CN113344648B - Advertisement recommendation method and system based on machine learning - Google Patents
Advertisement recommendation method and system based on machine learning Download PDFInfo
- Publication number
- CN113344648B CN113344648B CN202110895354.7A CN202110895354A CN113344648B CN 113344648 B CN113344648 B CN 113344648B CN 202110895354 A CN202110895354 A CN 202110895354A CN 113344648 B CN113344648 B CN 113344648B
- Authority
- CN
- China
- Prior art keywords
- advertisement
- recommendation
- target
- user
- monitoring
- 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 42
- 238000010801 machine learning Methods 0.000 title claims abstract description 15
- 238000012544 monitoring process Methods 0.000 claims abstract description 88
- 230000005012 migration Effects 0.000 claims abstract description 20
- 238000013508 migration Methods 0.000 claims abstract description 20
- 238000006243 chemical reaction Methods 0.000 claims description 46
- 238000007477 logistic regression Methods 0.000 claims description 13
- 238000004422 calculation algorithm Methods 0.000 claims description 11
- 238000004364 calculation method Methods 0.000 claims description 10
- 239000011159 matrix material Substances 0.000 claims description 9
- 230000009466 transformation Effects 0.000 claims description 9
- 238000012549 training Methods 0.000 claims description 8
- 238000005457 optimization Methods 0.000 claims description 7
- 241000820057 Ithone Species 0.000 claims description 6
- 230000006870 function Effects 0.000 claims description 6
- 238000012512 characterization method Methods 0.000 claims description 4
- 230000004913 activation Effects 0.000 claims description 3
- 230000006978 adaptation Effects 0.000 claims description 3
- 238000013507 mapping Methods 0.000 claims description 3
- 238000010276 construction Methods 0.000 claims description 2
- 239000002699 waste material Substances 0.000 abstract description 5
- 230000001360 synchronised effect Effects 0.000 abstract description 4
- 230000006399 behavior Effects 0.000 description 4
- 230000008859 change Effects 0.000 description 4
- 230000009471 action Effects 0.000 description 3
- 238000011161 development Methods 0.000 description 3
- 238000004891 communication Methods 0.000 description 2
- 238000012937 correction Methods 0.000 description 2
- 230000003993 interaction Effects 0.000 description 2
- 230000007334 memory performance Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 238000013500 data storage Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000010365 information processing Effects 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 230000014759 maintenance of location Effects 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000001737 promoting effect Effects 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 238000012163 sequencing technique Methods 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0241—Advertisements
- G06Q30/0251—Targeted advertisements
- G06Q30/0269—Targeted advertisements based on user profile or attribute
- G06Q30/0271—Personalized advertisement
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/048—Activation functions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0241—Advertisements
- G06Q30/0277—Online advertisement
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Strategic Management (AREA)
- Finance (AREA)
- Development Economics (AREA)
- Accounting & Taxation (AREA)
- Computational Linguistics (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Data Mining & Analysis (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Game Theory and Decision Science (AREA)
- Entrepreneurship & Innovation (AREA)
- Economics (AREA)
- Marketing (AREA)
- General Business, Economics & Management (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses an advertisement recommendation method and system based on machine learning, which comprises the following steps: step S1, constructing an advertisement recommendation model; step S2, setting a monitoring coefficient for the recommendation characteristic set, and updating the advertisement recommendation scheme of the target user based on the monitoring coefficient; and step S3, providing the advertisement recommendation opinions to the target users at the advertisement push terminal based on the advertisement recommendation scheme and correcting the advertisement recommendation model. According to the method and the device, the interest migration information of the target user is obtained by setting the monitoring coefficient, the recommendation and the update are carried out on the user with the interest migration, the recommendation and the update are not needed to be carried out on the user without the interest migration, the original recommendation scheme is continuously used, the waste of operation resources and the extension of operation time caused by the synchronous update of advertisement recommendations of all users can be avoided, the operation efficiency of a recommendation system is effectively improved, and accurate and diversified advertisement recommendations can be integrally carried out according to the interest of the user.
Description
Technical Field
The invention relates to the technical field of advertisement recommendation, in particular to an advertisement recommendation method and system based on machine learning.
Background
The value of machine learning mainly focuses on data steering and information processing capability of data. In the current development of the industry, the arrival of the big data era brings better technical support for data conversion, data processing data storage and the like, and the industrial upgrading and the new birth form a promoting force, so that the big data can be automatically planned aiming at a program capable of finding objects, and the coordination of human users with computer information is realized.
With the development of internet technology, obtaining information, life, entertainment and work through the internet is becoming a part of people's life. In order to improve popularity and promote commodities, the merchants often put advertisements on the internet, and in the case of popular advertisement recommendation systems in China, when a user searches by using a certain keyword, the content related to the keyword appears on a search result page. Keywords are prominent on a search results page only when a particular keyword is retrieved.
Meanwhile, the appearance and popularization of the internet bring a large amount of information to users, and the requirements of the users on the information in the information age are met, but the amount of information on the internet is greatly increased along with the rapid development of the network, so that the users cannot obtain the part of information which is really useful for the users when facing a large amount of information, and the use efficiency of the information is reduced on the contrary, which is the so-called information overload problem. Therefore, the work demand and the interest content of each user are intelligently judged according to the retrieval records left on the Internet by each user, so that the advertisements with the same type are adaptively recommended, the effective delivery rate of the advertisements is improved, personalized services are provided for the users, an affinity relation can be established between the advertisements and the users, and the users can rely on the recommendation.
