CN112446541A - Fusion classification model establishing method, marketing conversion rate gain prediction method and system - Google Patents

Fusion classification model establishing method, marketing conversion rate gain prediction method and system Download PDF

Info

Publication number
CN112446541A
CN112446541A CN202011348424.9A CN202011348424A CN112446541A CN 112446541 A CN112446541 A CN 112446541A CN 202011348424 A CN202011348424 A CN 202011348424A CN 112446541 A CN112446541 A CN 112446541A
Authority
CN
China
Prior art keywords
marketing
conversion
user
group
classification model
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202011348424.9A
Other languages
Chinese (zh)
Inventor
盛冲冲
樊婧逸
李健
张琛
万化
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shanghai Pudong Development Bank Co Ltd
Original Assignee
Shanghai Pudong Development Bank Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Shanghai Pudong Development Bank Co Ltd filed Critical Shanghai Pudong Development Bank Co Ltd
Priority to CN202011348424.9A priority Critical patent/CN112446541A/en
Publication of CN112446541A publication Critical patent/CN112446541A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06QINFORMATION 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
    • G06Q10/00Administration; Management
    • G06Q10/04Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00Pattern recognition
    • G06F18/20Analysing
    • G06F18/24Classification techniques
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00Pattern recognition
    • G06F18/20Analysing
    • G06F18/25Fusion techniques
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/04Architecture, e.g. interconnection topology
    • G06N3/045Combinations of networks
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/08Learning methods
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06QINFORMATION 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
    • G06Q10/00Administration; Management
    • G06Q10/06Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/067Enterprise or organisation modelling

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Business, Economics & Management (AREA)
  • Data Mining & Analysis (AREA)
  • General Physics & Mathematics (AREA)
  • Strategic Management (AREA)
  • Human Resources & Organizations (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Economics (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Biomedical Technology (AREA)
  • Game Theory and Decision Science (AREA)
  • Computational Linguistics (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Evolutionary Biology (AREA)
  • Development Economics (AREA)
  • Health & Medical Sciences (AREA)
  • Biophysics (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Marketing (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Tourism & Hospitality (AREA)
  • General Business, Economics & Management (AREA)
  • Educational Administration (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The text provides a fusion classification model establishing method, a marketing conversion rate gain prediction method and a system, wherein the marketing conversion rate gain prediction method comprises the following steps: determining conversion results of users in a control group and a marketing group to products related to the marketing activities after the marketing activities are carried out for a preset time period; converting the conversion result into a conversion variable by using a preset conversion rule for balancing the sample; taking the characteristics of the users in the control group and the marketing group as input, taking the value probability of the conversion variable of the users as output, and training to obtain a fusion classification model; calculating to obtain the value probability of the conversion variable of the user to be analyzed according to the characteristics of the user to be analyzed and the fusion classification model; and calculating to obtain the marketing conversion rate gain according to the probability of the value of the user conversion variable to be analyzed. The method can directly calculate the gain of the marketing conversion rate, enables the data of the control group and the marketing group to be fused, avoids error amplification caused by a double classification model, and can also relieve the problem of unbalance of positive samples and negative samples.

