 2.2GHz processor and 512G memory), and a simulation experiment is run to perform example analysis. In the example analysis, the adopted advertisement data is sourced from some internet advertisement delivery platform. The example analysis uses the mean square error RMSE as an evaluation index of the advertisement recommendation accuracy, the lower the RMSE, the higher the accuracy, and the shorter the time, the higher the efficiency by adopting the recommendation model running time as an evaluation index of the advertisement recommendation efficiency.
Fig. 3a and fig. 3b are graphs comparing the accuracy and efficiency of advertisement recommendation before and after the embodiment of the present invention is applied, respectively, where the greater the accuracy value, the higher the accuracy, the less the time required, and the higher the efficiency.
According to the technical scheme, the method is specially used for accurate recommendation of the advertisements, and can solve the problem of accuracy of webpage advertisement putting while guaranteeing real-time webpage advertisement putting.
Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art may still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some technical features. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims (4)
1. The utility model provides a webpage advertisement puts in device based on hidden feature model at random of multilayer which characterized in that: the device includes:
the advertisement data collection module (110) is used for collecting and storing advertisement behavior data of the user;
the data conversion module (120) is used for converting the advertisement behavior data into a target matrix;
the characteristic updating module (130) is used for initializing and updating relevant parameters involved in the user behavior characteristic matrix and the advertisement characteristic matrix; randomly generating weights and bias based on the target matrix, and updating the behavior characteristics of each user by using an activation function to form a user behavior characteristic matrix; combining the user behavior characteristic matrix and the target matrix to obtain an advertisement characteristic matrix;
the advertisement recommendation module (140) is used for obtaining advertisement weights by utilizing the user behavior characteristic matrix and the advertisement characteristic matrix, sequencing the advertisement recommendation sequence of each user according to the advertisement weights, and selecting the advertisement with the highest weight from the advertisement recommendation sequences to be delivered to the user.
2. The device for delivering webpage advertisements based on the multilayer random hidden feature model according to claim 1, wherein: the feature update module (130) comprises:
a parameter initialization unit (131) for initializing relevant parameters related to updating the user behavior characteristic matrix and the advertisement characteristic matrix;
a user random feature updating unit (132) which randomly generates weights and biases based on the target matrix, and updates the behavior feature of each user by using an activation function to form a user behavior feature matrix;
and an advertisement characteristic updating unit (133) which combines the user behavior characteristic matrix and the target matrix to obtain an advertisement characteristic matrix.
3. A webpage advertisement putting method based on a multilayer random hidden feature model is characterized by comprising the following steps: the method comprises the following steps:
s1, collecting and storing the advertisement behavior data of the user;
s2, converting the advertisement behavior data into a target matrix and storing for later use;
s3, randomly generating weights and biases based on the target matrix, generating a user behavior feature matrix by using an activation function, and then obtaining an advertisement feature matrix by using derivation of the user behavior feature matrix;
s4, obtaining advertisement weight by using the user behavior characteristic matrix and the advertisement characteristic matrix, sequencing the advertisement recommendation sequence of each user according to the advertisement weight, and selecting the advertisement with the highest weight from the advertisement recommendation sequences to launch to the user.
4. The web advertisement delivery method based on the multilayer random hidden feature model according to claim 3, characterized in that: step S3 includes:
s3-1, receiving an advertisement putting instruction sent by a server;
s3-2, parameter initialization: initializing relevant parameters related in the updating process of the user behavior characteristic matrix and the advertisement characteristic matrix;
s3-3, constructing a target loss function based on the known data set of the target matrix, and approximating the target loss function by using the inner product of the user behavior characteristic matrix and the advertisement characteristic matrix; establishing a distance function corresponding to the approximation value as an optimization target; using regularization to constrain the optimization process;
s3-4, judging whether the control variable of the multilayer structure layer number reaches the upper limit, if so, executing S4, and if not, executing S3-5;
s3-5, judging whether the target loss function has converged on the known data set, if yes, executing S4, if not, executing S3-6 steps;
s3-6, randomly generating weights and biases according to the target matrix and the initialized related parameters, and updating the user behavior characteristic matrix by using the randomly generated weights and biases and combining with the activation function;