However, the existing advertisement recommendation method and system have the following problems: the advertisement recommendation is regularly updated for all users to adapt to the interest change of the target user, and in practice, the interest change of the users does not occur all the time, and due to the huge target user quantity and the characteristic data quantity, huge calculation pressure can be caused at the updating time, if the advertisement recommendation of all users is synchronously updated, only the waste of calculation resources and the extension of calculation time can be caused, the operation efficiency of a recommendation system is reduced, and the user experience is finally reduced.
Disclosure of Invention
The invention aims to provide an advertisement recommendation method and system based on machine learning, and aims to solve the technical problems that in the prior art, synchronous updating of advertisement recommendations of all users only causes waste of operation resources and extension of operation time, reduces the operation efficiency of a recommendation system, and finally reduces user experience.
In order to solve the technical problems, the invention specifically provides the following technical scheme:
a machine learning-based advertisement recommendation method comprises the following steps:
step S1, establishing a recommendation feature set by using the user features of the target users, the advertisement features of the target advertisements and the cross features of the target advertisements and the target users, and establishing an advertisement recommendation model based on the recommendation feature set, wherein the advertisement recommendation model is used for matching out an advertisement recommendation scheme of the target advertisements for the target users;
step S2, setting a monitoring coefficient for the recommendation characteristic set, and updating the advertisement recommendation scheme of the target user based on the monitoring coefficient;
and step S3, providing the advertisement recommendation opinions to the target users at the advertisement push terminal based on the advertisement recommendation scheme, and recording the conversion result of the advertisement recommendation scheme for correcting the advertisement recommendation model.
As a preferable aspect of the present invention, in step S1, the method for creating the recommended feature set includes:
and carrying out linear combination on the user characteristics, the advertisement characteristics and the cross characteristics to obtain memory characteristics, wherein the operation formula of the linear combination is as follows:
wherein,
，
characterized by the k-th memorability characteristic,
characterized by an ith one of the user feature, the advertisement feature, and the cross-feature,
characterized in that said ith feature does not participate in a linear combination of the kth memorability feature,
characterized in that the ith feature participates in a linear combination of the kth memorability feature,
characterized by a product operator, d characterized by a total dimension of the user feature, the advertisement feature, and the cross feature;
carrying out deep combination on the user characteristics, the advertisement characteristics and the cross characteristics to obtain the expansibility characteristics, wherein the operation formula of the deep combination is as follows:
wherein,
characterized by (A)l+1) the characteristics of the expandability of the layer,
is characterized bylThe topological character of the layer(s),
an activation function characterized as a combination of depths,
is characterized bylThe combined weight of the layers is determined,
is characterized bylThe combined bias of the layers is such that,
characterized by the user features, advertisement features, and cross-over features;
and converging the user characteristics, the advertisement characteristics, the cross characteristics, the memorability characteristics and the expansibility characteristics into the same set to be used as recommendation characteristics, and using the set containing the recommendation characteristics as a recommendation characteristic set.
As a preferred aspect of the present invention, the method for constructing the advertisement recommendation model includes:
selecting a positive sample item and a negative sample item from the recommendation feature set, wherein the positive sample item is a set item of a target user having a conversion result on the target advertisement, and the negative sample item is a set item of the target user not having a conversion result on the target advertisement;
carrying out sample training on the positive sample item and the negative sample item based on the logistic regression algorithm to construct an advertisement recommendation model, wherein the model formula of the advertisement recommendation model is as follows:
wherein,
characterized as the output of the ad recommendation model,
characterized by a logistic regression function and,
，
，
characterized by the k-th memorability characteristic,
characterized by the ith one of the user, advertisement and cross features, and n is characterized by
U () is characterized as a union operator,
is characterized by
And
the transpose operator of the combined features,
is characterized bylThe final value of (a) is,
is characterized by
Transpose operator of the layer expansibility feature, b is the bias of the advertisement recommendation model;
the output of the advertisement recommendation model is the predicted probability of the target user to the conversion result of the target advertisement, wherein,
when the prediction probability is higher than the probability threshold value of the logistic regression algorithm, the target user can generate a conversion result on the target advertisement;
and when the prediction probability is lower than the probability threshold value of the logistic regression algorithm, the target user does not generate a conversion result on the target advertisement.
As a preferable aspect of the present invention, the method for generating an advertisement recommendation scheme includes:
counting all target advertisements of the conversion result generated by each target user, and performing descending order arrangement on the target advertisements with the conversion result according to the advertisement profit value to generate an advertisement recommendation sequence chain belonging to each target user;
and sequentially recommending the target advertisements on the advertisement recommendation sequence chain to the corresponding target users.
As a preferable aspect of the present invention, in step S2, the specific method for setting the monitoring coefficient includes:
setting a monitoring interval, and monitoring all the recommended features of each target user for a feature value once after each monitoring interval, wherein the feature value monitoring is used for monitoring the interest migration attribute of the target user;
calculating the overall similarity between all recommended features of each target user after monitoring and all recommended features of each target user before monitoring as a monitoring coefficient of each target user, wherein the calculation formula of the monitoring coefficient is as follows:
wherein,
the characterization is that the listening coefficient is,
characterized by the total number of recommended features,
、
respectively characterized as the jth recommended feature after and before the monitoring.