Description

Fusion classification model establishing method, marketing conversion rate gain prediction method and system
Technical Field
The present disclosure relates to the field of intelligent marketing, and in particular, to a method for building a fusion classification model, and a method and a system for predicting a marketing conversion rate gain.
Background
In the field of intelligent marketing, how to measure and predict increment promotion (Uplift) brought by marketing activities is to avoid wasting marketing budget on the parts of people who will transform originally, are immoderate to marketing and have negative marketing feedback, so that the method is the most important challenge of an intelligent marketing algorithm, and a marketing transformation rate gain prediction model is urgently needed to be constructed to improve the effect of marketing activities. The existing mainstream method for marketing conversion rate gain modeling is a differential response dual-model method based on a traditional machine learning method, and the method mainly establishes two classification models:
(1) constructing a classification model on a marketing (record G ═ T) sample: pT(Y|X)
(2) Another classification model was constructed on Control Group (denoted G ═ C) samples: pC(Y|X)
Where X is a characteristic of the user, Y is whether to convert (i.e., the result of the conversion, e.g., response, purchase, etc.), PTIndicating the probability of whether the user has converted in the marketing group. PCIndicating the probability of whether the user transformed in the control group. Determining the marketing conversion rate gain of a certain user as the difference between the conversion probability of marketing and the conversion probability of not marketing, namely:
Δ(X)=P(Y=1|X,G=T)-P(Y=1|X,G=C)=PT(Y=1|X)-PC(Y=1|X);
where P (Y ═ 1| X, G ═ T) represents the probability of user X transitioning in the marketing group, P (Y ═ 1| X, G ═ C) represents the probability of X transitioning in the control group, and P represents the probability of X transitioning in the control groupT(Y ═ 1| X) is shorthand for P (Y ═ 1| X, G ═ T), PC(Y ═ 1| X) is a abbreviation for P (Y ═ 1| X, G ═ C).
The dual-model method of differential response in the prior art has the following disadvantages:
(1) the dual model of the differential response determines marketing conversion gain by subtracting the predicted values of the two models, the two independent models are trained separately and easily generate accumulated errors, the incremental signal is a smaller value, the requirement on the accuracy of the two models is high, and otherwise the errors are easily amplified. In addition, because the conversion rate in marketing activities is often very low, positive samples (samples with successful conversion) are fewer, and are even more than an order of magnitude different than negative samples (samples without conversion), which causes the problem that two independent models are unbalanced in positive and negative samples.
(2) By adopting the traditional machine learning method, the requirements on the characteristic engineering are strict, and the performance of the machine learning system can be improved only by needing good characteristic design. Because the traditional machine learning method usually needs manual characteristic engineering design, the problem of low efficiency exists. The current online marketing activities have high frequency, fast demand change, large data accumulation and various data types, and the manual design mainly depends on the prior knowledge of designers, so that the complex demand is difficult to support, and the advantages of large data cannot be utilized. In addition, the manual parameter adjusting mode has the problem that the number of parameters is very limited.
Disclosure of Invention
The method is used for solving the problem that in the prior art, the marketing conversion rate gain determination depends on two classification models, and the problem of inaccurate prediction exists due to large errors of the classification models, so that over-delivery or less-delivery is caused, and partial user experience is poor. In addition, the problem of imbalance of positive and negative samples based on the establishment process of the existing classification model exists.
In order to solve the above technical problem, a first aspect of the present disclosure provides a fusion classification model building method, including:
determining a comparison group of the marketing activities and conversion results of users in the marketing groups to products related to the marketing activities;
converting the conversion result into a conversion variable by using a preset conversion rule for balancing the sample;
taking the characteristics of the users in the control group and the marketing group as input, taking the value probability of the conversion variables of the users as output, and training by adopting a deep learning method to obtain a fusion classification model;
the conversion variable values comprise a first value and/or a second value, the first value indicates that the user conversion result in the marketing group is converted or the user conversion result in the control group is not converted, and the second value indicates that the user conversion result in the marketing group is not converted and/or the user conversion result in the control group is converted.
In a further embodiment of this document, before determining the conversion results of the control group of the marketing campaign and the users in the marketing group to the products associated with the marketing campaign, the method further comprises:
determining user characteristics, a comparison group and a marketing group of the marketing campaign according to the characteristics of the marketing campaign;
wherein the number of users in the control group is the same as that in the marketing group, and the feature distribution of the users is consistent.
In a further embodiment herein, the predetermined transformation rules for balancing the samples comprise rules expressed by the following formula:
Figure BDA0002800662560000021
wherein Z represents a conversion variable, T represents an operation group, C represents a control group, Y represents a conversion result, G ═ T represents that a user comes from the operation group, G ═ C represents that the user comes from the control group, Y ═ 1 represents that the conversion result is, Y ═ 0 represents that the conversion result is unconverted, Z ═ 1 represents a first value, and Z ═ 0 represents a second value.
In a further embodiment of the present disclosure, the features of the users in the control group and the marketing group are used as inputs, the transition variable value probabilities of the users are used as outputs, and a deep learning method is adopted to train and obtain a fusion classification model, which includes:
establishing an N-layer neural network model, wherein the first N-1 layer adopts a ReLU function, and the last layer adopts a Sigmod activation function;
initializing the neural network model;
the characteristics of the users in the comparison group and the marketing group are used as input, and the conversion variable value probability of the users is used as output;
constructing a loss function according to the input, output and neural network model;
optimizing the neural network model according to the loss function;
and the optimized neural network model is the fusion classification model.
A second aspect herein provides a marketing conversion gain prediction method, comprising:
establishing a fusion classification model obtained by the method of any one of the embodiments according to the characteristics of the activity to be marketed;
calculating to obtain the value probability of the user conversion variable to be analyzed according to the characteristics of the user to be analyzed and the fusion classification model;
and calculating to obtain the marketing conversion rate gain according to the probability of the conversion variable value of the user to be analyzed.
In a further embodiment of the present invention, the calculating the marketing conversion rate gain according to the value probability of the user conversion variable to be analyzed includes calculating the marketing conversion rate gain by using the following formula:
Δ (X) ═ 2 × P (Z ═ 1| X) -1; or
Δ(X)=1-2×P(Z=0|X);
Wherein, X represents the feature of the user to be analyzed, Z ═ 1 represents a first value, P (Z ═ 1| X) represents the probability that the user to be analyzed converts the variable into the first value, and P (Z ═ 0| X) represents the probability that the user to be analyzed converts the variable into the second value.
In further embodiments herein, the marketing conversion gain method further comprises:
and judging whether the marketing conversion rate gain is greater than 0, if so, determining that the user to be analyzed is the user interested in the marketing activity, and providing the marketing activity to the user to be analyzed.
In a third aspect of this document, there is also provided a fusion classification model building system, including:
the analysis module is used for determining a comparison group of the marketing activities and conversion results of users in the marketing groups to products related to the marketing activities;
the data conversion processing module is used for converting the conversion result into a conversion variable by utilizing a preset conversion rule for balancing the sample;
the model construction and training module is used for taking the characteristics of the users in the control group and the marketing group as input, taking the value probability of the conversion variables of the users as output, and training by adopting a deep learning method to obtain a fusion classification model;
the conversion variable values comprise a first value and/or a second value, the first value indicates that the user conversion result in the marketing group is converted or the user conversion result in the control group is not converted, and the second value indicates that the user conversion result in the marketing group is not converted and/or the user conversion result in the control group is converted.
In a fourth aspect herein, there is also provided a marketing conversion gain prediction system comprising:
the classification model establishing module is used for establishing the fusion classification model of any one of the embodiments according to the characteristics of the activity to be marketed;
the model calculation module is used for calculating the value probability of the conversion variable of the user to be analyzed according to the characteristics of the user to be analyzed and the fusion classification model;
and the gain calculation module is used for calculating to obtain the marketing conversion rate gain according to the value probability of the user conversion variable to be analyzed.
In a fifth aspect of the present disclosure, there is also provided a computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the fusion classification model building method or the marketing conversion rate gain prediction method according to any of the foregoing embodiments when executing the computer program.