s3-7, combining the user behavior feature matrix and the target matrix to obtain an advertisement feature matrix, and returning to the step S3-4.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011012586.5A CN112036963B (en) | 2020-09-24 | 2020-09-24 | Webpage advertisement putting device and method based on multilayer random hidden feature model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011012586.5A CN112036963B (en) | 2020-09-24 | 2020-09-24 | Webpage advertisement putting device and method based on multilayer random hidden feature model |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112036963A true CN112036963A (en) | 2020-12-04 |
| CN112036963B CN112036963B (en) | 2023-12-08 |
Family
ID=73575598
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011012586.5A Active CN112036963B (en) | 2020-09-24 | 2020-09-24 | Webpage advertisement putting device and method based on multilayer random hidden feature model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112036963B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2022141456A1 (en) * | 2020-12-31 | 2022-07-07 | 百果园技术(新加坡)有限公司 | Advertisement delivery allocation method and apparatus, and electronic device and storage medium |
| CN117078312A (en) * | 2023-09-05 | 2023-11-17 | 北京玖众科技股份有限公司 | Advertisement putting management method and system based on artificial intelligence |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107506414A (en) * | 2017-08-11 | 2017-12-22 | 武汉大学 | A kind of code based on shot and long term memory network recommends method |
| CN109446430A (en) * | 2018-11-29 | 2019-03-08 | 西安电子科技大学 | Method, apparatus, computer equipment and the readable storage medium storing program for executing of Products Show |
| CN109615452A (en) * | 2018-10-29 | 2019-04-12 | 华中科技大学 | A kind of Products Show method based on matrix decomposition |
| CN110390561A (en) * | 2019-07-04 | 2019-10-29 | 四川金赞科技有限公司 | User-financial product of stochastic gradient descent is accelerated to select tendency ultra rapid predictions method and apparatus based on momentum |
-
2020
- 2020-09-24 CN CN202011012586.5A patent/CN112036963B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107506414A (en) * | 2017-08-11 | 2017-12-22 | 武汉大学 | A kind of code based on shot and long term memory network recommends method |
| CN109615452A (en) * | 2018-10-29 | 2019-04-12 | 华中科技大学 | A kind of Products Show method based on matrix decomposition |
| CN109446430A (en) * | 2018-11-29 | 2019-03-08 | 西安电子科技大学 | Method, apparatus, computer equipment and the readable storage medium storing program for executing of Products Show |
| CN110390561A (en) * | 2019-07-04 | 2019-10-29 | 四川金赞科技有限公司 | User-financial product of stochastic gradient descent is accelerated to select tendency ultra rapid predictions method and apparatus based on momentum |
Non-Patent Citations (2)
| Title |
|---|
| 涂丹丹;舒承椿;余海燕;: "基于联合概率矩阵分解的上下文广告推荐算法", 软件学报, no. 03 * |
| 解贵龙;张;于重重;赵霞;: "基于数字标牌广告数据的兴趣点推荐算法研究", 计算机应用与软件, no. 07 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2022141456A1 (en) * | 2020-12-31 | 2022-07-07 | 百果园技术(新加坡)有限公司 | Advertisement delivery allocation method and apparatus, and electronic device and storage medium |
| CN117078312A (en) * | 2023-09-05 | 2023-11-17 | 北京玖众科技股份有限公司 | Advertisement putting management method and system based on artificial intelligence |
| CN117078312B (en) * | 2023-09-05 | 2024-02-27 | 北京玖众科技股份有限公司 | Advertisement putting management method and system based on artificial intelligence |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112036963B (en) | 2023-12-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110941740B (en) | Video recommendation method and computer-readable storage medium | |
| US10609433B2 (en) | Recommendation information pushing method, server, and storage medium | |
| CN109902849B (en) | User behavior prediction method and device, and behavior prediction model training method and device | |
| CA3007853C (en) | End-to-end deep collaborative filtering | |
| US10489688B2 (en) | Personalized digital image aesthetics in a digital medium environment | |
| EP3862893A1 (en) | Recommendation model training method, recommendation method, device, and computer-readable medium | |
| US20180053210A1 (en) | Personalization of Digital Content Recommendations | |
| CN108629608B (en) | User data processing method and device | |
| CN110008397B (en) | Recommendation model training method and device | |
| CN113254679B (en) | Multimedia resource recommendation method and device, electronic equipment and storage medium | |
| CN108446297B (en) | Recommendation method and device and electronic equipment | |
| CN111008335B (en) | Information processing method, device, equipment and storage medium | |
| CN112989169B (en) | Target object identification method, information recommendation method, device, equipment and medium | |
| CN112288471A (en) | Advertisement click rate prediction method based on user historical behavior sequence | |
| CN112036963B (en) | Webpage advertisement putting device and method based on multilayer random hidden feature model | |
| CN109190040B (en) | Collaborative evolution-based personalized recommendation method and device | |
| US20220172083A1 (en) | Noise contrastive estimation for collaborative filtering | |
| CN114117232A (en) | Recommendation method and system for enhancing comparative learning of nodes and meta-paths | |
| CN114493674A (en) | Advertisement click rate prediction model and method | |
| CN108763515B (en) | Time-sensitive personalized recommendation method based on probability matrix decomposition | |
| CN113836390A (en) | Resource recommendation method and device, computer equipment and storage medium | |
| CN115659005A (en) | Product pushing method and device, computer equipment and storage medium | |
| CN114298734A (en) | Internet online advertisement accurate delivery method based on multitask learning | |
| CN111831890A (en) | User similarity generation method and device, storage medium and computer equipment | |
| CN111414535B (en) | Method and device for recommending target object to user |
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 |