As a preferred aspect of the present invention, in step S2, the method for updating the advertisement recommendation scheme of the target user based on the listening coefficient includes:
setting a monitoring threshold, and comparing the monitoring coefficient of each target user with the monitoring threshold, specifically:
if the monitoring coefficient is higher than the monitoring threshold value, the advertisement recommendation scheme corresponding to the target user does not need to be updated;
if the monitoring coefficient is lower than the monitoring threshold, the advertisement recommendation scheme corresponding to the target user needs to be updated, wherein:
calculating the single item similarity of each recommended feature of the corresponding target user after monitoring and each recommended feature of the corresponding target user before monitoring, and selecting all recommended features with the single item similarity higher than a monitoring threshold as recommended feature update chains of the corresponding target user, wherein the single item similarity calculation formula is as follows:
wherein,
characterized as the jth recommended feature after interception
And j recommendation feature before monitoring
Similarity of single items between the two;
and replacing the recommendation characteristic update chain to a corresponding recommendation characteristic item of the target user before monitoring, realizing the update of the recommendation characteristic representing the interest migration attribute of the target user, bringing all the recommendation characteristics of the target user after the update into the advertisement recommendation model, providing a new advertisement recommendation scheme for the target user, and realizing the adaptation of migrating the interest of the target user to the new interest.
As a preferred embodiment of the present invention, the specific method for modifying the advertisement recommendation model includes:
correcting the advertisement recommendation model by using a multi-objective optimization model by taking the AUC (acquired efficiency) index of the area under the model curve and the conversion rate of a target user as optimization indexes;
the conversion rate of the target users is the ratio of the number of the target users generating the conversion result for each target advertisement to the number of all the target users.
As a preferred aspect of the present invention, the step S1 further includes mapping the user feature, the advertisement feature and the cross feature to the same semantic space, wherein:
acquiring the size of a user characteristic graph corresponding to the user characteristic, the size of an advertisement characteristic graph corresponding to the advertisement characteristic and the size of a cross characteristic graph corresponding to the cross characteristic;
and performing matrix transformation on the convolutional layer with the user characteristic input convolutional kernel size as the user characteristic graph size, performing matrix transformation on the convolutional layer with the advertisement characteristic input convolutional kernel size as the advertisement characteristic graph size, and performing matrix transformation on the convolutional layer with the cross characteristic input convolutional kernel size as the cross characteristic graph size to convert the user characteristic, the advertisement characteristic and the cross characteristic into the same semantic space.
As a preferred aspect of the present invention, the present invention provides a recommendation system of the advertisement recommendation method based on machine learning, including:
the model unit is used for constructing an advertisement recommendation model and generating an advertisement recommendation scheme;
the monitoring unit is used for monitoring interest migration of a target user and updating a recommended scheme for the target user to adapt to new interest;
the recommendation unit comprises an advertisement push terminal, and the advertisement push terminal is used for providing advertisement recommendation opinions to the target users based on the advertisement recommendation scheme and recording conversion results of the advertisement recommendation scheme;
and the correcting unit is used for correcting the advertisement recommendation model according to the conversion result of the advertisement recommendation scheme.
As a preferred scheme of the invention, the model unit, the monitoring unit, the recommending unit and the correcting unit perform data interaction through network communication. .
Compared with the prior art, the invention has the following beneficial effects:
the invention utilizes linear combination and depth combination to discover memorability characteristics representing the accurate interest of the target user and expansibility characteristics representing the expansion interest of the target user in user characteristics, advertisement characteristics and cross characteristics, an advertisement recommendation model with memory performance and generalization performance is constructed based on the memory characteristic and the expansibility characteristic, accurate and diverse advertisement recommendations can be provided for users, and by setting the monitoring coefficient, the interest migration information of the target user is obtained, and the recommendation and update are carried out on the user with the interest migration, the recommendation and update are not needed to be carried out on the user without the interest migration, the original recommendation scheme is adopted, the method can avoid the waste of operation resources and the extension of operation time caused by the synchronous update of the advertisement recommendations of all users, effectively improve the operation efficiency of a recommendation system, and integrally and accurately recommend various advertisements according to the interests of the users.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below. It should be apparent that the drawings in the following description are merely exemplary, and that other embodiments can be derived from the drawings provided by those of ordinary skill in the art without inventive effort.
FIG. 1 is a flowchart of an advertisement recommendation method according to an embodiment of the present invention;
fig. 2 is a block diagram of a recommendation system according to an embodiment of the present invention.