In a sixth aspect of the present disclosure, a computer-readable storage medium is further provided, where the computer-readable storage medium stores a computer program for implementing the fusion classification model building method or the marketing conversion rate gain prediction method according to any one of the foregoing embodiments when the computer program is executed by a processor.
According to the fusion classification model establishing method, the marketing conversion rate gain predicting method and the system, the fusion classification model established by the conversion variables is introduced, the marketing conversion rate gain can be accurately determined, further user classification is achieved, and only the marketing activities are released to the users interested in the marketing activities subsequently, so that the marketing activity releasing efficiency can be improved, disturbance to the users not interested in the marketing activities is reduced, and user experience is improved. And the data of the control group and the marketing group can be fused, so that the error amplification caused by the double-classification model is avoided. In addition, the introduction of the conversion variable can also alleviate the problem of unbalance of positive samples (probability of conversion of marketing activity) and negative samples (probability of non-conversion of marketing activity).
In order to make the aforementioned and other objects, features and advantages of the present invention comprehensible, preferred embodiments accompanied with figures are described in detail below.
Drawings
In order to more clearly illustrate the embodiments or technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art that other drawings can be obtained according to the drawings without creative efforts.
FIG. 1 illustrates a first flowchart of a fusion classification model building method according to an embodiment herein;
FIG. 2 shows a second flowchart of a fusion classification model building method according to an embodiment of the present disclosure;
FIG. 3 illustrates a flow diagram of a fused classification model training process according to an embodiment herein;
FIG. 4 illustrates a first flowchart of a marketing conversion gain prediction method of embodiments herein;
FIG. 5 illustrates a second flowchart of a marketing conversion gain prediction method of embodiments herein;
FIG. 6 is a block diagram illustrating a marketing fusion classification model building system according to an embodiment of the present disclosure;
FIG. 7 illustrates a first architectural diagram of a marketing conversion gain prediction system of an embodiment herein;
FIG. 8 illustrates a second block diagram of a marketing conversion gain prediction system of an embodiment herein;
FIG. 9 is a flowchart of a marketing campaign object determination process, according to a specific embodiment herein;
FIG. 10 is a block diagram illustrating a computer device according to an embodiment of the present disclosure.
Description of the symbols of the drawings:
600. a data acquisition module;
610. an analysis module;
620. a data conversion processing module;
630. a model building and training module;
710. a classification model establishing module;
720. a model calculation module;
730. a gain calculation module;
740. a judgment module;
1002. a computer device;
1004. a processor;
1006. a memory;
1008. a drive mechanism;
1010. an input/output module;
1012. an input device;
1014. an output device;
1016. a presentation device;
1018. a graphical user interface;
1020. a network interface;
1022. a communication link;
1024. a communication bus.
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 herein without making any creative effort, shall fall within the scope of protection.
Marketing conversion gain in the prior art is determined by the following formula:
Δ(X)=PT(Y=1|X)-PC(Y=1|X);
wherein Δ (X) represents marketing conversion gain, PT(Y ═ 1| X) denotes the probability of user conversion in the marketing group, PC(Y ═ 1| X) represents the probability of user conversion in the control group.
PT(Y-1 | X) and PCThe method has the advantages that (Y is 1| X) depends on two independent classification models respectively, the marketing conversion rate gain calculation accuracy is low due to the fact that errors of the classification models are large, and in addition, the accuracy of the classification models is low due to the fact that positive samples and negative samples which are used as basis in the building process of the existing classification models are not balanced.
The purpose of the marketing campaign is to select the most suitable marketing object from the user group, and the marketing objects can be classified into the following four categories:
(1) the users are sensitive to the marketing activities and are easily influenced by the marketing activities;
(2) the conversion is unrelated to the marketing activity, and the users in the category are naturally converted and unrelated to the marketing activity;
(3) non-conversion and immoderate to marketing campaigns;
(4) such users are susceptible to negative impact of the marketing campaign by not converting or otherwise losing the marketing campaign.
Among the above four types of people, the first type of people is the target crowd of the marketing campaign, i.e., the interested crowd, the second and third types of people are the crowd with no marketing effect, unnecessary costs (inecessary costs) can be wasted when marketing is performed on the crowd, and the fourth type of people is the crowd with adverse effect on the marketing campaign, and the marketing campaign should be actively avoided from being performed on the crowd.
In view of the above technical drawbacks of the prior art and the identification of the above four types of people, an embodiment of the present disclosure, as shown in fig. 1, provides a fusion classification model building method, including:
step 110, determining a comparison group of the marketing activities and conversion results of users in the marketing groups to products related to the marketing activities;
step 120, converting the conversion result into a conversion variable by using a predetermined conversion rule for balancing the sample, wherein the conversion variable comprises a first value and a second value, the first value indicates that the user conversion result in the marketing pin group is converted or the user conversion result in the control group is not converted, and the second value indicates that the user conversion result in the marketing pin group is not converted and/or the user conversion result in the control group is converted;
and step 130, taking the characteristics of the users in the comparison group and the marketing group as input, taking the value probability of the conversion variable of the corresponding user as output, and training by adopting a deep learning method to obtain a fusion classification model. The conversion variable value comprises a first value and/or a second value, and the sum of the probability of the first value and the probability of the second value of the conversion variable is 1.
In detail, the marketing campaign described herein may be determined according to the applied business scenario, and the specific content of the marketing campaign differs from the applied business scenario, and taking the banking field as an example, the marketing campaign is, for example, an insurance business, a financial product recommendation, an online and offline consumption campaign, and the like, and the specific content of the marketing campaign is not limited herein.
The fusion classification model establishing method is suitable for the same type of marketing activities, and different fusion classification models are required to be established for different types of marketing activities.
The conversion result in the step 110 may be determined after the marketing campaign provides the marketing group users with a predetermined time period, for example, half a year, etc., which may be determined according to practical situations, and is not limited herein. The control group and the users in the marketing group are determined according to the characteristics of the marketing campaign, specifically, the control group is used for comparing the marketing effect of the marketing campaign, the users in the control group do not receive the propaganda of the marketing campaign, the marketing group is used for actually generating the marketing effect, and the users in the marketing group receive the propaganda of the marketing campaign.
The conversion of the marketing campaign by the user described herein includes, but is not limited to, purchasing, responding to the marketing campaign, depending on the marketing content, and the conversion results include: and (3) transformation and non-transformation, wherein in the specific implementation, 1 is used for transformation, and 0 is used for non-transformation. For example, if a user in the marketing group successfully converts (e.g., purchases, responds to, etc. a product associated with the marketing campaign), the conversion result for that user is set to 1, whereas if a user in the marketing group does not convert (e.g., does not purchase, does not respond to, etc. a product associated with the marketing campaign), the conversion result for that user is set to 0. For another example, if a user in the control group still successfully converts without being marketed, the conversion result of the user is set to 1, whereas if a user in the control group does not convert without being marketed, the conversion result of the user is set to 0.
The predetermined conversion rules for the balance samples described herein are used to balance the balance of the converted variable samples and to combine the data in the control and marketing groups. Any rule that can balance the samples can be used as the predetermined transformation rule for balancing the samples, which is not specifically limited herein.
The user characteristics used in modeling can be set according to the marketing activity requirements, taking the banking field as an example, the user characteristics include attribute quantity and statistic quantity, wherein the attribute quantity includes but is not limited to age, attribute, income, work type and the like, the statistic quantity includes but is not limited to historical consumption amount, historical transfer, loan amount and the like, and the specific content and number of the user characteristics are not limited herein.
The fusion classification model establishing method provided by this embodiment may be operated in an intelligent terminal and a server, including a smart phone, a tablet computer, a desktop computer, and the like, and the specific implementation manner is not limited herein. The configuration requirements of the intelligent terminal and the server are not high, and the configuration of the conventional dual-core 8GB RAM can be realized.
In the embodiment, the fusion classification model established by introducing the conversion variables can directly calculate the gain of the marketing conversion rate, and can fuse the data of the control group and the marketing group, thereby avoiding error amplification caused by the double classification model.
In an embodiment of this document, as shown in fig. 2, before the step 110 is executed, the method further includes: and step 100, determining the user characteristics, the marketing activity control group and the marketing group according to the marketing activity characteristics. The number of users in the control group is the same as that in the marketing group, and the feature distribution of the users is consistent.
In specific implementation, the matching relationship between the user characteristics and the marketing activity characteristics can be established by analyzing the characteristics of the users related to the historical marketing activity, and the characteristics of the users can be quickly determined according to the marketing activity characteristics and the matching relationship.
When the marketing group and the comparison group are selected, if the user characteristics of the marketing group and the comparison group are not distributed consistently, the assumption (P (G | X) ═ P (G)) that the marketing strategy and the user characteristics are independent mutually is not satisfied, therefore, the marketing group and the comparison group can randomly select a certain proportion of users from all user data according to a certain proportion to form the comparison group and the user group, or the marketing group and the comparison group are screened by a certain marketing strategy.
The embodiment can meet the independence of user characteristics and marketing strategies (whether marketing or not), improve the comparison effect of the comparison group and improve the precision of the fusion classification model. Meanwhile, the user characteristics are determined by utilizing big data analysis, and the design of complex characteristics can be avoided.
In one embodiment, for ease of calculation and representation, the predetermined transformation rules for balancing the samples include rules expressed by the following formula:
Figure BDA0002800662560000091
wherein Z represents a conversion variable, T represents an operation group, C represents a control group, Y represents a conversion result, G ═ T represents that a user comes from the operation group, G ═ C represents that the user comes from the control group, Y ═ 1 represents that the conversion result is conversion, Y ═ 0 represents that the conversion result is not conversion, Z ═ 1 represents a first value, and Z ═ 0 represents a second value.
In other embodiments, the specific numerical values in the above conversion rules may be replaced by other numerical values or characters, which are not specifically limited herein.
As can be seen from the above conversion rule, the positive samples (Z ═ 1 indicates positive samples) of the conversion variable Z include samples in the control group in which the user conversion result is not converted, and the negative samples (Z ═ 0 indicates positive samples) of the conversion variable Z include samples in the marketing group which are not converted, but the samples which are not converted often account for a large part, so the positive and negative samples of the conversion variable Z are relatively balanced.
According to the above rules, a calculation formula of the marketing conversion rate gain can be derived by combining probability theory, specifically, assuming that the intervention strategy G is independent from the user, i.e. G is independent from the user characteristic XP (G | X) ═ p (G):
P(Z=1|X)=P(Z=1|X,G=T)×P(G=T)+P(Z=1|X,G=C)×P(G=C)
=P(Y=1|X,G=T)×P(G=T)+P(Y=0|X,G=C)×P(G=C)
=PT(Y=1|X)×P(G=T)+PC(Y=0|X)×P(G=C) (1)
where P (Z ═ 1| X) represents the probability that the user conversion result in the marketing group is converted and the user conversion result in the control group is unconverted, that is, the probability that Z is 1. P (Z ═ 1| X, G ═ T) ═ P (Y ═ 1| X, G ═ T ═ Y ═ 1| X) denotes the probability that the user conversion result in the pin group is conversion, P (G ═ T) denotes the probability that the user belongs to the pin group, P (Z ═ 1| X, G ═ C) ═ P (Y ═ 0| X, G ═ C) ═ PC(Y ═ 0| X) represents the probability that the user conversion result in the control group is unconverted, and P (G ═ C) represents the probability that the user belongs to the control group.
As can be seen from the setting of the marketing group and the comparison group, the general marketing group and the comparison group are set to be equal in number, and P (G ═ T) ═ P (G ═ C) ═ 0.5, that is, the probability that one user is assigned to the marketing group and the comparison group is equal. Under this assumption, (1) can be rewritten as:
P(Z=1|X)=0.5×(PT(Y=1|X)+PC(Y=0|X))
=0.5×(PT(Y=1|X)+1-PC(Y=1|X)) (2)
(2) multiplying both sides of the equation by 2 can deduce:
Δ(X)=PT(Y=1|X)-PC(Y=1|X)=2×P(Z=1|X)-1 (3)
where Δ (X) represents the marketing conversion rate gain, and P (Z ═ 1| X) represents the probability that the user conversion result in the marketing group is conversion and the user conversion result in the control group is non-conversion.
As can be seen from equation (3), the calculation of the marketing conversion gain can be achieved by only knowing the probability of one variable Z being 1, i.e., by directly modeling Z (which is equivalent to optimizing P (Z being 1| X)), without separately modeling the marketing group PT(Y-1 | X) and control group PC(Y ═ 1| X) was modeled separately.
The user data of the marketing group and the control group can be directly merged and trained by adding a conversion variable, and a model is used for modeling, so that the data level and the model level are communicated.
In an embodiment of the present disclosure, to facilitate online deployment and one-step calculation, the fusion classification model and the calculation result of formula (3) may be combined together to form a marketing conversion rate gain model, and the marketing conversion rate gain model is deployed online as a whole.
To verify the effectiveness of the marketing conversion rate gain model, in a further embodiment, the marketing conversion rate gain model is evaluated by using a conventional AUC (Area Under the ROC Curve) index, and if the AUC result is not ideal, the step 120 may be returned to adjust the predetermined conversion rule for balancing the samples until the AUC result is ideal (e.g., reaches a set value). Taking the following predetermined conversion rule for the balance samples as an example for testing, the AUC of the test set can reach 0.72, and the result shows that the predetermined conversion rule for the balance samples is properly selected, and the marketing conversion rate gain model has higher accuracy.
Figure BDA0002800662560000101
Wherein, Z represents a conversion variable, T represents a marketing group, C represents a control group, Y represents a conversion result, G ═ T represents that a user comes from the marketing group, G ═ C represents that the user comes from the control group, Y ═ 1 represents conversion, Y ═ 0 represents non-conversion, Z ═ 1 represents that the user conversion result in the marketing group is conversion and the user conversion result in the control group is non-conversion, Z ═ 0 represents that the user conversion result in the marketing group is non-conversion and/or the user conversion result in the control group is conversion.
In an embodiment of this document, as shown in fig. 3, in the step 130, the features of the users in the control group and the marketing group are used as inputs, the transition variable value probabilities of the users are used as outputs, and a deep learning method is adopted to train and obtain a fusion classification model, which includes:
step 310, establishing an N-layer neural network model, wherein the first N-1 layer adopts a ReLU function, and the last layer adopts a Sigmod activation function;
step 320, initializing a neural network model;
step 330, taking the characteristics of the users in the comparison group and the marketing group as input, and converting the value probability of the variable as output;
step 340, constructing a loss function (Sigmoid cross entry) according to the input, the output and the neural network model;
step 350, optimizing a neural network model according to the loss function;
and step 360, optimizing to obtain the neural network model which is the fusion classification model.
In a specific embodiment, after tuning, the neural network model established in step 310 finally selects 4 layers of neural networks, and the first 3 layers adopt a RuLU function (Rectified Linear Unit). In other embodiments, the number of layers of the neural network model may be set according to requirements, which is not limited herein. In the embodiment, the calculation of the modeling process can be simplified by setting the front N-3 layers as the RuLU function, the calculation speed of modeling is improved, the disappearance of the gradient is prevented, the network has sparsity, and overfitting is reduced. The solution can be facilitated by setting the last layer to a Sigmod activation function, ensuring data amplitude.
In one embodiment, the step 320 initializes the neural network model using Kaiming uniform distribution.
In one specific embodiment, in the step 330, the transition variable value probability includes P (Z ═ 1| X) and/or P (Z ═ 0| X), where P (Z ═ 1| X) represents a probability that the user transition result in the marketing group is the transition and the user transition result in the control group is the non-transition, and P (Z ═ 0| X) represents a probability that the user transition result in the marketing group is the non-transition and/or the user transition result in the control group is the transition. The characteristics of the users may be determined according to the characteristics of the marketing campaign, and the characteristics of the users of the same type of marketing campaign may be the same, which is not limited herein.
The loss function in step 340 is, for example, a binary cross-entropy loss function. This embodiment is through setting up the loss function into two categorised cross entropy loss functions, can be convenient for derive, improves data and seeks speed. In a specific implementation, in the step 340, other types of loss functions may be further selected, which is not limited herein.
In some embodiments, in step 350, an Adam (Adaptive moment estimation) optimizer can be used to optimize the neural network model, and the optimization method can accelerate the fitting of high-dimensional data without setting the attenuation of step length, so that the obtained parameter value is relatively stable. In other embodiments, algorithms such as SGD may be selected according to a random gradient descent method, which is not specifically limited herein.
And the finally optimized neural network model is a fusion classification model, and the fusion classification model can calculate the probability of the value of the conversion variable according to the characteristics of the user. According to the probability of the value of the conversion variable, and by combining the formula (3), the marketing conversion rate gain can be rapidly obtained.
In the specific implementation of this embodiment, the data (including the user characteristics and the transformation variables) obtained by the analysis in step 120 may be divided into three parts: training set, verification set and test set. For example, the training set accounts for 60%, the validation set accounts for 20%, and the test set accounts for 20%. In practice, the training set, the validation set, and the test set may have other ratios, which is not limited herein.