The reference numerals in the drawings denote the following, respectively:
1-a model unit; 2-a monitoring unit; 3-a recommendation unit; 4-a correction unit.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
As shown in fig. 1, the present invention provides a method for recommending advertisements based on machine learning, comprising the following steps:
step S1, establishing a recommendation feature set by using the user features of the target users, the advertisement features of the target advertisements and the cross features of the target advertisements and the target users, and establishing an advertisement recommendation model based on the recommendation feature set, wherein the advertisement recommendation model is used for matching an advertisement recommendation scheme of the target advertisements for the target users;
in step S1, the method for creating the recommended feature set includes:
the user characteristics, the advertisement characteristics and the cross characteristics are linearly combined to obtain the memorability characteristics, and the operation formula of the linear combination is as follows:
wherein,
，
characterized by the k-th memorability characteristic,
characterized by an ith one of a user characteristic, an advertisement characteristic and a cross-over characteristic,
characterized in that the ith feature does not participate in the linear combination of the kth memorability feature,
characterized in that the ith feature participates in the linear combination of the kth memorability feature,
the characterization is a product operator, and the d is the total dimension of the user characteristic, the advertisement characteristic and the cross characteristic;
the user characteristics of the target user include attribute characteristics (such as gender, age, and the like) of the target user, interest characteristics (such as a commodity category of interest, a common APP category, and the like) of the target user, the advertisement characteristics of the target advertisement include attribute characteristics (such as a recommended commodity category, a profit value, and the like) of the target advertisement, behavior characteristics (such as playing duration, playing platform, and the like) of the target advertisement, and cross characteristics of the target advertisement and the target user, and the cross characteristics mainly refer to behavior characteristics (such as duration, number of times of watching the target advertisement, whether the target user purchases the recommended commodity, and the like) generated by the target user on the target advertisement.
The user characteristics, the advertisement characteristics and the cross characteristics comprise memory attributes of direct interest of the target users, and the strength of the memory attributes can be enhanced by linearly combining the user characteristics, the advertisement characteristics and the cross characteristics, so that the characteristics which frequently and simultaneously appear can be learned by carrying out advertisement model training on the memory characteristics which are linearly combined based on the user characteristics, the advertisement characteristics and the cross characteristics, the co-occurrence performance existing in historical data is explored, namely the direct interest of the target users is mined from the behavior actions of the target users, the accurate capture of the user interest can be realized, and an accurate advertisement recommendation scheme is made for the target users according to the direct interest of the target users.
If the contents recommended by the advertisements generated by model training only depending on the memorability characteristics are accurate contents, the interest of the user is convergent, the freshness is not generated, and the long-term retention of the user is not facilitated.
Carrying out deep combination on the user characteristics, the advertisement characteristics and the cross characteristics to obtain the expansibility characteristics, wherein the operation formula of the deep combination is as follows:
wherein,
characterized by (A)l+1) the characteristics of the expandability of the layer,
is characterized bylThe topological character of the layer(s),
an activation function characterized as a combination of depths,
is characterized bylThe combined weight of the layers is determined,
is characterized bylThe combined bias of the layers is such that,
characterized as user features, advertisement features, and cross-features;
the user characteristics, the advertisement characteristics and the cross characteristics also comprise generalization attributes of hidden interests of the target users, and the intensity of the generalization attributes can be enhanced by deeply combining the user characteristics, the advertisement characteristics and the cross characteristics, so that the training of an advertisement model based on the extension characteristics of linear combination of the user characteristics, the advertisement characteristics and the cross characteristics can explore and learn the characteristics which never appear, namely, the hidden interests of the target users are mined from the behavior actions of the target users, the hidden interests of the users can be captured, and diversified advertisement recommendation schemes can be formulated for the target users according to the hidden interests of the target users.
The generalization attribute is more prone to improving diversity of recommended contents than the memory attribute, if the recommended contents of the advertisement generated by model training only depending on the memory characteristics are too generalized, the accurate interest of the user cannot be met, and the risk of user loss is great. Compared with the accuracy of recommendation, the expansibility tends to improve the diversity of the recommendation system.
Therefore, only by combining the memorability characteristic and the expansibility characteristic, the advertisement recommendation model generated by training can obtain the accuracy and the expansibility of the recommendation result at the same time.
The user characteristics, the advertisement characteristics, the cross characteristics, the memorability characteristics and the expansibility characteristics are gathered into the same set to be used as recommendation characteristics, and the set containing the recommendation characteristics is used as a recommendation characteristic set.
The specific method for constructing the advertisement recommendation model comprises the following steps:
selecting a positive sample item and a negative sample item in the recommendation characteristic set, wherein the positive sample item is a set item of a target user having a conversion result on the target advertisement, and the negative sample item is a set item of the target user not having the conversion result on the target advertisement;
specifically, the positive sample items and the negative sample items are mixed according to an original rule, for example, if the original rule of the positive sample items and the negative sample items in the recommended feature set is that the positive sample items are far more than the negative sample items, the positive sample items are also selected and kept far more than the negative sample items for mixing, and if the original rule of the positive sample items and the negative sample items in the recommended feature set is that the positive sample items are far less than the negative sample items, the positive sample items are also selected and kept far less than the negative sample items for mixing.
The target advertisement generating the conversion result is in accordance with the interest of the target user in a large probability for the target user, and the target advertisement not generating the conversion result is in accordance with the interest of the target user in a small probability for the target user, so that the conversion result represents the interest of the user and serves as a distinguishing point of positive and negative sample items.