In an embodiment of this document, there is also provided a marketing conversion rate gain prediction method, as shown in fig. 4, including:
step 410, establishing a fusion classification model established by the fusion classification model establishing method according to any one of the embodiments according to the characteristics of the activity to be marketed;
step 420, calculating to obtain the value probability of the conversion variable of the user to be analyzed according to the characteristics of the user to be analyzed and the fusion classification model;
and 430, calculating the marketing conversion rate gain according to the probability of the conversion variable value of the user to be analyzed.
In detail, the process of building the fusion classification model in step 410 is described in the foregoing embodiments, and will not be described in detail here.
In step 420, the user characteristics to be analyzed are input into the fusion classification model, and the fusion classification model outputs the value probability of the user transformation variable to be analyzed. In one embodiment, if the values of the transition variable Z are 1 and 0, the transition variable values probabilities are P (Z ═ 1| X) and P (Z ═ 0| X).
The method and the device can accurately and efficiently calculate the marketing conversion rate gain, and provide a basis for subsequent customer classification and marketing activity object determination.
In this embodiment, in the step 430, calculating the marketing conversion rate gain according to the probability of the user conversion variable value to be analyzed includes calculating the marketing conversion rate gain by using the following formula:
Δ(X)=2×P(Z=1|X)-1;
wherein, X represents the feature of the user to be analyzed, Z ═ 1 represents a first value, and P (Z ═ 1| X) represents the probability that the user to be analyzed converts the variable into the first value.
In other embodiments, the above formula may be modified as follows:
Δ(X)=2×(1-P(Z=0|X))-1=1-2×P(Z=0|X);
and P (Z ═ 0| X) represents the probability that the user to be analyzed converts the variable into the second value. Marketing conversion gain can also be calculated by P (Z ═ 0| X) as well.
A further purpose of calculating the marketing conversion rate gain is to determine the type of the user to be analyzed and whether to perform a marketing campaign on the user to be analyzed, and therefore, in an embodiment of the present invention, as shown in fig. 5, in addition to the above steps 410 to 430, the method further includes:
step 440, judging whether the marketing conversion rate gain is greater than 0, if so, determining that the user to be analyzed is the user interested in the marketing activity, namely, the user to be analyzed has a high possibility of conversion due to the marketing activity, and providing the user to be analyzed with the marketing activity.
The mode of providing the marketing campaign to the user is, for example, placing an advertisement, sending a short message, and the like, and the specific mode of providing the marketing campaign is not specifically limited herein.
According to the embodiment, audience crowds interested in the marketing campaign can be accurately determined, the marketing campaign to be marketed is provided only for the user with high conversion possibility, the conversion rate of the marketing campaign can be improved, interference on the user who feels dislike to the marketing campaign is avoided, and user experience is improved.
Based on the same inventive concept, a fusion classification model building system and a marketing conversion rate gain prediction system are also provided, as described in the following embodiments. Because the principle of solving the problems of the fusion classification model establishing system and the marketing conversion rate gain prediction system is similar to that of the fusion classification model establishing method and the marketing conversion rate gain prediction method, the fusion classification model establishing system and the marketing conversion rate gain prediction system can be implemented by referring to the fusion classification model establishing method and the marketing conversion rate gain prediction method, and repeated parts are not repeated.
The fusion classification model establishing system and the marketing conversion rate gain predicting system comprise a plurality of functional modules, which can be realized by a special or general chip, and can also be realized by a software program, and the text does not limit the functions.
Specifically, as shown in fig. 6, the marketing fusion classification model establishing system includes:
the data acquisition module 600 determines the user characteristics and selects a control group and a marketing group of the marketing campaign according to the marketing campaign characteristics, so that the number of users in the control group is the same as that in the marketing group and the distribution of the characteristics of the users is consistent
The analysis module 610 is used for determining the conversion results of the users to the marketing activities in the control group and the marketing group after the marketing activities are carried out for a preset time period;
a data conversion processing module 620, configured to convert the conversion result into a conversion variable according to a predetermined conversion rule for balancing the sample;
the model building and training module 630 is used for training the characteristics of the users in the control group and the marketing group as input and the value probability of the conversion variable of the corresponding user as output by adopting a deep learning method to obtain a fusion classification model;
the conversion variable comprises a first value and a second value, the first value indicates that the user conversion result in the marketing group is converted or the user conversion result in the control group is not converted, and the second value indicates that the user conversion result in the marketing group is not converted and/or the user conversion result in the control group is converted.
In some embodiments, the data transformation module 620 transforms the transformation result of the user using a predetermined transformation rule for balancing the samples as follows:
Figure BDA0002800662560000131
wherein, Z represents a conversion variable, T represents an operation group, C represents a control group, Y represents a conversion result, G ═ T represents that a user comes from the operation group, G ═ C represents that a user comes from the control group, Y ═ 1 represents that the conversion result is conversion, Y ═ 0 represents that the conversion result is non-conversion, Z ═ 1 represents that a user conversion result in the operation group is conversion and a user conversion result in the control group is non-conversion, Z ═ 0 represents that a user conversion result in the operation group is non-conversion and/or a user conversion result in the control group is conversion.
In some embodiments, the process of training the fusion classification model by the model building and training module 630 using a deep learning method includes: establishing an N-layer neural network model, wherein the first N-1 layer adopts a ReLU function, and the last layer adopts a Sigmod activation function; initializing a neural network model; the characteristics of the users in the comparison group and the marketing group are used as input, and the variable value probability is converted to be used as output; constructing a two-class cross entropy loss function according to the input, the output and the neural network model; optimizing a neural network model according to a two-classification cross entropy loss function; and the optimized neural network model is the fusion classification model.
The marketing fusion classification model establishing system can directly calculate the marketing conversion rate gain of a user by introducing the fusion classification model established by the conversion variable, can ensure that the control group and the marketing group data are fused, avoids error amplification caused by the double classification model, and can relieve the problem of unbalance of a positive sample (the probability of marketing activity conversion) and a negative sample (the probability of marketing activity non-conversion) by introducing the conversion variable.
As shown in fig. 7, the marketing conversion rate gain prediction system includes:
the classification model establishing module 710 is configured to establish a fusion classification model according to characteristics of a to-be-marketed activity by using the fusion classification model establishing method according to any one of the embodiments;
the model calculation module 720 is used for calculating the value probability of the conversion variable of the user to be analyzed according to the characteristics of the user to be analyzed and the fusion classification model;
and the gain calculation module 730 is used for calculating the marketing conversion rate gain according to the probability of the value of the user conversion variable to be analyzed.
In one embodiment, the gain calculation module 730 may calculate the marketing conversion gain using the following formula:
Δ (X) ═ 2 × P (Z ═ 1| X) -1; or
Δ(X)=1-2×P(Z=0|X);
Wherein, X represents a user to be analyzed, Z ═ 1 represents that the user conversion result in the control group is that the user receives the marketing product and the user conversion result in the control group is that the user does not receive the marketing product, P (Z ═ 1| X) represents the probability that the user to be analyzed has a conversion variable of 1, and P (Z ═ 0| X) represents the probability that the user to be analyzed has a conversion variable of the second value.
The method and the device can accurately and efficiently calculate the marketing conversion rate gain, and provide a basis for subsequent customer classification and marketing activity object determination.
In a further embodiment, as shown in fig. 8, the marketing conversion gain prediction system comprises: further comprising:
the determining module 740 is configured to determine whether the marketing conversion rate gain is greater than 0, and if the marketing conversion rate gain is greater than 0, determine that the user to be analyzed is a user interested in the marketing campaign, that is, if the probability that the user to be analyzed is converted due to the marketing campaign is high, provide the user to be analyzed with the marketing campaign.
According to the embodiment, interested audience crowds of the marketing campaign can be accurately determined, the marketing campaign to be marketed is provided only for the user with high conversion possibility, the conversion rate of the marketing campaign can be improved, interference on the user who feels dislike to the marketing campaign is avoided, and user experience is improved.
In an embodiment of the present disclosure, as shown in fig. 9, fig. 9 is a flowchart illustrating a process of determining a marketing campaign object according to an embodiment of the present disclosure, and in particular, the process of determining a marketing campaign object includes:
step 910, establishing a fusion classification model according to the characteristics of the activity to be marketed, specifically, the process of establishing the fusion classification model includes:
(1) determining user characteristics, a control group and a marketing group according to the characteristics of the activities to be marketed so as to enable the number of users in the control group and the marketing group to be the same and the characteristic distribution of the users to be consistent;
(2) determining conversion results Y of the users to the marketing activities in a control group and a marketing group of the marketing activities, wherein Y is 1 to represent conversion, and Y is 0 to represent no conversion;
when the step is implemented, the marketing group can be tried and marketed for a period of time through the activity to be marketed, and the step can also be determined according to the historical data of users in the control group and the marketing group;
(3) converting the conversion result by using the following rules to obtain a conversion variable:
Figure BDA0002800662560000151
wherein, Z represents a conversion variable, T represents a marketing group, C represents a control group, G ═ T represents that a user comes from the marketing group, G ═ C represents that the user comes from the control group, Y ═ 1 represents conversion, Y ═ 0 represents non-conversion, Z ═ 1 represents that the user conversion result in the marketing group is conversion and the user conversion result in the control group is non-conversion, Z ═ 0 represents that the user conversion result in the marketing group is non-conversion and/or the user conversion result in the control group is conversion;
(4) the characteristics of the users in the control group and the marketing group are used as input, the probability of the conversion variable Z of the corresponding user being 1 is used as output, a fusion classification model is obtained by adopting a deep learning method for training, the training process is referred to the embodiment, and the detailed description is omitted;
step 920, acquiring user data to be analyzed, inputting each user feature to be analyzed into the fusion classification model, and calculating to obtain the probability that a conversion variable Z corresponding to the user to be analyzed is 1;
step 930, calculating the marketing conversion rate gain of each user to be analyzed by using the following formula according to the probability that the conversion variable Z corresponding to each user to be analyzed is equal to 1:
Δ(X)=2×P(Z=1|X)-1;
wherein, X represents a user to be analyzed, Z ═ 1 represents that the user conversion result in the control group is that the marketing product is accepted and the user conversion result in the control group is that the marketing product is not accepted, and P (Z ═ 1| X) represents the probability that the user to be analyzed has a conversion variable of 1;
step 940, judging whether the marketing conversion rate gain delta (X) of each user to be analyzed is greater than 0, if the marketing conversion rate gain of the user to be analyzed is greater than 0, dividing the user to be analyzed into marketing objects, if the marketing conversion rate gain of the user to be analyzed is less than 0, not providing the user to be analyzed with a marketing activity, if the marketing conversion rate gain of the user to be analyzed is equal to 0, providing the user with the marketing activity to be analyzed, and not providing the user with the marketing activity to be analyzed;
step 950, providing the user in the set of marketing objects with a pending marketing campaign.
The embodiment can achieve the following technical effects:
(1) a conversion variable is introduced to convert the problem into a single model for directly modeling marketing classification, so that an amplification error caused by double-model difference is avoided;
(2) the fusion classification model is established by using a deep learning method, so that the method has stronger learning ability and high-efficiency feature expression ability, and can quickly learn new effective feature expression from training data aiming at new application;
in the deep neural network method, the feature representation and the classifier are optimized in a combined manner, so that the optimization effect can be exerted to a greater extent;
(3) by introducing the conversion variable, according to the conversion rule of the conversion variable Z and the conversion result Y, the proportion of positive and negative samples (Z is positive when 1, and Z is negative when 0) of Z is very close, and the problem of unbalanced samples in machine learning is effectively avoided.
In an embodiment herein, as shown in fig. 10, there is also provided a computer device, the computer device 1002 may include one or more processors 1004, such as one or more Central Processing Units (CPUs), each of which may implement one or more hardware threads. The computer device 1002 may also include any memory 1006 for storing any kind of information, such as code, settings, data, etc. For example, and without limitation, the memory 1006 may include any one or more of the following in combination: any type of RAM, any type of ROM, flash memory devices, hard disks, optical disks, etc. More generally, any memory may use any technology to store information. Further, any memory may provide volatile or non-volatile retention of information. Further, any memory may represent fixed or removable components of computer device 1002. In one case, when the processor 1004 executes the associated instructions, which are stored in any memory or combination of memories, the computer device 1002 can perform any of the operations of the associated instructions. The computer device 1002 also includes one or more drive mechanisms 1008, such as a hard disk drive mechanism, an optical disk drive mechanism, or the like, for interacting with any memory.
Computer device 1002 may also include an input/output module 1010(I/O) for receiving various inputs (via input device 1012) and for providing various outputs (via output device 1014)). One particular output mechanism may include a presentation device 1016 and an associated graphical user interface 1018 (GUI). In other embodiments, input/output module 1010(I/O), input device 1012, and output device 1014 may also be excluded, as only one computer device in a network. Computer device 1002 can also include one or more network interfaces 1020 for exchanging data with other devices via one or more communication links 1022. One or more communication buses 1024 couple the above-described components together.
Communication link 1022 may be implemented in any manner, such as over a local area network, a wide area network (e.g., the Internet), a point-to-point connection, etc., or any combination thereof. Communications link 1022 may include any combination of hardwired links, wireless links, routers, gateway functions, name servers, etc., governed by any protocol or combination of protocols.
In an embodiment of the present disclosure, a computer-readable storage medium is further provided, where a computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, the method for building a fusion classification model or predicting a marketing conversion rate gain according to any of the foregoing embodiments is performed.
In an embodiment herein, there is also provided computer readable instructions, wherein when executed by a processor, the program causes the processor to perform the fusion classification model building method or the marketing conversion rate gain prediction method of any of the preceding embodiments.
It should be understood that, in various embodiments herein, the sequence numbers of the above-mentioned processes do not mean the execution sequence, and the execution sequence of each process should be determined by its function and inherent logic, and should not constitute any limitation to the implementation process of the embodiments herein.
It should also be understood that, in the embodiments herein, the term "and/or" is only one kind of association relation describing an associated object, meaning that three kinds of relations may exist. For example, a and/or B, may represent: a exists alone, A and B exist simultaneously, and B exists alone. In addition, the character "/" herein generally indicates that the former and latter related objects are in an "or" relationship.
Those of ordinary skill in the art will appreciate that the elements and algorithm steps of the examples described in connection with the embodiments disclosed herein may be embodied in electronic hardware, computer software, or combinations of both, and that the components and steps of the examples have been described in a functional general in the foregoing description for the purpose of illustrating clearly the interchangeability of hardware and software. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the implementation. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present disclosure.
It is clear to those skilled in the art that, for convenience and brevity of description, the specific working processes of the above-described systems, apparatuses and units may refer to the corresponding processes in the foregoing method embodiments, and are not described herein again.
In the several embodiments provided herein, it should be understood that the disclosed system, apparatus, and method may be implemented in other ways. For example, the above-described apparatus embodiments are merely illustrative, and for example, the division of the units is only one logical division, and other divisions may be realized in practice, for example, a plurality of units or components may be combined or integrated into another system, or some features may be omitted, or not executed. In addition, the shown or discussed mutual coupling or direct coupling or communication connection may be an indirect coupling or communication connection through some interfaces, devices or units, and may also be an electric, mechanical or other form of connection.
The units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the units can be selected according to actual needs to achieve the purposes of the embodiments herein.
In addition, functional units in the embodiments herein may be integrated into one processing unit, or each unit may exist alone physically, or two or more units are integrated into one unit. The integrated unit can be realized in a form of hardware, and can also be realized in a form of a software functional unit.
The integrated unit, if implemented in the form of a software functional unit and sold or used as a stand-alone product, may be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present invention may be implemented in a form of a software product, which is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the methods described in the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk or an optical disk, and other various media capable of storing program codes.
The principles and embodiments of this document are explained herein using specific examples, which are presented only to aid in understanding the methods and their core concepts; meanwhile, for the general technical personnel in the field, according to the idea of this document, there may be changes in the concrete implementation and the application scope, in summary, this description should not be understood as the limitation of this document.