The construction method of the advertisement recommendation model comprises the following steps:
selecting a positive sample item and a negative sample item from the recommendation feature set, wherein the positive sample item is a set item of a target user having a conversion result on the target advertisement, and the negative sample item is a set item of the target user not having a conversion result on the target advertisement;
carrying out sample training on the positive sample item and the negative sample item based on a logistic regression algorithm to construct an advertisement recommendation model, wherein the model formula of the advertisement recommendation model is as follows:
wherein,
characterized as the output of the ad recommendation model,
characterized by a logistic regression function and,
，
，
characterized by the k-th memorability characteristic,
characterized by the ith one of a user feature, an advertisement feature and a cross feature, and n is characterized by
U () is characterized as a union operator,
is characterized by
And
the transpose operator of the combined features,
is characterized bylThe final value of (a) is,
is characterized by
Transpose operator of the layer expansibility feature, b is the bias of the advertisement recommendation model;
the method has the advantages that the logistic regression algorithm is utilized to establish the advertisement recommendation model, all target users can be operated in parallel, the operation speed is high, whether a conversion result is generated on a certain target advertisement can be visually displayed, and the method is suitable for the concurrent system for advertisement recommendation.
The output of the advertisement recommendation model is the predicted probability of the target user to the conversion result of the target advertisement, wherein,
when the prediction probability is higher than the probability threshold of the logistic regression algorithm, the target user can generate a conversion result on the target advertisement;
when the prediction probability is lower than the probability threshold of the logistic regression algorithm, the target user does not generate a conversion result on the target advertisement.
The method for generating the advertisement recommendation scheme comprises the following steps:
counting all target advertisements of a conversion result generated by each target user, and performing descending order arrangement on the target advertisements with the conversion result according to the advertisement profit value to generate an advertisement recommendation sequence chain belonging to each target user;
and sequentially recommending the target advertisements on the advertisement recommendation sequence chain to the corresponding target users.
In all the target advertisements of the predicted target users for generating the conversion results, the recommendation sequencing is carried out based on the maximum profit, the accurate and various user interests of the users can be met to the maximum extent, the profits of advertisement publishers can be guaranteed to the maximum extent, and the two purposes are achieved by one action.
Step S2, setting a monitoring coefficient for the recommendation characteristic set, and updating the advertisement recommendation scheme of the target user based on the monitoring coefficient;
in step S2, the specific method for setting the monitoring coefficient includes:
setting a monitoring interval, and monitoring all recommended features of each target user for a feature value once after each monitoring interval, wherein the feature value monitoring is used for monitoring the interest migration attribute of the target user;
calculating the overall similarity between all recommended features of each target user after monitoring and all recommended features of each target user before monitoring as a monitoring coefficient of each target user, wherein the calculation formula of the monitoring coefficient is as follows:
wherein,
the characterization is that the listening coefficient is,
characterized by the total number of recommended features,
、
respectively characterized as the jth recommended feature after and before the monitoring.
In step S2, the specific method for updating the advertisement recommendation scheme of the target user based on the monitoring coefficient includes:
setting a monitoring threshold, and comparing the monitoring coefficient of each target user with the monitoring threshold, specifically:
the monitoring threshold is self-defined by the user, and this embodiment is not limited.
If the monitoring coefficient is higher than the monitoring threshold value, the advertisement recommendation scheme corresponding to the target user does not need to be updated;
if the monitoring coefficient is lower than the monitoring threshold, the advertisement recommendation scheme corresponding to the target user needs to be updated, specifically:
calculating the single item similarity of each recommended feature of the corresponding target user after monitoring and each recommended feature of the corresponding target user before monitoring, and selecting all recommended features with the single item similarity higher than a monitoring threshold as recommended feature update chains of the corresponding target user, wherein the calculation formula of the single item similarity is as follows:
wherein,
characterized as the jth recommended feature after interception
And listeningPrevious jth recommendation feature
The similarity of the single items.
The single similarity and the listening coefficient are calculated by using the euclidean distance, and other calculation algorithms representing the similarity may also be used.
And replacing the recommendation characteristic update chain to the corresponding recommendation characteristic item of the target user before monitoring, realizing the update of the recommendation characteristic representing the interest migration attribute of the target user, bringing all the recommendation characteristics of the target user after the update into an advertisement recommendation model, providing a new advertisement recommendation scheme for the target user, and realizing the adaptation to the new interest of the target user.
Setting a monitoring coefficient, rapidly identifying whether interest migration of a target user occurs, triggering an updating mechanism if the interest migration occurs, changing and replacing the recommendation characteristics of the target user with larger change, keeping the original change smaller, realizing the updating of the recommendation characteristics of the target user at the moment, enabling the updated recommendation characteristics to reflect the recommendation characteristics of the changed interest characteristics and also have the recommendation characteristics of the original static interest characteristics, bringing the recommendation characteristics of the target user at the moment into an advertisement recommendation scheme obtained in an advertisement recommendation model, not only having target advertisements suitable for the new interest of the target user, but also having target advertisements of old interests which are not transferred by the target user to a certain extent, providing new recommendations for the target user, keeping a certain old recommendation, instead of pursuing to explore the new interests, discarding the old interests which are not transferred, and capturing the interests of the target user more comprehensively, the shopping psychology of the user is better met.
Only the recommendation features with large changes are updated, and the recommendation features with small changes are reserved, because the features with large changes can well meet the characteristics of interest migration, data calculation can be further reduced, the system operation pressure is reduced, the response speed is improved, and an advertisement recommendation scheme which is suitable for the new interests of the target user is quickly generated.
And step S3, providing advertisement recommendation to the target user based on the advertisement recommendation scheme, and recording the real generation data of the conversion result of the advertisement recommendation scheme to feed back to the advertisement recommendation model so as to modify the advertisement recommendation model.
The specific method for modifying the advertisement recommendation model comprises the following steps:
correcting the advertisement recommendation model by using a multi-objective optimization model by taking the AUC (acquired efficiency) index of the area under the model curve and the conversion rate of a target user as optimization indexes;
the conversion rate of the target users is the ratio of the number of the target users generating the conversion result for each target advertisement to the number of all the target users.
And performing multi-objective optimization correction on the advertisement recommendation model by using the AUC (effective product) index of the area under the model curve and the conversion rate of the target user, so as to obtain the advertisement recommendation model with the optimal performance.
Step S1 further includes mapping the user features, advertisement features, and cross features to the same semantic space, wherein:
acquiring the size of a user characteristic graph corresponding to the user characteristic, the size of an advertisement characteristic graph corresponding to the advertisement characteristic and the size of a cross characteristic graph corresponding to the cross characteristic;
and performing matrix transformation on the convolutional layer with the user characteristic input convolutional kernel size as the user characteristic graph size, performing matrix transformation on the convolutional layer with the advertisement characteristic input convolutional kernel size as the advertisement characteristic graph size, and performing matrix transformation on the convolutional layer with the cross characteristic input convolutional kernel size as the cross characteristic graph size to convert the user characteristic, the advertisement characteristic and the cross characteristic into the same semantic space.
The semantic space is unified, and the subsequent operation of non-numerical characteristics can be better realized.
Based on the advertisement recommendation method based on machine learning, the invention provides a recommendation system, which comprises the following steps:
the model unit 1 is used for constructing an advertisement recommendation model and generating an advertisement recommendation scheme;
the monitoring unit 2 is used for monitoring interest migration of the target user and updating a recommended scheme for the target user to adapt to new interest;
the recommendation unit 3 comprises an advertisement push terminal, and the advertisement push terminal is used for providing advertisement recommendation opinions to target users based on an advertisement recommendation scheme and recording conversion results of the advertisement recommendation scheme;
and the correcting unit 4 is used for correcting the advertisement recommendation model according to the conversion result of the advertisement recommendation scheme.
The model unit 1, the monitoring unit 2, the recommending unit 3 and the correcting unit 4 perform data interaction through network communication.
The invention utilizes linear combination and depth combination to discover memorability characteristics representing the accurate interest of the target user and expansibility characteristics representing the expansion interest of the target user in user characteristics, advertisement characteristics and cross characteristics, an advertisement recommendation model with memory performance and generalization performance is constructed based on the memory characteristic and the expansibility characteristic, accurate and diverse advertisement recommendations can be provided for users, and by setting the monitoring coefficient, the interest migration information of the target user is obtained, and the recommendation and update are carried out on the user with the interest migration, the recommendation and update are not needed to be carried out on the user without the interest migration, the original recommendation scheme is adopted, the method can avoid the waste of operation resources and the extension of operation time caused by the synchronous update of the advertisement recommendations of all users, effectively improve the operation efficiency of a recommendation system, and integrally and accurately recommend various advertisements according to the interests of the users.
The above embodiments are only exemplary embodiments of the present application, and are not intended to limit the present application, and the protection scope of the present application is defined by the claims. Various modifications and equivalents may be made by those skilled in the art within the spirit and scope of the present application and such modifications and equivalents should also be considered to be within the scope of the present application.
Claims (7)
1. A machine learning-based advertisement recommendation method is characterized by comprising the following steps:
step S1, establishing a recommendation feature set by using the user features of the target users, the advertisement features of the target advertisements and the cross features of the target advertisements and the target users, and establishing an advertisement recommendation model based on the recommendation feature set, wherein the advertisement recommendation model is used for matching out the advertisement recommendation scheme of the target advertisements for the target users, and the establishment method of the recommendation feature set comprises the following steps:
and carrying out linear combination on the user characteristics, the advertisement characteristics and the cross characteristics to obtain memory characteristics, wherein the operation formula of the linear combination is as follows:
wherein,
，
characterized by the k-th memorability characteristic,
characterized by an ith one of the user feature, the advertisement feature, and the cross-feature,
characterized in that said ith feature does not participate in a linear combination of the kth memorability feature,
characterized in that the ith feature participates in a linear combination of the kth memorability feature,
characterized by a product operator, d characterized by a total dimension of the user feature, the advertisement feature, and the cross feature;
carrying out deep combination on the user characteristics, the advertisement characteristics and the cross characteristics to obtain the expansibility characteristics, wherein the operation formula of the deep combination is as follows:
wherein,
characterized by (A)
+1) the characteristics of the expandability of the layer,
is characterized by
The topological character of the layer(s),
an activation function characterized as a combination of depths,
is characterized by
The combined weight of the layers is determined,
is characterized by
The combined bias of the layers is such that,
characterized by the user features, advertisement features, and cross-over features;