Claims (10)

1. A fusion classification model building method is characterized by comprising the following steps:
determining a comparison group of the marketing activities and conversion results of users in the marketing groups to products related to the marketing activities;
converting the conversion result into a conversion variable by using a preset conversion rule for balancing the sample;
taking the characteristics of the users in the control group and the marketing group as input, taking the value probability of the conversion variables of the users as output, and training by adopting a deep learning method to obtain a fusion classification model;
the conversion variable values comprise a first value and/or a second value, the first value indicates that the user conversion result in the marketing group is converted or the user conversion result in the control group is not converted, and the second value indicates that the user conversion result in the marketing group is not converted and/or the user conversion result in the control group is converted.
2. The method of claim 1, wherein prior to determining the conversion results of the control group of the marketing campaign and the users in the marketing group to the products associated with the marketing campaign, further comprising:
determining user characteristics, a comparison group and a marketing group of the marketing campaign according to the characteristics of the marketing campaign;
wherein the number of users in the control group is the same as that in the marketing group, and the feature distribution of the users is consistent.
3. The method of claim 1, wherein the predetermined conversion rules for balancing the samples comprise rules expressed using the following formula:
Figure FDA0002800662550000011
wherein Z represents a conversion variable, T represents an operation group, C represents a control group, Y represents a conversion result, G ═ T represents that a user comes from the operation group, G ═ C represents that the user comes from the control group, Y ═ 1 represents that the conversion result is conversion, Y ═ 0 represents that the conversion result is not conversion, Z ═ 1 represents a first value, and Z ═ 0 represents a second value.
4. The method of claim 1, wherein the characteristics of the users in the control group and the marketing group are used as input, the conversion variable value probability of the users is used as output, and a fusion classification model is obtained by training by a deep learning method, and the method comprises the following steps:
establishing an N-layer neural network model, wherein the first N-1 layer adopts a ReLU function, and the last layer adopts a Sigmod activation function;
initializing the neural network model;
the characteristics of the users in the comparison group and the marketing group are used as input, and the conversion variable value probability of the users is used as output;
constructing a loss function according to the input, output and neural network model;
optimizing the neural network model according to the loss function;
and the optimized neural network model is the fusion classification model.
5. A marketing conversion rate gain prediction method is characterized by comprising the following steps:
establishing a fusion classification model obtained by the method of any one of claims 1 to 4 according to the characteristics of the activity to be marketed;
calculating to obtain the value probability of the user conversion variable to be analyzed according to the characteristics of the user to be analyzed and the fusion classification model;
and calculating the marketing conversion rate gain according to the value probability of the user conversion variable to be analyzed.
6. The method of claim 5, wherein calculating a marketing conversion rate gain according to the value probability of the user transformation variable to be analyzed comprises calculating the marketing conversion rate gain using the following formula:
Δ (X) ═ 2 × P (Z ═ 1| X) -1; or
Δ(X)=1-2×P(Z=0|X);
Wherein, X represents the feature of the user to be analyzed, Z ═ 1 represents a first value, P (Z ═ 1| X) represents the probability that the user to be analyzed converts the variable into the first value, and P (Z ═ 0| X) represents the probability that the user to be analyzed converts the variable into the second value.
7. The method of claim 6, further comprising:
and judging whether the marketing conversion rate gain is greater than 0, if so, determining that the user to be analyzed is the user interested in the marketing activity, and providing the marketing activity to the user to be analyzed.
8. A marketing conversion gain prediction system, comprising:
the classification model establishing module is used for establishing a fusion classification model obtained by the method of any one of claims 1 to 4 according to the characteristics of the activity to be marketed;
the model calculation module is used for calculating the value probability of the conversion variable of the user to be analyzed according to the characteristics of the user to be analyzed and the fusion classification model;
and the gain calculation module is used for calculating to obtain the marketing conversion rate gain according to the value probability of the user conversion variable to be analyzed.
9. A computer device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor when executing the computer program implements the fusion classification model building method of any one of claims 1 to 4 or the marketing conversion rate gain prediction method of claim 5.
10. A computer-readable storage medium storing an executable computer program which when executed by a processor implements the fusion classification model building method of any one of claims 1 to 4 or the marketing conversion rate gain prediction method of claim 5.
CN202011348424.9A 2020-11-26 2020-11-26 Fusion classification model establishing method, marketing conversion rate gain prediction method and system Pending CN112446541A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202011348424.9A CN112446541A (en) 2020-11-26 2020-11-26 Fusion classification model establishing method, marketing conversion rate gain prediction method and system

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202011348424.9A CN112446541A (en) 2020-11-26 2020-11-26 Fusion classification model establishing method, marketing conversion rate gain prediction method and system

Publications (1)

Publication Number Publication Date
CN112446541A true CN112446541A (en) 2021-03-05

Family

ID=74738467

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202011348424.9A Pending CN112446541A (en) 2020-11-26 2020-11-26 Fusion classification model establishing method, marketing conversion rate gain prediction method and system

Country Status (1)

Country Link
CN (1) CN112446541A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113254644A (en) * 2021-06-07 2021-08-13 成都数之联科技有限公司 Model training method, non-complaint work order processing method, system, device and medium
CN116805253A (en) * 2023-08-18 2023-09-26 腾讯科技(深圳)有限公司 Intervention gain prediction method, device, storage medium and computer equipment

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20170140417A1 (en) * 2015-11-18 2017-05-18 Adobe Systems Incorporated Campaign Effectiveness Determination using Dimension Reduction
CN107527240A (en) * 2016-08-18 2017-12-29 南京坦道信息科技有限公司 A kind of operator's industry product Praise effect identification system and method
CN110335057A (en) * 2019-04-30 2019-10-15 广发证券股份有限公司 A kind of fund Precision Marketing Method that machine learning is merged with artificial rule
CN110837847A (en) * 2019-10-12 2020-02-25 上海上湖信息技术有限公司 User classification method and device, storage medium and server
CN111507395A (en) * 2020-04-15 2020-08-07 赛诺数据科技(南京)有限公司 Marketing big data modeling method and platform
CN111951044A (en) * 2020-07-30 2020-11-17 中国工商银行股份有限公司 Bank terminal interaction method and system

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20170140417A1 (en) * 2015-11-18 2017-05-18 Adobe Systems Incorporated Campaign Effectiveness Determination using Dimension Reduction
CN107527240A (en) * 2016-08-18 2017-12-29 南京坦道信息科技有限公司 A kind of operator's industry product Praise effect identification system and method
CN110335057A (en) * 2019-04-30 2019-10-15 广发证券股份有限公司 A kind of fund Precision Marketing Method that machine learning is merged with artificial rule
CN110837847A (en) * 2019-10-12 2020-02-25 上海上湖信息技术有限公司 User classification method and device, storage medium and server
CN111507395A (en) * 2020-04-15 2020-08-07 赛诺数据科技(南京)有限公司 Marketing big data modeling method and platform
CN111951044A (en) * 2020-07-30 2020-11-17 中国工商银行股份有限公司 Bank terminal interaction method and system

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113254644A (en) * 2021-06-07 2021-08-13 成都数之联科技有限公司 Model training method, non-complaint work order processing method, system, device and medium
CN116805253A (en) * 2023-08-18 2023-09-26 腾讯科技(深圳)有限公司 Intervention gain prediction method, device, storage medium and computer equipment
CN116805253B (en) * 2023-08-18 2023-11-24 腾讯科技(深圳)有限公司 Intervention gain prediction method, device, storage medium and computer equipment

Similar Documents

Publication Publication Date Title
US11373233B2 (en) Item recommendations using convolutions on weighted graphs
Bilal Zorić Predicting customer churn in banking industry using neural networks
US9171259B1 (en) Enhancing classification and prediction using predictive modeling
Yang Adaptive credit scoring with kernel learning methods
US20070124236A1 (en) Credit risk profiling method and system
US20200134716A1 (en) Systems and methods for determining credit worthiness of a borrower
US9483734B2 (en) Transforming predictive models
CN112446541A (en) Fusion classification model establishing method, marketing conversion rate gain prediction method and system
US20190362354A1 (en) Real-time updating of predictive analytics engine
CN111275514A (en) Intelligent purchasing method and system, storage medium and electronic device
CN111274791B (en) Modeling method of user loss early warning model in online home decoration scene
US11921821B2 (en) System and method for labelling data for trigger identification
CN116151857A (en) Marketing model construction method and device
Wu et al. Customer churn prediction for commercial banks using customer-value-weighted machine learning models
CN111667307B (en) Method and device for predicting financial product sales volume
Kazemi et al. Estimation of optimum thresholds for binary classification using genetic algorithm: An application to solve a credit scoring problem
Nagashima et al. Data Imputation Method based on Programming by Example: APREP-S
Chu An Empirical Analysis of Corporate Financial Management Risk Prediction Based on Associative Memory Neural Network
Doumpos et al. Data Analytics for Developing and Validating Credit Models
CN112508608B (en) Popularization activity configuration method, system, computer equipment and storage medium
US20230342617A1 (en) Using machine learning to predict appropriate actions
US11991183B2 (en) Optimizing resource utilization
US20230342597A1 (en) Using machine learning to extract subsets of interaction data for triggering development actions
US20230394512A1 (en) Methods and systems for profit optimization
US20230351493A1 (en) Efficient processing of extreme inputs for real-time prediction of future events

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
RJ01 Rejection of invention patent application after publication

Application publication date: 20210305

RJ01 Rejection of invention patent application after publication