the user characteristics, the advertisement characteristics, the cross characteristics, the memorability characteristics and the expansibility characteristics are gathered into the same set to be used as recommendation characteristics, the set containing the recommendation characteristics is used as a recommendation characteristic set,
the construction method of the advertisement recommendation model comprises the following steps:
selecting a positive sample item and a negative sample item from the recommendation feature set, wherein the positive sample item is a set item of a target user having a conversion result on the target advertisement, and the negative sample item is a set item of the target user not having a conversion result on the target advertisement;
carrying out sample training on the positive sample item and the negative sample item based on a logistic regression algorithm to construct an advertisement recommendation model;
the output of the advertisement recommendation model is the predicted probability of the target user to the conversion result of the target advertisement, wherein,
when the prediction probability is higher than a probability threshold value of a logistic regression algorithm, a target user can generate a conversion result on the target advertisement;
when the prediction probability is lower than the probability threshold of the logistic regression algorithm, the target user does not generate a conversion result on the target advertisement;
step S2, setting a monitoring coefficient for the recommendation characteristic set, and updating the advertisement recommendation scheme of the target user based on the monitoring coefficient,
calculating the single item similarity of each recommended feature of the corresponding target user after monitoring and each recommended feature of the corresponding target user before monitoring, and selecting all recommended features with the single item similarity higher than a monitoring threshold as recommended feature update chains of the corresponding target user, wherein the single item similarity calculation formula is as follows:
wherein,
characterized as the jth recommended feature after interception
And j push before snoopingFeatures of recommendation
Similarity of single items between the two;
replacing the recommendation characteristic update chain to a corresponding recommendation characteristic item of the target user before monitoring, realizing the update of recommendation characteristics representing interest migration attributes of the target user, bringing all recommendation characteristics of the target user after the update into the advertisement recommendation model to provide a new advertisement recommendation scheme for the target user, and realizing the adaptation of migrating the interest of the target user to new interest;
and step S3, providing the advertisement recommendation opinions to the target users at the advertisement push terminal based on the advertisement recommendation scheme, and recording the conversion result of the advertisement recommendation scheme for correcting the advertisement recommendation model.
2. The method of claim 1, wherein the model formula of the advertisement recommendation model is as follows:
wherein,
characterized as the output of the ad recommendation model,
characterized by a logistic regression function and,
，
，
is characterized by the kth noteThe characteristics of the memory are that,
characterized by the ith one of the user, advertisement and cross features, and n is characterized by
U () is characterized as a union operator,
is characterized by
And
the transpose operator of the combined features,
is characterized by
The final value of (a) is,
is characterized by
Transpose operator of the layer's scalability features, b is the bias of the advertisement recommendation model.
3. The machine learning-based advertisement recommendation method according to claim 2, wherein: the method for generating the advertisement recommendation scheme comprises the following steps:
counting all target advertisements of the conversion result generated by each target user, and performing descending order arrangement on the target advertisements with the conversion result according to the advertisement profit value to generate an advertisement recommendation sequence chain belonging to each target user;
and sequentially recommending the target advertisements on the advertisement recommendation sequence chain to the corresponding target users.
4. The machine learning-based advertisement recommendation method according to claim 3, wherein: in step S2, the specific method for setting the monitoring coefficient includes:
setting a monitoring interval, and monitoring all the recommended features of each target user for a feature value once after each monitoring interval, wherein the feature value monitoring is used for monitoring the interest migration attribute of the target user;
calculating the overall similarity between all recommended features of each target user after monitoring and all recommended features of each target user before monitoring as a monitoring coefficient of each target user, wherein the calculation formula of the monitoring coefficient is as follows:
wherein,
the characterization is that the listening coefficient is,
characterized by the total number of recommended features,
、
respectively characterized as the jth recommended feature after and before the monitoring.
5. The method of claim 4, wherein the advertisement recommendation method based on machine learning is characterized in that: in step S2, the method for updating the advertisement recommendation scheme of the target user based on the listening coefficient includes:
setting a monitoring threshold, and comparing the monitoring coefficient of each target user with the monitoring threshold, specifically:
if the monitoring coefficient is higher than the monitoring threshold value, the advertisement recommendation scheme corresponding to the target user does not need to be updated;
and if the monitoring coefficient is lower than the monitoring threshold value, the advertisement recommendation scheme corresponding to the target user needs to be updated.
6. The method of claim 5, wherein the advertisement recommendation method based on machine learning is characterized in that: the specific method for modifying the advertisement recommendation model comprises the following steps:
correcting the advertisement recommendation model by using a multi-objective optimization model by taking the AUC (acquired efficiency) index of the area under the model curve and the conversion rate of a target user as optimization indexes;
the conversion rate of the target users is the ratio of the number of the target users generating the conversion result for each target advertisement to the number of all the target users.
7. The method of claim 6, wherein the step S1 further comprises mapping the user feature, the advertisement feature and the cross feature to a same semantic space, wherein:
acquiring the size of a user characteristic graph corresponding to the user characteristic, the size of an advertisement characteristic graph corresponding to the advertisement characteristic and the size of a cross characteristic graph corresponding to the cross characteristic;
and performing matrix transformation on the convolutional layer with the user characteristic input convolutional kernel size as the user characteristic graph size, performing matrix transformation on the convolutional layer with the advertisement characteristic input convolutional kernel size as the advertisement characteristic graph size, and performing matrix transformation on the convolutional layer with the cross characteristic input convolutional kernel size as the cross characteristic graph size to convert the user characteristic, the advertisement characteristic and the cross characteristic into the same semantic space.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110895354.7A CN113344648B (en) | 2021-08-05 | 2021-08-05 | Advertisement recommendation method and system based on machine learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110895354.7A CN113344648B (en) | 2021-08-05 | 2021-08-05 | Advertisement recommendation method and system based on machine learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113344648A CN113344648A (en) | 2021-09-03 |
| CN113344648B true CN113344648B (en) | 2021-11-30 |
Family
ID=77480847
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110895354.7A Active CN113344648B (en) | 2021-08-05 | 2021-08-05 | Advertisement recommendation method and system based on machine learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113344648B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114387041B (en) * | 2022-03-22 | 2022-06-17 | 北京鑫宇创世科技有限公司 | Multimedia data acquisition method and system |
| CN115542986B (en) * | 2022-11-24 | 2023-07-04 | 南通中铁华宇电气有限公司 | Environment simulation test system for lamp detection and sequential model building method |
Family Cites Families (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108629630B (en) * | 2018-05-08 | 2020-05-12 | 广州太平洋电脑信息咨询有限公司 | Advertisement recommendation method based on feature cross-combination deep neural network |
| CN110363590A (en) * | 2019-07-16 | 2019-10-22 | 深圳乐信软件技术有限公司 | A kind of advertisement recommended method, device, terminal and storage medium |
| CN111159564A (en) * | 2019-12-31 | 2020-05-15 | 联想(北京)有限公司 | Information recommendation method and device, storage medium and computer equipment |
| CN111598627A (en) * | 2020-05-26 | 2020-08-28 | 揭阳职业技术学院 | Personalized advertisement pushing method for elevator media terminal |
| CN111833096A (en) * | 2020-06-10 | 2020-10-27 | 北京龙云科技有限公司 | Advertisement recommendation method and system based on machine learning |
| CN111798280B (en) * | 2020-09-08 | 2020-12-15 | 腾讯科技（深圳）有限公司 | Multimedia information recommendation method, device and equipment and storage medium |
| CN112348592A (en) * | 2020-11-24 | 2021-02-09 | 腾讯科技（深圳）有限公司 | Advertisement recommendation method and device, electronic equipment and medium |
-
2021
- 2021-08-05 CN CN202110895354.7A patent/CN113344648B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN113344648A (en) | 2021-09-03 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US10783361B2 (en) | Predictive analysis of target behaviors utilizing RNN-based user embeddings | |
| CN111339415B (en) | Click rate prediction method and device based on multi-interactive attention network | |
| Zhu et al. | Online purchase decisions for tourism e-commerce | |
| CN112765480B (en) | Information pushing method and device and computer readable storage medium | |
| CN109783539A (en) | Usage mining and its model building method, device and computer equipment | |
| CN106104512A (en) | System and method for active obtaining social data | |
| CN113344648B (en) | Advertisement recommendation method and system based on machine learning | |
| CN106651544A (en) | Conversational recommendation system for minimum user interaction | |
| CN111651678B (en) | Personalized recommendation method based on knowledge graph | |
| CN111506820A (en) | Recommendation model, method, device, equipment and storage medium | |
| CN112749330A (en) | Information pushing method and device, computer equipment and storage medium | |
| CN111506821A (en) | Recommendation model, method, device, equipment and storage medium | |
| Xiong et al. | DNCP: An attention-based deep learning approach enhanced with attractiveness and timeliness of News for online news click prediction | |
| CN115618101A (en) | Streaming media content recommendation method and device based on negative feedback and electronic equipment | |
| WO2024061073A1 (en) | Multimedia information generation method and apparatus, and computer-readable storage medium | |
| CN111209725A (en) | Text information generation method and device and computing equipment | |
| CN110851694A (en) | Personalized recommendation system based on user memory network and tree structure depth model | |
| Sun | Music Individualization Recommendation System Based on Big Data Analysis | |
| CN116342228A (en) | Related recommendation method based on directed graph neural network | |
| CN114298118B (en) | Data processing method based on deep learning, related equipment and storage medium | |
| CN114547480A (en) | Deep learning recommendation method and system based on multi-platform fusion | |
| CN114996566A (en) | Intelligent recommendation system and method for industrial internet platform | |
| Liu et al. | Ada: adaptive depth attention model for click-through rate prediction | |
| Xiao et al. | Personalized Travel Product Recommendation Based on Embedding of Multi-Behavior Interaction Network and Product Information Knowledge Graph | |
| KR101985603B1 (en) | Recommendation method based on tripartite graph |
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 | ||
| CP03 | Change of name, title or address |
Address after: 101300 Room 5058, 5th Floor, Room 101, 1st to 6th Floor, Building 1, Yard 15, Xingtian Road, Shunyi District, Beijing Patentee after: Beijing Longyun Technology Co.,Ltd. Address before: 100160 floor 4, block C, Huaxia happiness innovation and entrepreneurship center, Fengtai District, Beijing Patentee before: Beijing Longyun Technology Co.,Ltd. |
|
| CP03 | Change of name, title or address | ||
| CP03 | Change of name, title or address |
Address after: Unit 1108A, 11th Floor, Building 3, No.16 Lize Road, Fengtai District, Beijing, 100000 Patentee after: Beijing Longyun Technology Group Co.,Ltd. Country or region after: China Address before: 101300 Room 5058, 5th Floor, Room 101, 1st to 6th Floor, Building 1, Yard 15, Xingtian Road, Shunyi District, Beijing Patentee before: Beijing Longyun Technology Co.,Ltd. Country or region before: China |
|
| CP03 | Change of name, title or address |